Every opportunity found across four rounds, in one place. Read the shapes first — they matter more than any single opportunity below them.
6 August 2026 · 77 opportunities · 23 strong · 15 recurring shapes
Fifteen shapes, strongest first. For each: what it is, how many industries it turned up in, why the same job keeps appearing, and the one thing you would build for it.
14 industries · 15 opportunities · 7 of them strong
What it is. Somebody sends you a charge, a claim or a deduction. There is a public document — a government regulation, a published price list, a state law, an industry grading standard — that says what they were allowed to charge you and by when. Almost nobody opens it. The job is to read the bill against the rulebook, find the line that fails, and write the argument quoting the rule back at them, before the clock runs out.
Where it turned up. Shipping containers · part-load trucking · farm and construction equipment dealers · oil and gas part-owners · hardwood lumber · car auctions · digging contractors · truck parts · fresh produce · contractors working near railways · rail freight · railcar owners · contractors buying workers' compensation insurance · cable companies renting space on power poles.
Why the same job keeps appearing. Wherever two companies argue about money at scale, somebody eventually publishes a rulebook so the arguments can be settled. That rulebook is then written for the party sending the bill, quoted by the party sending the bill, and never read by the party receiving it — because reading it costs more than the individual charge. So a scrap yard, a container importer, a digging contractor and a tractor dealer are structurally the same customer: each is facing a professional counterparty computing a charge from a document the recipient has never opened.
Better still, several of these rulebooks contain either a checklist or a countdown. Container demurrage bills must contain thirteen specific pieces of information, and leaving one out cancels the obligation to pay. In at least five states, an equipment warranty claim that is not refused in writing within thirty days is automatically approved. Those turn a matter of opinion into arithmetic, which is why they are the best members of the shape.
🔴 THE PRODUCT THESIS. Build one engine that loads a published rulebook and turns it into a list of checkable conditions; reads the other side's bill whatever shape it arrives in; produces a line-by-line finding with the exact rule quoted; drafts the rebuttal in the language that party's claims desk actually answers; and runs a countdown on every open item. Charge a share of what comes off the bill — every one of these markets already accepts that.
What is genuinely the same in all fourteen: the rulebook loader, the argument writer that quotes the rule (the single hardest and most reusable piece — an argument that quotes the rule is worth several times one that just asserts), the deadline clock, the case file, and the share-of-recovery pricing.
What has to differ: which rulebook, the document layouts, where you file it, and the one thing that cannot be built and only learned — how much the other side concedes before a lawyer gets involved.
Where to start: the members whose rulebook contains a checklist or a forfeiture clock, because they need the least judgement and produce the most defensible wins — container charges, equipment warranty, hardwood lumber. The ones where the argument is genuinely a matter of opinion, like freight damage and produce grading, come second.
Members: Utility damage defence · Equipment warranty argument and clock · Container demurrage disputes · Part-load reweigh and reclassification · Car auction arbitration · Truck part core credits · Produce rejection claims · Oil and gas part-owner billing · Railway flagging charges · Power pole attachments · Hardwood lumber grade claims · Freight damage claims · Railcar repair billing · Rail demurrage · Workers' compensation premium audits.
14 industries · 14 opportunities · 5 of them strong
What it is. Same act of arguing with somebody's invoice, but there is no public rulebook. The only thing saying what they were allowed to charge is a private agreement in a drawer. Uniform rental contracts, supplier rebate deals, hotel booking terms, subcontracts, milk purchase terms, union labour agreements. The bill arrives weekly or monthly forever, and nobody checks it against the paper that governs it.
Where it turned up. Multi-site restaurants and hotels on uniform rental · wholesale distribution · independent drinks brands · independent hotels · trade contractors · contractors renting machines · dairy farming · shipping agents · sales rep agencies · timber · hospital laundry · concrete buyers · dock labour · oil and gas part-owners.
Why the same job keeps appearing. This is what happens when a recurring commercial relationship outlives the person who negotiated it. A contract is signed once by somebody senior; the bills are then processed forever by somebody junior who has never seen it. The other side knows this. So a restaurant group's uniform bill, a drinks brand's supplier deductions, a hotel's booking commission and a dairy farmer's milk cheque all break the same way: prices creep, deductions appear, counts drift, and there is no reference document in the room.
The economics are better than Shape 1 — the bills recur weekly or monthly, so one contract-reading exercise produces a permanent stream of checks. The difficulty is that the reference document is bespoke, so the reading has to be redone per customer rather than per industry.
🔴 THE PRODUCT THESIS. Build a contract-to-invoice reconciler. Step one is a reader that turns whatever agreements the customer actually has — a signed PDF, an emailed rate sheet, a letter, a spreadsheet of deals — into a machine-checkable model of what may be charged, at what rate, to whom, when. Step two runs every recurring bill against that model on its own cadence and produces the credit claim and the argument.
Genuinely common: the contract reader, the fuzzy line matcher (the same item is described three different ways in three systems — this piece recurs literally everywhere in this shape), the exception queue, the credit-claim writer, and the engine that runs weekly rather than annually.
🔴 THE SHARPEST OBSERVATION IN THE WHOLE CATALOGUE, AND IT IS AN ARGUMENT FOR BUILDING THIS. The human competitors here — the cost-reduction consultancies — all work monthly or quarterly, by hand, for large accounts, on a share of savings for up to five years. The bills arrive weekly. A monthly human cannot check a weekly bill. That is arithmetic, not opinion, and it is the clearest structural gap in the corpus. Build the weekly version and the human version becomes the price to beat rather than the competitor.
Members: Uniform and linen invoice audit · Weekly uniform reconciliation · Supplier rebates and deductions · Drinks distributor reporting · Hotel booking commission recovery · Subcontractor deduction defence · Idle rented equipment · Milk cheque settlement · Ship agency disbursements · Sales rep commissions · Timber scale settlement · Hospital linen · Concrete tickets · Dock gang timekeeping.
8 industries · 9 opportunities · touches several other shapes
What it is. A letter arrives about something that happened weeks or months ago — a hit cable, a mess on a floor, a bad engine, a warm truckload, a slip in a car park. It has a number on it and a short deadline. To answer it you have to rebuild one specific day out of things nobody filed on purpose: a foreman's camera roll, a group chat, a daily report, a delivery ticket, a temperature trace, a timesheet. Most companies simply pay, because building the answer costs more than the bill.
Where it turned up. Digging contractors · trade contractors on building sites · car auctions and dealer groups · regional trucking · employers facing insurance audits · produce growers · commercial snow clearing · rural cable operators.
🔴 WHY THIS IS THE MOST LOPSIDED SHAPE IN THE CATALOGUE. The party making the claim has a payday. The party defending has a cost centre. So professional, no-win-no-fee, increasingly AI-assisted machinery grows on the attacking side of every one of these arguments, and nothing at all grows on the defending side. One utility recovery firm invoices 81,000 claims a year. A buyer-side car auction claims service advertises a $300 average concession and says over half of claims result in one. Meanwhile the defendant is a foreman scrolling his phone under a deadline. That asymmetry is identical whether the defendant digs trenches, hangs drywall, hauls freight, wholesales cars or grows lettuce.
🔴 THE PRODUCT THESIS. Build a defence-file builder, triggered by an accusation arriving rather than by a routine. It reads the claim letter, works out which day and which job it concerns, reaches into everything the company already has — phone photos, texts, daily logs, tickets, timesheets, gauge printouts — assembles a timeline of what actually happened, tests it against the contract or the law, and drafts the rebuttal inside the response window.
The hard part nobody sells: working BACKWARDS. Every product on the market in these industries works forwards — capture the evidence properly as the job happens so the file exists later. Not one works backwards, which is the situation the defendant is actually in, because the incident nobody photographed on purpose is precisely the one that generates a claim. Build the backward reader once and it serves all eight industries.
🔴 A FREE SELECTION RULE. Go and look at who is selling to the attacking side. Wherever there is a funded, no-win-no-fee attacker and a silent defendant, there is an opportunity — and you can see that pattern from outside an industry without knowing anything about it.
Members: Utility damage defence · Subcontractor deduction defence · Car auction arbitration (seller side) · Freight damage claims · Workers' compensation premium audits · Produce rejection claims · Snow service verification · Power pole attachments.
10 industries · 9 opportunities · 3 of them strong
What it is. Money has been earned and somebody else is sitting on it until the paperwork satisfies their checklist. A mobile phone company will not release a final invoice until the closeout photos are right. A power company will not pay a rebate until the equipment photo is legible. A state highway department will not pay for extra work until the labour and equipment arithmetic matches its own manual. The job is assembling and pricing that pack, and doing it again when it comes back short.
Where it turned up. Mobile phone mast construction · insulation, lighting and mechanical contractors · highway contractors · rural power co-operatives · importers and exporters · food wholesale · aircraft parts · commercial snow · steel and metal suppliers · specialist building subcontractors.
Why the same job keeps appearing. Any time a large organisation pays a smaller one for work it did not directly supervise, it defends itself with an evidence requirement — written for its own convenience rather than the payee's. The payee is then a small business whose cash is frozen behind somebody else's checklist. The pain is never disorganisation; it is a frozen final payment on 60- or 90-day terms while payroll runs fortnightly. That is the single most quoted complaint across this whole shape.
🔴 THE PRODUCT THESIS. Build a pack assembler that knows the payer's acceptance criteria as data. It reads the payer's own requirement document, reads what the crew or the yard actually produced, and — this is the part nobody sells — judges whether it is good enough, rather than whether it exists. The named competitor in mast closeout tracks whether a photo exists in a category; every rejection in the trade press is about whether the photo is acceptable. Telling a crew what is missing while they are still on site is the whole product.
What must differ: who the payer is, what the medium is (photographs, priced field tickets, mill certificates, logbooks), and the pricing arithmetic where one exists.
Pricing falls out naturally: a share of money released, because the amount being unlocked is already known and dated.
Members: Mast closeout packs · Utility rebate filing · Highway extra-work claims · Storm cost recovery · Import duty refunds · Damaged-goods claims at food distributors · Aircraft records gaps · Steel certificate packs · Snow service verification · Subcontractor payment portals.
8 industries · 2 opportunities here, 6 shared with other shapes · the cleanest commercial fit in the catalogue
What it is. You already did the work and already filed the claim. It came back with a terse reason code and sat there. Somebody has to read the reason, work out from the original job what actually went wrong, correct it and send it again — and the software that submitted it shows you a red row and stops.
Where it turned up. Commercial tyre dealers · truck parts · equipment dealers · contractors filing rebates · rural power co-operatives · wholesale distribution · oilfield services · mast contractors.
Why the same job keeps appearing. Every large payer that automates its own intake creates a stream of rejections and pushes the cost of it onto the claimant. The vendor who built the submission pipe had no reason to build the diagnosis, because a rejected claim is not their problem — and the customer's own help documentation says so out loud. One tyre software vendor's help centre literally instructs the user to open the red row, read the reason, click modify, retype and resubmit. That same sentence would describe a truck part credit, a refused warranty claim, a rebate resubmission and a short-paid supplier deduction.
🔴 THE PRODUCT THESIS. Build a reject worker. It watches the payer's rejection feed, works out the real cause by reading the reason code against the original job record, corrects the submission, resends it, tracks whether it landed, and escalates the ones that bounce twice. It is not a submission tool and must never become one — the submission is exactly the part every competitor already ships.
Genuinely common: interpreting reason codes (a small, learnable vocabulary per payer that nobody publishes and everybody discovers by getting rejected), rebuilding the correct submission from the underlying job, and the resend-and-chase loop. Plus an appeal layer for payers who have a formal one — the federal disaster agency runs an open, downloadable appeals record, meaning the argument can be built from the payer's own history of what it has previously accepted.
What must differ: the payer's portal and its reason-code vocabulary. That is genuinely almost all of it, which makes this the most transferable shape here.
🔴 WHY THIS IS THE BEST BUSINESS-MODEL FIT OF ANY SHAPE. The customer can measure the result the day you start, the money is unambiguous, and the alternative is a red row nobody has time for. It is also the shape where "we run it and take a share" is obviously the right delivery, because no dealer wants another portal.
Members: Tyre national-account rebilling · Oilfield tickets · plus core credits, equipment warranty, utility rebates, storm cost recovery, supplier deductions and mast closeout, all listed under other shapes.
10 industries · 3 opportunities here, 7 shared · the most under-appreciated shape in the catalogue
What it is. Not an argument about a bill — a silence. Money the company was entitled to and never asked for: rent never switched on for a tank sitting at a customer's yard, a rebate the sales team qualified for and nobody claimed, a duty refund on goods that were exported, a chargeable job done on a warehouse floor that never reached an invoice, a commission statement that simply never arrived, an audit right that quietly expired. Nobody is disputing it because nobody knows it exists.
Where it turned up. Wholesale distribution · importers and exporters · trade contractors · fertiliser and feed makers · sales rep agencies · propane and welding gas · cold storage · truck parts · highway contractors · oil and gas part-owners.
Why the same job keeps appearing. An error announces itself; an absence does not. Every one of these companies has a process for handling a document that arrives and no process at all for a document that never came. So the leak is invisible by construction and stays invisible for years. A rebate vendor's own marketing says the largest source of margin erosion for distributors is qualifying sales that never generated a claim. Only about a fifth of eligible duty refunds are claimed. Gas distributors installing tracking systems discover accounts holding more cylinders than they were billing for. In every case two records exist that would reveal the gap if anyone laid them side by side, and nobody ever has.
🔴 THE PRODUCT THESIS. Build an absence detector. It takes two records that were never designed to be compared — a sales ledger and a pile of supplier agreements, a delivery history and a billing file, a shipment record and a customs entry, a floor's paper tickets and a month's issued invoices — and produces the list of things that should have generated a claim and did not. The output is not a dispute; it is a claim that has never been filed.
Genuinely common: matching names across two vocabularies that do not agree, pulling entitlements out of agreements written by hundreds of different counterparties, and — the genuinely distinctive capability — reasoning about what is missing rather than what is wrong. Detecting absence is a materially harder and rarer thing than detecting error, and no competitor in any of these ten industries claims it. Every product found reconciles what arrived.
🔴 WHY THIS SHAPE IS WORTH MORE THAN ITS ENTRIES. It is the only shape where the customer cannot currently see the size of their own problem, which means the first delivery IS the sales pitch. And the test that would settle each of the ten entries is the same test: put the two records side by side for one month and count. That test is cheap, identical in all ten industries, and converts directly into the product.
Members: Cold storage unbilled work · Cylinder and tank rent · Product registration and tonnage reporting · plus supplier rebates, duty refunds, utility rebates, sales rep commissions, core credits, railway flagging refunds and oil and gas audit rights, listed elsewhere.
8 industries · 8 opportunities · 3 of them strong
What it is. A regulator obliges you to hold a complete, current, dated record proving a routine thing keeps happening — refrigerant added and leaks repaired, chemicals on site, fuel tanks checked every thirty days, miners trained, drills run on boats, waste paperwork returned signed, grant-funded people's time certified. The duty sits on you. The evidence is generated by a contractor, a crew or another organisation, in free text, in their own format, for their own purposes. Somebody has to turn their paperwork into your file before somebody knocks on the door.
Where it turned up. Cold storage, food plants, breweries, dairies and ice rinks · small manufacturers holding chemicals · fuel tank owners like school districts and city yards · quarries · tug and towing fleets · waste-producing industrial sites · grant-funded charities and workforce agencies · industrial contractors keeping their site approvals alive.
Why the same job keeps appearing. Regulators write duties onto the party with the assets, because that party is findable and permanent. But the actual work is done by whoever holds the spanner, the truck or the sub-award — and they write it up their own way. So the duty and the evidence permanently sit in two different companies. The compliance software market has settled on the deadline — the filing, the dashboard, the reminder — because the deadline is visible. Nobody has taken the register behind it, because the register is invisible until it is wrong.
🔴 THE PRODUCT THESIS. Build an evidence keeper that works from inbound mess rather than from data entry. It knows the requirement as a model — what must exist, produced by whom, at what interval, kept how long — reads whatever actually arrives (a contractor's PDF invoice, a photographed walkthrough sheet, a training certificate on another company's letterhead, a receipt pile), decides whether each item satisfies the requirement, chases what has not arrived, and holds an inspection-ready file.
The hard part is the same as Shape 6: detecting what did NOT arrive. Every product in this space stores and schedules. The failure in the field is never a wrong document, it is a missing one — and the industry's own guidance says so. The most common refrigerant audit failure is described as records that exist but cannot be produced, sitting in a contractor's spreadsheet the owner has never seen.
🔴 THE SELECTION RULE INSIDE THIS SHAPE, AND IT IS DECISIVE. Pick the members with a SHORT CLOCK. Refrigerant leak arithmetic is triggered on every single top-up with a thirty-day repair window. Waste paperwork runs 45- and 60-day clocks. Fuel tank walkthroughs are every thirty days. Those are continuous jobs. The chemical inventory filing is annual and fails the frequency test outright — treat it as the weak member, not the flagship. A continuous duty is a product; an annual duty is a consultancy engagement in disguise.
Members: Refrigerant leak file · Waste paperwork loop closure · Grant time and sub-recipient evidence · Chemical inventory filing · Fuel tank monthly file · Mine training records · Towing vessel evidence · Contractor site approvals.
10 industries · 9 opportunities · none strong, and the reason why is the most useful warning here
What it is. One set of facts — a load, a shipment, a product, a vehicle, a payroll — has to be reshaped into a different form for every state, county, city, township, court or fund it touches, each with its own layout, deadline, fee and website. A company operating in twenty places does the same job twenty times a month.
Where it turned up. Heavy haulage and cranes · tobacco wholesale · fertiliser and pet food makers · multi-site vehicle repair groups · unionised builders · sign makers · water backflow testers · multi-state employers · scrap yards and pawn shops · funeral homes.
Why the same job keeps appearing. Jurisdictions are sovereign and none of them coordinate. Any business crossing their boundaries inherits the multiplication, and the multiplication does not care what the business does.
🔴 THE PRODUCT THESIS — the moat is a CORPUS, not a workflow. In every member of this shape, the real asset is that nobody has read the jurisdictions into one machine-readable model: three thousand county and city rules for oversize loads, fifty state tonnage forms, fifty vehicle lien laws, hundreds of union fund forms. Reading thousands of inconsistent local government documents into one consistent answer is exactly the thing that was impossible before AI and is a weekend's work to copy afterwards. The corpus is the product. The form-filling is the commodity.
🔴 THE STRONGEST NEGATIVE SIGNAL IN THE ENTIRE CATALOGUE, AND IT LIVES HERE. A public form with a legal deadline attracts vendors EARLY, because the requirement is guaranteed and the form is published. Sign permits went from an empty room to three competing products in twenty-three months, one with a published price list. Water backflow reporting is closed, eleven vendors, $29 a month. Subcontractor payment portals: closed. Scrap yard police reporting: closed by a connector the industry's own software already ships. Four of the nine members here came back closed or weak.
So the rule for choosing inside this shape: go to the layer nobody has mapped, which is almost always the LOCAL one. Every state-level version of this job is taken. The county, city and township layer beneath it is untouched — and that is precisely where the reading is hardest and the corpus is most valuable.
Members: Oversize load permits · Tobacco excise filings · Vehicle liens and abandoned titles · Union benefit fund remittance · Wage garnishment orders · Funeral home death records · Sign permits (closed) · Backflow reporting (closed) · Scrap yard police reporting (closed).
6 industries · 2 opportunities here, 4 shared · both of the ones here are strong
What it is. You are legally or commercially obliged to hold a piece of paper that another company has to produce — a made-in-America certificate, a signed waste form coming back, a missing page of an aircraft's history, a doctor's signature, an insurance document matching your contract. They have no incentive to send it, often do not know the answer, and when something finally arrives somebody has to judge whether it is good enough or start again.
Where it turned up. Highway, water and transit contractors and their suppliers · waste-producing sites · aircraft parts traders · funeral homes · general contractors and facilities departments · steel and metal suppliers.
Why the same job keeps appearing. The obligation and the knowledge sit in different companies, and the penalty falls entirely on the side that does not hold the knowledge. In every case the market has built a tool that tells you what is missing and then hands you the phone. One vendor's own material says it plainly: rebuilding incomplete records requires contacting previous owners, maintenance providers or manufacturers. That sentence is the handover point, and it is the same sentence in six industries.
🔴 THE PRODUCT THESIS. Build a chaser with judgement attached. It knows what document is owed, by whom, by when; runs the outbound campaign across whatever channels that counterparty actually uses; reads the reply and decides whether it is sufficient — which is where the value is, because a generic "all our products are American made" is now explicitly not an answer under the current rules; and writes the follow-up asking for the specific missing element.
🔴 THE HONEST WEAKNESS. A real portion of the chasing ends in a phone call to somebody who does not answer email, and people are a distant third priority. This shape is strong on the reading and judging and weak on the last mile. The two members where the last mile is least human are made-in-America certification (email to a supplier who wants to keep the business) and waste paperwork closure (a disposal site with a legal duty of its own) — build those first. The aircraft and funeral-home versions lean hardest on human chasing.
Members: Made-in-America supplier certification · Insurance endorsement verification · plus waste paperwork, aircraft records, funeral home records and steel certificates, listed elsewhere.
5 industries · 2 opportunities here, 3 shared · contains what may be the single best entry in the catalogue
What it is. A part, a load of steel, a subcontractor's crew or a policy document turns up, and with it a certificate that is supposed to prove it is what it claims to be. Somebody has to read that certificate and decide whether it is complete, internally consistent, and issued by somebody who was actually allowed to issue it — then accept the thing or quarantine it. Getting it wrong is not a paperwork error; it is a bad part on an aircraft, an uninsured subcontractor on your site, or a claim denied at exactly the moment it matters.
Where it turned up. Aircraft parts distribution and repair · steel and metal suppliers · general contractors, property owners, hospital and university facilities · quarries · insurance agencies.
Why the same job keeps appearing. Any supply chain that cannot inspect quality by looking at the object substitutes a document for the inspection. The document then becomes the control, and the control is only as good as whoever reads it at the receiving desk — one person, under time pressure, checking a form they have seen in forty different layouts. That is how an aviation counterfeit scandal involving roughly sixty thousand parts with forged certificates was possible, and why the industry's own coalition has written "develop software to electronically validate the key document fields" onto its list of things that do not yet exist.
🔴 THE PRODUCT THESIS. Build a receiving desk that judges rather than files. It reads the certificate whatever the layout, pulls out the identifiers, checks internal consistency, verifies that the issuing organisation actually holds a live approval covering that item, compares the described item against what physically arrived and what was ordered, and returns an accept-or-reject with the reasons written out for a human to sign.
The part nobody sells is authority and eligibility judgement, not extraction. The one vendor claiming to read aviation certificates demonstrates its product on a purchase order. The insurance certificate category verifies that a cover sheet exists and is in date, while a competitor's own comparison admits the tools do not check the actual coverage documents. Extraction is a commodity. Deciding whether the issuer was permitted to issue this form for this article — or whether an insurance document actually satisfies the clause a lawyer wrote — is not.
🔴 THE STRONGEST SINGLE MEMBER OF THIS SHAPE, AND ARGUABLY OF THE WHOLE CATALOGUE: checking a supplier's actual insurance coverage documents against the contract that supplier signed. Every certificate-tracking product on the market checks a one-page document that is legally worthless; an entire funded category exists to do that badly; and the customer finds out at the claim. Same document, opposite half, empty room.
Members: Aviation parts certificate checking · Insurance policy checking (closed) · plus insurance endorsement verification, steel certificates and mine training certificates, listed elsewhere.
11 industries · 6 opportunities here, 5 shared · hands you a hard selection rule
What it is. Money is earned in a truck, a pit, a plant or on a dock, and it is recorded there by somebody paid to do the work rather than to write it up — a scrawled ticket, a photographed slip, a radio call, a hand-marked manifest. Somebody in an office then has to turn that into a priced, coded, customer-specific invoice, and whatever they cannot read or did not receive is simply never billed.
Where it turned up. Oilfield services · cold storage · food and drink delivery routes · ready-mix concrete · fire protection · car salvage · court reporting · machine shops · dental and optical labs · dock labour · highway contractors.
Why the same job keeps appearing. The person who creates the billable event and the person who bills it are never the same person and rarely in the same building. And the loss is silent: unlike a disputed invoice, an unbilled event generates no argument and no record, so nobody knows the size of it. A vendor in this space names the failure exactly — omission errors, where labour and materials are never recorded at all.
🔴 THE SELECTION RULE, WHICH IS WORTH MORE THAN THE SHAPE. Where the vendor could move the capture UPSTREAM into an app on the crew's phone, the job is closed or closing: oilfield ticketing has eight vendors and one that owns both ends; grain settlement is fully automated at the weighbridge; concrete extras are captured by the dominant plant system; fire inspection quoting was shipped by the market leader in May 2026. Where the paper comes from a party you do NOT control — a customer's own prescription, a supplier's delivery ticket, a dentist who will never adopt your app, an independent driver — the job cannot be moved upstream and it is still open.
The test before building any member: can the person who writes the note be made to write it differently? If yes, a software vendor will eventually do that and the opportunity evaporates. If no — because they work for somebody else — it is durable.
Members: Delivery route settlement · Machine shop quoting from drawings · Fire inspection quoting (weak) · Dental and optical lab intake (weak) · Court reporting invoices (weak) · Salvage part descriptions (weak) · plus cold storage, concrete tickets, oilfield tickets, dock timekeeping and highway extra-work claims, listed elsewhere.
4 industries · 2 opportunities here, 2 shared · one member wide open, one closed, and that contrast is the lesson
What it is. A big customer will not buy, or will not let you on site, until you answer their form — a fifty-page food safety questionnaire, a safety programme upload to a vetting website, an insurance and approval pack. You have answered nearly all of it before, for a different customer, in a different layout, and nobody wrote it down anywhere useful. Then it renews every year, and there are four different websites.
Where it turned up. Food, drink and ingredient makers · industrial contractors serving heavy industry · insurance agencies · specialist building subcontractors.
Why the same job keeps appearing. Large buyers push their own risk management down onto their suppliers, and each one designs its own form. The supplier is small, has no compliance department, and answers the same questions forever in different words.
🔴 THE PRODUCT THESIS. Build an answer library with a fix-the-rejection loop. It holds every answer the company has previously given with its supporting evidence, matches an incoming question to a previously approved answer even when the wording is entirely different, drafts the response, flags the handful that genuinely need a human, keeps the renewal calendar, and — the part the competitors explicitly decline — reads the reviewer's rejection of a specific document and rewrites that document to pass.
What must differ, and it differs a lot: the subject knowledge. A generic questionnaire tool cannot tell an allergen control question from a sustainability one, which is why the generic vendors serving software companies have never crossed into food.
🔴 THE HONEST READ. This shape's two best members point in opposite directions. Food supplier questionnaires are served in the United States by nobody — the only two companies doing this exact job are Dutch and Belgian, with roughly €600,000 between them, selling European data hosting into a country with over forty thousand food plants. Contractor site approvals are the opposite: a funded AI-native product aimed at exactly this customer at $69 a month, both major vetting platforms shipping their own AI, and a comparison-shopping industry ranking them. Same shape, one wide-open member and one closed one — and you can only see that by putting them side by side, which is the argument for organising by shape.
Members: Food supplier questionnaires (strong) · Insurance agency carrier portals · plus contractor site approvals and subcontractor payment portals, listed elsewhere.
8 industries · 3 opportunities here, 5 shared · THE WORST HIT RATE IN THE CATALOGUE, and that is the finding
What it is. A person is paid to carry facts between two pieces of software, or between a physical reality and a billing record, because the two will never speak.
Where it turned up. Equipment renters · propane and welding gas · power pole owners and cable attachers · insurance agencies · vending · cotton ginning · grain elevators · hospital laundry.
🔴 THIS SHAPE IS LISTED AS A WARNING, AND THE WARNING IS THE PRODUCT THESIS. Of eight members, four came back genuinely closed — vending, cotton gins, grain elevators, hospital linen — and closed for the same reason every time. When two systems inside ONE industry stay unjoined long enough, either an industry data standard appears or the industry's own software vendor builds the connector, because their customer will otherwise leave over it. The gap closes from the inside.
So the rule, which is worth more than the shape: a two-systems gap is only durable when the two systems belong to DIFFERENT COMPANIES WITH OPPOSING INTERESTS. The rental supplier has no reason to help you prove the machine was idle. The pole owner has no reason to help the cable company reconstruct a permit history. The insurance company has no reason to write its quote back into the agency's system. Those three are the live members. Everything where both systems sit inside one industry's own toolchain has closed or will.
If anything is built here: build the version where you sit on the side being billed and hold the evidence the biller does not want joined.
Members: Vending route settlement (closed) · Cotton gin data (closed) · Grain elevator settlement (closed) · plus idle rented equipment, cylinder tracking, pole attachments, insurance agency portals and hospital linen, listed elsewhere.
4 industries · 0 opportunities of its own · a variant, not a separate build
What it is. The mirror image of arguing with a bill: you are the party who has to READ what somebody else submitted and decide whether it is allowable, correct or payable — a sub-recipient's invoice against a grant's rules, a buyer's arbitration claim against the rulebook, a repair bill against a published price list, a contractor's payroll against an agreement.
Why it is thin, honestly. Three of its four members are better opportunities from the OTHER side of the same table — defend the auction claim rather than adjudicate it, argue the railcar bill rather than issue it. The one genuinely strong member, monthly monitoring of the organisations a grant recipient passes money down to, carries a live risk that must be checked before anything is built: some of these organisations hold sensitive client records, and jobs that only exist inside regulated personal data are out of scope. It must be scoped to money and time evidence only, and if the allowability judgement cannot be made without opening client files, it dies. That is a question to answer in an afternoon, not after a build.
If it is built, what is common is reading an unstructured submission against a rulebook and ruling on it — which is Shape 1's engine pointed the other way. A variant, not a separate product.
Members: all listed elsewhere — grant sub-recipient monitoring, car auction arbitration, railcar repair billing, workers' compensation premium audits.
4 industries · 1 opportunity of its own · useful mainly as a diagnostic
What it is. You are paid by volume, weight or quantity, and somebody else measures it. Your machine, your crew or your truck says one thing; their survey, their scale or their inspector says another. The difference is your income, and you have to build the argument for your own number.
Why the same job keeps appearing. Wherever a job is priced per unit of something physical, the buyer insists on measuring it and the seller has a continuous record the buyer does not accept. Dredging, earthmoving, timber, grain and scrap are the same transaction wearing different clothes.
Why it is thin. The member that surfaced it — dredging — dies on population: roughly thirty companies in the entire trade association. Grain died because the elevator's own software computes the settlement at the weighbridge before the truck leaves.
🔴 SO THE VALUE HERE IS A DIAGNOSTIC, NOT A PRODUCT. In this family, ask first whether the settlement is already computed automatically by the buyer's software at the point of measurement. Where it is, the room is closed. Where a human still eyeballs a grade or a survey lands weeks later, it is open — which is exactly the split between grain (closed) and timber and scrap (open), and it is the same distinction that made the scrap-grade finding the best result of the previous round.
🔴 THE UNTESTED THREAD WORTH CARRYING FORWARD: the same argument in EARTHMOVING, where the population is orders of magnitude larger than dredging and a construction engineering firm writes publicly that contractors commonly raise concerns over pay quantities. Nobody has researched it.
Members: Dredging pay quantities (weak) · plus timber scale settlement, grain elevator settlement and highway extra-work claims, listed elsewhere.
Tap any row to jump to the full entry. Sort by clicking a column heading.
| # | The op, in a line | St | Ocean | Size | AI |
|---|
Scores out of 5. Ocean 5 = nobody is looking, 1 = crowded named category. Size 5 = clearly seven figures a year. AI 5 = messy judgement where AI is the whole job, 1 = plumbing.
78 opportunities, grouped by shape, strongest first inside each group. Each group opens with a compact table, then the entries.
How to read the scores — all out of 5. OCEAN: 5 means nobody is looking, 1 means a crowded named category. SIZE: 5 means clearly seven figures a year. SAME: 5 means near-identical in every company. AI: 5 means messy judgement where AI is the whole job, 1 means plumbing.
Status: 🟢 STRONG · 🔵 WORTH A LOOK · 🟡 WEAK · ⚫ GENUINELY CLOSED. Nothing has been deleted — weak entries are simply four lines.
15 opportunities · 7 strong
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 1.1 | Defend the utility's bill after a crew hits a buried line | Digging contractors | 🟢 | 5·3·5·4 | one part-time investigator |
| 1.2 | Argue refused warranty claims and run the 30-day legal clock | Farm and plant dealers | 🟢 | 3·3·4·4 | claim-typing services |
| 1.3 | Cancel container storage bills that break the federal invoice rules | Importers, container haulers | 🟢 | 3·4·5·4 | flag-only software |
| 1.4 | Rebut carrier reweigh and reclass bills with real product data | Bulky-goods shippers | 🟢 | 3·3·4·4 | invoice audit firms |
| 1.5 | Defend sellers against buyer claims at car auctions | Dealer and fleet remarketers | 🟢 | 4·2·3·5 | buyer-side only |
| 1.6 | Recover deposits kept on returned truck parts | Truck parts distributors | 🟢 | 4·4·4·4 | car-shop software |
| 1.7 | Build the produce shipper's side of a load rejection | Grower-shippers, packers | 🟢 | 4·2·5·5 | neutral referees only |
| 1.8 | Audit monthly oil and gas bills before the audit right expires | Small non-operating owners | 🔵 | 3·3·4·4 | fly-in audit firms |
| 1.9 | Reclaim unused railway flagging deposits | Heavy civil contractors | 🔵 | 4·3·4·3 | railroad-side only |
| 1.10 | Cut back-billed pole rent by rebuilding permit history | Rural cable, small internet | 🔵 | 3·3·4·4 | pole-owner side |
| 1.11 | Price a lumber grade claim inside the 14-day window | Hardwood mills and buyers | 🔵 | 4·2·5·3 | day-rate inspectors |
| 1.12 | Decline the cargo claim instead of paying it | Regional and final-mile carriers | 🔵 | 3·3·4·4 | shipper-side clerks |
| 1.13 | Argue railcar repair bills | Railcar owners, lessors | 🟡 | 2·2·4·2 | platform ships it |
| 1.14 | Dispute rail storage charges from small railroads | Plants on short-line rail | 🟡 | 2·2·4·3 | nine players |
| 1.15 | Fight the insurance premium audit | Sub-heavy contractors | ⚫ | 1·2·4·2 | nine no-win-no-fee firms |
(this is one job that the corpus recorded twice, as excavator-side-utility-damage-claim-defence and utility-third-party-damage-recovery)
THE OP — Sell digging contractors the defence of the bill the utility sends them after a line gets hit: read the dig ticket, the utility's own "we marked it" response codes, the crew's site photos and the repair invoice, and produce the rebuttal inside the claim window, paid on a share of what comes off the bill.
THE OP — Sell multi-brand farm and construction equipment dealers the ARGUING and the CLOCK rather than the filing: an agent that reads every refusal and short-payment letter, writes the rebuttal in that manufacturer's own vocabulary, and enforces the state law saying a warranty claim is automatically approved if the manufacturer did not refuse it in writing within thirty days.
THE OP — Sell importers and container haulers a standing challenge desk for every storage and equipment bill the ports and shipping lines send: read each invoice against the federal rule's mandatory content list and against their own gate records, write the refund demand, file it inside the 30-day window and chase it, taking a share of what gets cancelled.
THE OP — Sell mid-size shippers the evidence they can never find in time — a maintained, photo-and-spec-backed record of what every one of their products actually weighs and measures in its real packaging — and use it to auto-rebut every reweigh and reclassification bill inside the 30-day window, on a share of what gets reversed.
THE OP — Sell the SELLER side of a wholesale car auction claim the defence file nobody sells them: an AI that reads the claim against the announcements, the condition report, the check-in photos and the rulebook clock, and files the rebuttal inside a two-to-five-day window, on a share of the chargebacks it stops.
THE OP — Sell heavy-duty truck parts distributors and dealer groups the recovery of core deposit money the remanufacturer quietly kept: read every core shipment against the supplier's lump monthly credit and the supplier's own published chargeback schedule, and hand back a line-by-line claim for every core credited short, late or never — paid on a share of what comes back.
THE OP — Sell produce shippers the one thing the industry's three referees structurally cannot be — their own side — by reading the refrigeration download, the loading temperatures, the federal inspection certificate and the sales contract inside the eight-hour rejection clock and telling them whether to concede and how much, paid on a share of what they stop giving away.
THE OP — Sell small and mid-size non-operating owners the audit they never get: read every monthly joint-interest bill against their operating agreement and file the exception list before the two-year audit right expires, on a share of what comes back.
The customers are not the operators — they are family partnerships, ranch and estate interests, doctors' investment groups and small independents holding a few percent of a well. The round-three kill applied the "too technical" screen to the operator, who is not the customer; the customer is somebody opening a PDF at a kitchen table. Who is there: an established audit profession — Martindale ("the go-to firm for joint interest audits"), BRI, Malone, Weaver — but they are people flying to a location to sample records, so an audit only happens when the money justifies the trip. 🔴 Where they stop: industry practice is that audits are "commonly organized by the largest non-operator", with the others invited to share the cost. If you hold three percent of a well, nobody organises one for you and you may never hear about it. Meanwhile the standard clause lets you audit for two full years after the year the bills were rendered — so every year, quietly, a small owner's right to question a year of charges expires. Forced work with a deadline. AI: reading a decades-old negotiated agreement and testing every line of a monthly bill against it. Size: NOT SIZED; nobody publishes recovery rates and Martindale's own site carries no figures or fee model. What would settle it: count how many owners hold interests big enough to be worth auditing but too small for anyone to organise one — that count is the entire question and nobody has it.
THE OP — Stop selling contractors the FIGHT and sell them the REFUND they are already owed: reconcile flagger days actually worked against the deposit prepaid weeks earlier, claim back the unused balance the railroad's own form says it will return, and assemble the pass-through claim to the state highway department — on a share of what lands.
Crews working near live track must hire a railroad flagger, booked weeks ahead against a guess and often prepaid. Who is there: railway coordination firms (Wilson & Company, Bartlett & West, RailPros and others), and an established audit practice — but for the RAILROADS and the state highway departments, not the contractor paying the bill. Five searches for anyone auditing flagging bills for contractors returned nothing; one returned no results at all. 🔴 Where they stop: everyone sits between contractor and railroad on scheduling and access; the one firm that touches cost auditing sells it to railroads and project owners. Why this got reopened: round three worried a railroad would never concede a disputed charge, which is fair — the contractor cannot go to a competing railway. But that is the wrong money. The railroads' own paperwork already concedes the other money: one flagging agreement says that where work finishes in less time than estimated the deposit funds "shall be returned", and a major railroad's own form says any prepayment not used can be refunded. An unclaimed prepayment refund is not a dispute — it is an entitlement nobody collects. Day rates are published and large: $1,600 a day on one short line, $1,300 for ten hours on another, around $1,800 for a big railroad's flagger. Size: NOT SIZED — the day rate is solid, the number of unused days is guesswork. What would settle it: ask three heavy-civil controllers whether they have ever got money back when a crew finished early. "We have never asked" means it is live.
THE OP — Stop selling to the pole owner and sell to the company being back-billed: when a rural cable operator or small internet provider receives an audit finding hundreds of unauthorised attachments and a demand for years of back rent, rebuild the permit history out of decades of scanned paperwork and half-migrated databases and knock the count down, paid on the reduction.
Who is there: a full landscape — Alden Systems, 3-GIS, Joint Use 365, ikeGPS, Katapult, plus Osmose and others selling physical audits and "revenue recapture" as a service, with case studies quoting recovered revenue. 🔴 Where they stop: every product and every service is sold to the POLE OWNER, and the pitch is finding more to bill. The company at the other end of that bill has lawyers if it is a large carrier and nothing at all if it is a rural operator. The record-reconciliation work is identical in both directions and is only sold in one. Two facts make the empty side interesting: one large utility's published procedure gives the attacher 30 days to submit documentation showing the attachments were authorised or are not theirs, and the federal regulator has removed the cap on back-rent penalties under new agreements, raising the exposure on the defending side. A law firm's own description says these disputes "typically involve disputes over records rather than intentional acts" and that validating a claim is expensive and time-consuming — a lawyer describing a document-reconciliation job at lawyer rates. ⚠️ The honest strain: the buyers are cable and internet companies, which is the closest anything in this catalogue comes to the technology exclusion. A rural cable co-op is not a software firm, but that is a call for Nick. What would settle it: find one small operator who received a back-bill in the last two years and ask what they did. "Paid it because we could not prove otherwise" makes it real.
THE OP — Sell hardwood sawmills the defence of a grade claim and lumber buyers the preparation of one: turn a piece tally and an invoice into the exact value difference under the national grading rules inside the fourteen-day window, so mills stop paying claims they do not owe and buyers stop losing claims on paperwork rather than on lumber.
Who is there: human referees priced by the day and nothing else — the national association's inspector at $890 a day for members and $1,335 for non-members plus travel, and equivalent bureaus for softwood and the west coast. 🔴 Where they stop: they adjudicate; nobody prepares. The inspector arrives at the end of the argument. Everything before that — assembling the tally, computing what the difference is worth under the rules, deciding whether the claim is even procedurally valid — is done by hand by whoever picked up the phone. And the procedural trap is severe: if the buyer does not hold the shipment intact, report within fourteen days and furnish a piece tally, then in the trade's own words "the buyer must pay the seller's original invoice — no exceptions." A claim can be lost on a calendar rather than on lumber, and nothing exists to stop that. The association is explicit that it takes no responsibility for settlement, which deliberately leaves the whole preparation job vacant. Why it is only worth a look: claims are episodic rather than weekly, and the grading judgement itself needs eyes on the board — what we would argue is arithmetic, tally accuracy and procedure. Probably a feature of the timber entry rather than a business of its own. Size: NOT SIZED; the only hard number is the referee's day rate.
THE OP — Sell trucking companies and final-mile fleets the defence half of the cargo-claim argument: an AI that reads every claim filed against them, tests it against the five statutory defences and the delivery paperwork, and drafts the refusal or the counter-offer — because every vendor in this market currently works for the person filing the claim.
Who is there: on the shipper side an open outsourced market of 3PLs selling claims management (AFS, Translogistics advertising over $1.2M recovered annually, Intek, freightclaims.com, Flat World, Redwood) — people typing, attached to a freight relationship. On the carrier side there is software but it is thin: TranSolutions' product, now absorbed into a larger supply-chain group, and one from Cass. 🔴 Where they stop: the carrier-side product page describes a filing cabinet — record, report, resolve, store documents, workflows, reports. There is no AI on it and no judgement in it. A carrier receiving claims today gets a place to PUT the claim and no help DECIDING it, while the law gives it five specific defences that require reading the bill of lading, the delivery receipt exceptions, the packaging description, weather and temperature records and the photographs. Honest caution: a carrier's claims desk is also a customer-relationship desk — paying a small claim to keep an account is a commercial decision, so the product recommends and a person signs. Size: NOT SIZED. Two widely repeated figures (a 30% abandonment rate and a $50bn annual cost) were traced to a blog with no study behind them and are flagged as unusable. What would settle it: ask two regional carriers how many claims they received last month, how many they refused in writing, and who wrote the refusals. "We pay most under $500 because arguing costs more" — that threshold IS the product.
NO OP YET. The arguing half — the usual gap in this shape — is already shipped here: the main platform advertises automatic exception letters, sits on the car owner's side, and manages insights for over 3,000 fleets. Who is there: Transmetriq's CarLogix; the railroad industry's own body publishes the rulebook and price master free; IntelliTrans, Kaleris, Wabtec and RAS sell outsourced audits. Why it is not interesting: unusually, almost nothing is left — the incumbent is on the correct side and already argues, the rulebook is public so there is no privileged knowledge to sell, and the buyer universe is small (roughly 600 short lines plus lessors). Note the original kill said "already crowded", which under the corrected brief is not a kill; this one fails on the right grounds. What would change it: count private fleet owners with fewer than a hundred cars. Low hundreds keeps it closed; thousands makes the small-owner version worth a second look.
THE OP, narrow — Sell plants served by SHORT-LINE and REGIONAL railroads the demurrage argument the market skipped, because the federal invoicing rule that made this job tractable applies only to the giant railroads. Who is there: the most crowded named category in the catalogue — three no-win-no-fee services (Demurrage Defense Company, Eller, Demurrage Recovery Group), TransAudit, and five software products, with two new launches in the last three months and a GPS-based settlement standard arriving in October 2026. 🔴 Where they stop: at the Class I boundary, and it is a real line with a legal definition. The 2021 rule forcing railroads to disclose billing detail and provide machine-readable access applies only to railroads above roughly $900 million of revenue. Every smaller railroad sits outside it — so their customers still receive the opaque invoice the rule was written to fix, and the software vendors' whole approach depends on structured data those railroads are not obliged to produce. Why it stays weak: short lines are generally less aggressive about demurrage, the two small no-win-no-fee firms would take one of these customers tomorrow, and this is nine players deep. What would change it: find out how many short lines exist and whether their customers get demurrage bills with no detail at all.
GENUINELY CLOSED, and argued rather than asserted. Who is there: nine no-win-no-fee firms doing this exact job for this exact customer — one claiming more than 10,000 clients nationwide, another with two decades of testimonials from $17,000 to over $500,000 — plus ten certificate-tracking products preventing the problem and eight companies pointing AI at it. Why it is closed on three independent grounds, any one of which would be serious. (1) A well-resourced field does this exact job well and charges nothing unless it recovers — which is also the pricing model we would have used, so we would arrive second with an identical offer. (2) The frequency test fails on its face: the audit is annual. The continuous job underneath — collecting a certificate from every subcontractor all year — is exactly what the ten certificate vendors sell. (3) AI is not really the lever: the genuinely AI-shaped work is matching a name in a ledger to a name on a certificate and checking dates. A broker's own handbook says the fix is to produce the certificates, which clients likely have, just not organised. That is a filing problem. What would change it: only discovering that the real cycle is not annual — monthly self-reporting or pay-as-you-go premium reporting would reopen it. One conversation with a premium auditor answers that.
14 opportunities · 5 strong
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 2.1 | Check the weekly uniform bill against the signed contract | Multi-site restaurants, hotels | 🟢 | 2·4·5·4 | monthly human auditors |
| 2.2 | Rebuild the day a builder deducted money for | Trade contractors | 🟢 | 3·3·4·4 | forward-capture tools |
| 2.3 | Read the ugly commission statements and chase what is missing | Sales rep agencies | 🟢 | 3·3·4·5 | typing wizards |
| 2.4 | Re-run the milk cheque against published federal prices | Large dairy farms | 🟢 | 5·3·4·4 | co-op side only |
| 2.5 | Audit the mill's timber settlement statement | Southern wood dealers | 🟢 | 4·3·4·4 | seven forestry systems |
| 2.6 | Find idle rented machines still being billed | Contractors renting widely | 🔵 | 2·3·5·3 | four recent entrants |
| 2.7 | Dispute distributor billback charges, not just report sales | Independent drinks brands | 🔵 | 3·2·4·4 | reporting dashboards |
| 2.8 | Audit the concrete bill for unsupported extras | General and concrete contractors | 🔵 | 4·2·4·3 | seller-side only |
| 2.9 | Run the weekly uniform reconciliation the consultants cannot | Mid-market multi-site | 🔵 | 2·3·4·4 | quarterly consultants |
| 2.10 | Recover wrong booking commissions beyond the big three | Independent hotels | 🔵 | 2·2·5·3 | one good incumbent |
| 2.11 | File the supplier claims that were never submitted at all | Mid-market distributors | 🔵 | 2·4·3·4 | seven rebate platforms |
| 2.12 | Apply the dock labour contract to pay AND to the vessel bill | Stevedores, terminals | 🔵 | 4·2·5·3 | worker-side app |
| 2.13 | Read port vendor invoices against the estimate | Small ship agencies | 🟡 | 3·2·3·5 | owner-side giant |
| 2.14 | Reconcile hospital linen losses | Hospitals, laundries | ⚫ | 1·3·4·2 | free from the laundry |
THE OP — Sell multi-site restaurant, hotel, food-plant and dealer groups a machine that reads every weekly uniform and linen invoice line against their own signed contract and their own delivery tickets and produces the credit claim automatically, on a share of what comes back — because the cost-recovery consultants are people reading bills once a month, and the bill arrives every week.
THE OP — Sell trade contractors the BACKWARD half nobody sells: a deduction lands today about a morning four months ago, so read their subcontract to see whether the builder was even allowed to charge it and whether they served the notices their own contract requires, then rebuild that morning out of the foreman's phone, the group chat, the daily reports, the delivery tickets and the time sheets, and produce the rebuttal before the final payment is released — on a share of what gets put back.
THE OP — Sell independent sales-rep agencies the reading of the twenty ugly commission statements their own software openly refuses to read — and the chase for what is missing from them — paid on a share of the commission recovered.
THE OP — Sell the dairy FARMER, not the co-op, a monthly check on his own milk cheque: re-run the federal component-pricing arithmetic against the plant's own reported test results, price the haulage against what the neighbours pay, and flag every deduction he cannot account for, on a share of what gets corrected or renegotiated.
THE OP — Sell Southern wood dealers and logging contractors the audit of the mill's settlement statement that no forestry software performs — reading every mill ticket against the woods load report, the tract's contract terms and the mill's own deduction rules, and handing over the short list of loads that were misclassified, credited to the wrong tract, over-deducted or never paid at all — on a share of what gets recovered.
THE OP — Sell contractors the JOIN nobody sells: the supplier's bill says the machine was on rent, the tracker and the site diary say it sat still, so produce the credit demand across every supplier at once, on a share of the credits. Who is there: genuinely crowded and recently so. TrueRent sells software plus an outsourced rental coordinator claiming 15–30% savings, but names three Ohio cities as its focus. ProcIndex (Y Combinator-backed) sells AI accounts-payable agents for construction. Nimble sells the managed version to large contractors. Yardz, Tenna and Clue hold location data. United Rentals' own listing inside the software these contractors already run promises to stop costly rental overages. 🔴 Where they stop: three narrow but real stops. The bill readers hold no evidence a machine sat still, so they can find a rate error but not an idle machine. The tracking platforms never read the supplier's bill. And each supplier portal sees only its own fleet, so a contractor renting from five suppliers has five windows and no combined view. Why it is only worth a look: this is the least blue water in the catalogue — four separately funded parties arrived in eighteen months, the largest rental company has shipped a version of the pitch, and reading a messy PDF bill is now cheap commodity capability. Half the job is not AI at all. Size: partly sized — one advertised Equipment Coordinator role at $25.00/hour full time is roughly $65,000 all-in; the leaked-money half is NOT SIZED because three vendors selling the same cure publish figures that disagree by a factor of four. What would settle it: ask three contractors renting from four or more suppliers whether anyone reconciles all four bills against one picture of where the machines were. Then test whether the location proof even exists — if the machines are untracked and call-offs were phone calls, there is nothing to join.
THE OP — Sell independent drinks brands the argument, not the dashboard: read every chargeback and billback invoice their distributors send, match each line against the deal actually agreed and the sales that actually happened, and dispute the ones that do not reconcile, paid out of what we claw back. Who is there: on consolidation, four named vendors (Overproof, Liquid Vision, Datex, Ontrak); on reconciliation, Andavi and Park Street; American Spirits Exchange publishes a guide to billbacks; generic deduction vendors exist but are built for consumer-goods suppliers dealing with retailers, not the alcohol three-tier structure. 🔴 Where they stop: Andavi tells you the size of the hole in its own marketing and then does not fill it — most winery accounting staff "plow through spreadsheets" and typically only about 85% of invoices can be matched up against depletions. What Andavi sells is a pricing-management product giving visibility. That is a better spreadsheet with a better memory. It is not somebody reading the 15% that will not match and telling the distributor the charge is wrong. The consolidation vendors serve the brand's marketing question (which market is working), not its money question (am I being charged for promotions that did not happen). Why it is only worth a look: the 85% figure comes from a vendor selling the fix, and billbacks are reconciled quarterly in some houses, which weakens frequency — though the depletion reports themselves arrive monthly per distributor, so a brand in fifteen states has fifteen a month. What would settle it: one brand's last twelve months of billback invoices and deal records, reconciled by hand. Either a meaningful share fails to match, or the 85% is marketing.
TWO OPS SIT UNDER ONE KILL, AND THE SECOND IS THE EMPTY ROOM. (a) Sell extras-capture to the thousands of small independent concrete producers who will never write a $50,000-a-plant cheque to the vendor that supposedly closed this market. (b) Better: sell the general contractor the audit of the concrete bill — read every delivery ticket and invoice for short-load fees, standby, surcharges and returned-concrete charges, and argue the ones the ticket does not support, on a share of what comes back. Who is there: on the producer side, Command Alkon dominates and its Deliveries product explicitly documents water added, extra charges and returned concrete; alternatives include Sysdyne, Dispatch360, Kunzite, Veloxiti and Scalable. On the buyer side, a search for anyone auditing concrete invoices for the party being billed returned invoice templates, a Texas tax rule and generic construction-audit consulting — not one firm doing this. 🔴 Where they stop: at the price (comparison guides put the dominant system at $50,000+ per plant per year plus six-figure implementation, so a producer with two plants is not buying it) and at the side of the transaction — every product listed is sold to the company producing the invoice. Concrete is bought by the yard on tickets written by the seller's driver, with a menu of extras the buyer has almost no practical way to verify at volume. That is the same structure as freight bill auditing, which is a mature and profitable business on the buyer's side, and nobody has pointed it at concrete. AI is moderate, not high — much of it is extraction, and the judgement is narrow. What would settle it: one general contractor's last quarter of concrete invoices and matching tickets, hand-checked for unsupported extras.
(a spin-out found while reopening the hospital linen candidate)
THE OP — Sell mid-sized multi-site businesses the weekly line-by-line argument with their uniform and linen bill — matching every garment, mat, lost-item charge and surcharge against what the driver actually delivered and collected — on a share of what we claw back. Who is there: the same cost-reduction consultants as 2.1 — P3, Fine Tune, UniformBright, The Laundry Guy — all serving the customer being billed, all on contingency, none publishing a price. 🔴 Where they stop: Fine Tune is the closest and says in its own words that it audits "invoices and delivery records to verify counts and identify overcharges and unapproved items", which IS the job — but it is people, on a project, for large accounts clearing category minimums, sharing savings over a long engagement. P3 shares savings for 12–60 months. The volume defeats them: one consultant's own material says a weekly invoice for a 50-site operation carries 30 or more line items per location. That is 1,500 lines a week that no human team reads in full. Why it is only worth a look and not strong: unlike 2.1 this is head-on against a competitor doing nearly the right thing; the wedge is cadence and client size, not capability. What would settle it: will Fine Tune or P3 take a customer spending $150k a year across twelve sites, or do their minimums exclude that whole band? And in one month of real invoices and tickets, how many of the 30-plus lines per site per week are actually wrong?
THE OP — Sell independent and small-group hotels the commission argument the existing recovery firms do not cover — the wholesaler and bed-bank channel, group and corporate rate mismatches, and the booking sites outside the big three — on the same share-of-recovery terms, to the 97% of US properties the leading vendor has not signed. Who is there: OTA Recovery is a real and good incumbent — audits every reservation, files the disputes itself, reactivates closed virtual cards, charges purely on performance with no upfront fee, and states it serves 1,500+ properties with city-specific playbooks. Also OTA Systems and Xquic. Separately, a rate-monitoring category exists (RateGain, Lighthouse, SiteMinder, Mews) but that is monitoring, not recovery. 🔴 Where they stop: three checkable places. Channel — the incumbent names three booking sites and says nothing about the wholesaler and bed-bank layer, where the published problem is discounted wholesale rates resold publicly so the hotel pays high commission on a cheap room. Segment — hotels only; nothing for holiday-rental management companies with the identical problem on a faster-growing channel. Penetration — 1,500 properties is a small fraction of the US estate. Also nobody joins the monitoring vendors to the recovery vendors, which is the same read-versus-argue split that recurs throughout this catalogue. Honest read: this is a wedge behind a competent, correctly-priced incumbent, not an ocean. What would settle it: audit three hotels' wholesaler statements alongside their big-three statements. If the wholesale channel errs as much or more, the incumbent has left the better half on the table.
THE OP — Run the whole supplier-claim function for mid-market distributors as a service on a share of what comes back: an AI that reads the pile of supplier agreements out of the shared drive and the email, matches every qualifying sale to the claim it should have generated, files them, and re-argues every line the manufacturer short-pays. Who is there: the most heavily vendored candidate in the catalogue — a named category with its own vocabulary and its own data standard. Vistex, Vistaar, IMA360, Enable, Computer Market Research, SolveXia, Iptor; foodservice-specific Meal Ticket, Speedy Labs, Ontrak; manual services from Park Street and Vyas; and adjacent no-win-no-fee auditors like Vendormint at 30% contingency. 🔴 Where they stop: three places, each an opening. They sell a PLATFORM you run yourself — and running it means loading your agreements into it, which is the step that makes it work. Enable's own marketing names the failure: trade agreements "are often lost, fragmented across the business or hard to replicate in ERP systems, which forces companies to leave valuable rebates unclaimed." A $30M distributor with 400 agreements in a shared drive and a sales director's inbox cannot do that loading step, and nobody does it for them. The money-side services that do run it are pointed at the wrong transaction — supplier-versus-retailer deductions, not distributor-versus-manufacturer. And the re-arguing of a short-paid line is line-level work nobody performs. The biggest leak is not the denied claim but the claim never submitted, because nobody joined the agreement terms to the sales ledger. Why worth a look and not strong: any opportunity here has to be "we run it", never "we sell software", and the price has to beat everything from $20 a month to enterprise. What would settle it: ask two distributors how many of their rebate agreements are loaded into a system that automatically generates the claim. "Some, in a spreadsheet" means the never-claimed leak is countable from their own ledger in an afternoon.
THE OP — Sell East and Gulf Coast stevedores the same union-contract pay engine their own dockers already carry in their pockets to catch a short cheque — but pointed at the employer's job: turn last night's gang sheet into both the correct payroll and the correct bill to the shipping line, so guarantees, meal reliefs and equipment premiums get charged to the vessel rather than quietly absorbed. Who is there: on the WORKER side, a real and recent product — PortHQ, a members' app that logs a shift and "does the math the way the ILA contract actually pays", covering straight time, overtime, meal hours at double, hour guarantees, equipment premiums and roll-through shifts, and producing a simulated pay stub. On the EMPLOYER side, searches returned the contracts themselves, a regulator's payroll form and academic comparisons — no vendor, no service, no software. 🔴 Where they stop: PortHQ is the whole finding in one line. It exists so a docker "can catch a short check before it costs you" — direct evidence, published by somebody who built a product on it, that the employer's number comes out wrong often enough to be worth catching. It is explicitly personal: it does not generate payroll and does not bill shipping lines. So the contract's pay arithmetic has been encoded and shipped — to the party who receives the money and to nobody who computes it. The employer loses twice: a rule missed in the worker's favour is a grievance, and a rule missed in the employer's own favour is a guarantee paid out to labour and never charged on to the vessel. 🔴 The weakness is size: the employers' association lists 47 member companies; the true count including local port associations is larger but plausibly under a few hundred nationally. What would settle it: ask one stevedore's controller whether they bill the shipping line the guarantee hours they actually pay out. And check whether the terminal operating systems already do labour costing — that could close it.
THE OP — Sell small independent US port agencies the reading of the vendor invoices, matching every pilotage, towage, berth and launch charge that arrives after a ship leaves against the estimate and the port's published price list, and drafting the variance explanation the shipowner will demand. Who is there: DA-Desk (part of Marcura) claiming 7,000 users, and Base, a port agency product founded 2022. 🔴 Where they stop, and they serve opposite sides: DA-Desk's customers are shipowners and charterers — it audits the agent's account on the owner's behalf — and its page actively tells agents "please don't use this form". It is also explicitly a hybrid of software and people. On the agent side, Base's own page describes an activity log entry showing charges matched without describing any mechanism, and mentions no document reading, no AI and no format recognition. So the reading of the invoice against the published tariff — the actual work — is unclaimed by either side, and DA-Desk is not a competitor at all; it is in the room opposite. 🔴 Why it stays weak: population. The relevant US association publishes no member total and its own staff number 2–10; this is very likely a few hundred small companies at most. What would settle it: count the certified agents and the independent agencies at the top twenty US ports. Over about three hundred real firms and this reopens properly, because the competitive picture is much better than it looked.
GENUINELY CLOSED, and this one is confirmed rather than reopened. Who is there: software (InvoTech, LinenMaster, LinenTech, ABS, Laundry Logic, Positek, Datamars, ClearVu-IQ), services (HHS's ThreadCount priced purely on a share of savings, UniformBright, Medline, Crown), consultants, and the trade association's free toolkit of six audit templates and four calculators. Why it is closed: the laundries give loss-reduction audits, training and tracking software away free to keep the hospital contract; the leading service charges only a share of what it saves; the association gives the templates away. Selling into a market where the same work is available free from two directions is not an opportunity. The reconciliation job is also the detective, not the thief — the residual loss is linen binned, hoarded on a ward or leaving with a patient, which is behaviour, not documents. The honest yield from reopening it was the discovery that the adjacent, larger uniform-rental version of the same shape is NOT served that way — which is now entry 2.9.
1 opportunity of its own; the other 8 members are listed under Groups 1 and 2
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 3.1 | Rebuild a snow season you never captured, to answer a claim | Commercial snow contractors | 🔵 | 2·2·3·3 | forward-capture apps |
THE OP, narrow — Sell northern commercial snow contractors the reconstruction of a service record for a season they did not capture in an app, either to answer a property manager deducting from an invoice now or a slip-and-fall claim letter arriving two winters later. Who is there, and this is a correction to the previous round: Nektyd sells exactly the product the last round said did not exist — GPS pings every 20 seconds, timestamped before-and-after photos, signatures, checklists, "weather logs for snow and liability defense", and shareable proof packets, claiming 1,000+ operators. Aspire claims 70,000 users across nearly 1,500 locations. Certified Snowfall Totals sells meteorologist-verified snowfall data to contractors AND to insurers and property managers at $205 a year for one location. 🔴 Where they stop: everything is forward capture — it works if you were on it that night. A claim about an incident 24 months ago covers a season living in trucks, texts, paper and possibly another vendor's system, and nothing reconstructs that. Nobody reads the contract either: whether a visit is billable turns on trigger depths, per-inch tiers and zero-tolerance clauses, and that interpretation is still a person with a PDF. And nobody serves the party being billed. Why it is weak-ish: the reconstruction is episodic, not repeated; the season is five months long in a limited geography; and the anchoring weather evidence retails at $205 a year, which caps what this industry expects to pay for proof. What would settle it: ask five contractors how many claim letters or deductions they had to answer about work done BEFORE their current software, and what each cost in hours. One or two a season means it is a favour, not a business.
9 opportunities · 3 strong
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 4.1 | Run the closeout desk that unfreezes the final mast invoice | Mast contractors | 🟢 | 4·3·5·5 | one checklist app |
| 4.2 | File every utility rebate outside California and New York | Insulation, lighting, refrigeration | 🟢 | 4·3·4·4 | three heat-pump startups |
| 4.3 | Build and defend the highway extra-work claim | Heavy civil contractors | 🟢 | 3·4·4·4 | ticket-capture apps |
| 4.4 | Close the gap in an aircraft's records, not just find it | Parts traders, repair shops | 🔵 | 3·3·3·4 | reading-only vendors |
| 4.5 | Build the outbound steel document pack per customer rules | Metal service centres | 🔵 | 2·2·4·4 | fifteen inbound readers |
| 4.6 | Appeal the disaster claim after it is denied or clawed back | Rural power co-ops | 🔵 | 4·3·4·4 | assembly-only vendors |
| 4.7 | Claim damaged-goods credits back from suppliers | Food and goods distributors | 🔵 | 4·2·3·3 | the giant just left |
| 4.8 | Reclaim import duty on exported goods | Importer-exporters | 🟡 | 1·4·4·5 | AI-native, five months old |
| 4.9 | File the same pay application into every builder's portal | Trade subcontractors | ⚫ | 1·3·5·2 | one purpose-built winner |
THE OP — Run the closeout desk for mast contractors: take the crew's photo dump and paperwork, judge it against that phone company's own acceptance list BEFORE it is submitted, tell them exactly which shot is missing while the crew is still on site, and get paid out of the final invoice we unfreeze — which is a different business from the $99-a-month checklist app the one named vendor sells.
THE OP — Sell insulation, commercial lighting, refrigeration and mechanical contractors outside California and New York the thing the three funded startups only do for heat pumps in two states: file every utility and state rebate each finished job qualifies for, straight from the photos and invoice the crew already took, and work the rejections — paid on a share of the rebate, well under the 25% the leading player charges.
THE OP — Sell heavy civil and highway contractors the finished extra-work bill and the argument behind it: take their signed daily field tickets and rebuild the claim against the state's own rate book, certified payroll records, delivery slips and allowed markups, then answer the engineer's line-by-line objections — paid on a share of what actually gets approved.
THE OP — Sell aircraft owners, lessees, parts traders and teardown shops the CLOSING of the records gap: take the exception list the existing scanning tools produce and go and get the missing paper out of the manufacturer, the repair station and every previous operator, paid on a share of the value we put back on the asset. Who is there: ProvenAir (AI scanning of maintenance records producing back-to-birth timelines, integrated into an aerospace marketplace since April 2025), Bluetail (digital logbooks that "instantly flag missing data, incorrect entries, and gaps"), SPI Aviation, Aircraft Cloud, plus lease-return advisory consultancies with no published prices. 🔴 Where they stop: all of them sell the READING and stop at the exception list. Bluetail's own material draws the line for us — rebuilding incomplete logbooks "requires significant time and expense, including contacting previous owners, maintenance providers, or OEMs". That sentence IS the handover point. Second, every one sells to the party HOLDING the asset; nobody sells to the lessee facing an end-of-lease compensation claim, who has the deadline and the money on the line. Third, the dispute layer that exists is consultants with no published price — proof of budget, not a closed door. The value anchor is blunt: losing logbooks can cut an aircraft's value by 25% or more. Against it: leasing companies are financial firms and fail the technology test, so the customer has to be the shop, the trader or the operator; and the chasing step is partly human. What would settle it: ask five independent parts traders and one regional airline's records manager who does the chasing today, how long a gap takes to close, and what an unclosed gap costs at sale — then ask ProvenAir directly whether they have added a remediation service.
THE OP — Sell metal service centres the outbound half of the certificate job that the fifteen inbound vendors do not sell: build the document pack for a shipment spanning several heats against that specific customer's own document rules, and draft the conformance certificate when a mill certificate is missing, so the truck leaves on time and does not come back. Who is there: fifteen named vendors doing the INBOUND job — MTR.AI, Aekam, GoSmarter (from £295 a month), Star Software, Algoscale, Openmind, Nox Metals, PixDynamics, Arhivix, Revolution Data Systems, Certivo, Apolo — plus the industry's own business software, which already attaches certificates to orders and inherits heat numbers through cutting. 🔴 Where they stop: almost every one points INBOUND — reading a certificate as it arrives from a mill and checking it against a published standard. The outbound side — building the pack for a customer order spanning several heats, applying that customer's own document rules, and drafting a conformance certificate when one is missing — is claimed thinly or not at all. GoSmarter answers the outbound problem in a single sentence about automatically associating certificate sections with orders. The industry systems attach documents to orders; they do not check whether the pack satisfies what this particular customer demands, and those rules are genuinely per-customer — one large buyer refuses more than one batch number on a single test report. Second stop: price. £295 a month and $400–900 a week are not the prices of a won market. Why only worth a look: thinner than it looks, and it is a contested seam between an existing software feature and a dozen vendors expanding toward it. Size: NOT SIZED — vendor load figures disagree by a factor of thirty. What would settle it: show the last ten outbound shipments at one service centre and count how many needed a customer-specific rule, a hand-written conformance certificate, or held a truck at the dock.
THE OP — Stop selling the assembly of the storm claim, which is crowded, and sell small electric utilities the APPEAL: when the federal agency denies a cost or claws money back after the fact, write the 60-day appeal from the utility's own evidence and the agency's own published record of what it has previously accepted, for a share of what gets reinstated. Who is there on ASSEMBLY (correctly found crowded): disaster cost-recovery consultancies (Hagerty, Tidal Basin, Witt O'Brien's), utility software marketed against the problem, one vendor built for this exact niche but apparently a solo founder offering free pilots, and AI-native entrants including a policy question-answering agent. On APPEALS: one specialist consultancy publishing an article on improving appeal odds, and no productised, contingency-priced appeal-writing service aimed at electric utilities. 🔴 Where they stop: at submission. The whole landscape is built on the premise that if you capture and assemble well enough you get paid — and the industry's own evidence says that premise is incomplete. The co-ops' national body told the federal government in writing in May 2025 that shifting requirements force them "to resubmit claims months after they had first been approved" and that the agency has taken "back millions from unsuspecting communities and applicants." A perfectly assembled pack can still lose money years later, and nobody sells the answer to that moment — the moment with a hard deadline, a defined dollar amount, and therefore a natural contingency fee. 🔴 Two things make appeals better than assembly: the precedent is published and machine-readable — the agency runs an open, downloadable second-appeals record — so the argument can be built from the agency's own history of what it has accepted; and the trigger is a denial letter with a dollar figure and a 60-day clock, which converts the candidate's worst weakness (a bad buying moment) into its best feature. The weakness survives: it is episodic per client, and the co-ops' own five asks of government are about the rules, not better paperwork. What would settle it: download the appeals record, filter it for electric co-ops and municipal utilities, and read off how many appeals a year, what share are granted, and for how much. One afternoon, no money.
THE OP — Sell food and consumer-goods distributors the vendor-claim half of their damaged-goods desk: photograph and describe what came back, match each item to the purchase record and that supplier's own return terms, produce the claim, and argue the disputed credits. Who is there — and this changed materially in our favour. The decades-long giant of grocery reclamation states on its own site that in January 2025 its supply-chain arm was acquired and it has "completed the full divestiture of all post-purchase and retail returns operations", retaining only pharmaceutical returns. On the software side, deduction-management platforms are mature and well funded, naming Ferrero, P&G, Duracell and Hershey's. 🔴 Where they stop: two clean stops. Side of the transaction, textbook: every deduction platform is sold to the SUPPLIER defending itself against claims — its whole pitch is recovering invalid claims made against you. The distributor MAKING the claim against its own vendors is the opposite party in the same argument and has nothing. Customer size: those platforms are enterprise, and a $40M regional food distributor will never be a customer. And whoever now runs the physical warehouse work is in the business of moving and disposing of goods, not arguing money. Honest limit: the advertised job is half physical — assessing re-salability and disposing of what is not — which software cannot do. Only the vendor-claim and disputed-credit half is ours. Size: NOT SIZED; the only hard signal is labour — 263 openings for "reclamation clerk" on one job board and dedicated full-time roles at single regional distributors. What would settle it: what percentage of damaged goods results in a vendor claim, and what percentage of those gets disputed or short-paid? Any reclamation supervisor knows both.
NO OP LEFT. The core job is taken and the taker is formidable. Who is there: Zollback — AI parsing of invoices, bills of lading and system exports, algorithmic matching of imports to exports, electronic filing, and pay only when the refund is delivered, no upfront fees, no minimums — sold direct and through customs brokers, forwarders and accountants. Launched 17 March 2026, claims 15–20% more recovered and 10–15 working days versus 9–12 months, with $10M already processed; founders are a Stanford PhD in combinatorial optimisation, a former head of trade analytics at a global shipping line, and an engineer who built matching engines at a payments company. Plus DutyCalc and several established providers. Why it is not interesting, and this is the entry to be most careful about: the temptation is to point at the $11–15 billion of unclaimed refunds and Zollback's mere $10M and call it wide open. That is wrong twice. The unclaimed figure appears only in vendor marketing and could not be traced to a government source. And an AI-native, contingency-priced, five-month-old, technically formidable team is running our exact business model on our exact job — the rare case where a well-resourced player does this job for this niche and does it well. The 2025–26 tariff environment has also made this the most-hunted recovery niche in the country. The only residue: companies exporting through a distributor or forwarder often cannot produce proof of export and need it reconstructed — a genuine documentary problem, but a component of Zollback's job rather than a market beside it. What would change it: two conversations with manufacturers who export through distributors. Half a day, no more.
GENUINELY CLOSED, and recorded in full so nobody rediscovers it. The job: retyping the same monthly payment application into a different builder's portal for every job, plus the waivers and compliance documents each demands. Who is there: Siteline, launched 2019, describing itself as the only payment-application and waiver software built specifically for trade subcontractors — no per-seat or per-location fee, unlimited users and offices, onboarding done by them with most customers live within two weeks. On the builder side, three major payment platforms publishing comparison content. Why it is closed: a well-resourced player does this exact job for this exact niche, does it well, does the onboarding, and prices by company rather than per seat — so the usual "too expensive for a small sub" opening is not there. AI is also thin: this is mapping a schedule of values into another system's fields with each builder's retention rules — ordinary software could have done it in 2015 and a company did. What would change it: one email to Siteline asking two questions — do you handle certified payroll on public jobs, and what is the smallest customer you take? Certified payroll out of scope plus a high floor leaves a narrow sliver. Otherwise it stays shut.
2 opportunities of their own; six more members are listed under other groups
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 5.1 | Fix and resubmit rejected tyre national-account claims | Commercial tyre dealers | 🟢 | 3·3·4·4 | two dealer systems |
| 5.2 | Read whatever the crew actually sends and make the invoice | Small oilfield, water hauling | 🟡 | 2·2·5·3 | one vendor owns both ends |
THE OP — Sell independent commercial tyre dealers the rejected-claim desk they do not have: take the red rejection queue their tyre software already shows them but will not fix, work out from the original job and the manufacturer's own rules why each claim bounced, correct and resubmit it across every manufacturer they deal with rather than only the one or two with a built-in link, and take a share of the money recovered.
NO STRONG OP. The narrow one: read whatever the crew actually sends — a photo, a text, a scrawled ticket — and produce the clean invoice, for the small service and hauling firms who will never get their crews onto an app. Who is there: eight-plus vendors selling digital field ticketing into exactly this niche (Enverus OpenTicket, Pandell, Spira, Riger, Engage Mobilize, IronSight, Cantrell Jackson, WeighPay, FieldCap). Every search returned vendors before it returned job postings, which is the signature of a solved market. 🔴 Where they stop, narrowly: all of them require the crew to adopt an app in a field with no reception — the pitch presupposes that digital entry happened. Tickets that arrive as a photograph texted at 11pm are outside all of it. And the rejection loop after submission is described as something their software prevents, not a service anybody performs. Why it is weak: the serious incumbent owns BOTH ends of the transaction — the supplier's ticketing tool and the operator's invoicing portal, sold as one suite claiming to cut payment times by up to half. A network with both sides is the hardest kind of incumbent to work around, and the intake gap is a feature they could add rather than a room they cannot enter. The buyers who need it most are also the smallest and least able to pay. What would change it: ask five small service or hauling companies whether their crews actually use the app and what share of invoices get rejected first time. High adoption and low rejection closes it for good.
3 opportunities of their own; seven more members are listed under other groups
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 6.1 | Bill the chargeable work the warehouse floor never invoiced | Independent cold stores | 🟢 | 4·4·4·5 | forward-capture software |
| 6.2 | Switch on rent for tanks nobody is billing for | Propane, welding gas | 🔵 | 3·2·4·3 | barcode tracking systems |
| 6.3 | Stop over-remitting state tonnage fees, and check labels | Fertiliser, feed, pet food | 🔵 | 2·3·5·3 | a thirty-year incumbent |
THE OP — Sell independent refrigerated warehouses the recovery of chargeable work they did and never invoiced — by reading the paper the floor already produces (dock tickets, signed delivery notes, rework sheets, blast-freeze logs, driver wait times) and turning it into billable lines, paid as a share of what we get onto the invoice — because their warehouse software can only bill what somebody remembered to type into it.
THE OP — Sell independent propane retailers and welding-gas distributors a recovery sweep they can buy without buying a system: read their existing delivery history, invoices and customer file and hand back a list of company-owned tanks and cylinders sitting at addresses nobody is billing, paid on a share of the rent we switch on. Who is there: TrackAbout (now owned by Datacor), DataWeld's AcuTrax, TIMS, Datacor itself, Airpinpoint — established, with case studies in the industry's own magazines, prices undisclosed. 🔴 Where they stop, in three places that matter: these are FORWARD-LOOKING tracking systems requiring a barcode or tag on every asset and a scan at every touch — hardware, driver process change, and a rollout. They cannot tell you what is already wrong, only what happens from installation day. The recovery everyone quotes is a BY-PRODUCT of installing them, not a product anyone sells: distributors "discovered accounts with more cylinders than they were billing for" and rental billings then rose three to five percent — but nobody sells that discovery on its own to a distributor not ready to buy a tracking system. Second, the propane half is thinner than the welding-gas half. Third, all of it is software the distributor buys and runs. Why it is attractive despite that: the money is rent, which recurs monthly forever once switched on, so a one-week read of existing records produces an annuity. Honest caution: AI is moderate here — messy address-and-customer matching, real work but not heavy judgement — and the three-to-five-percent figure comes from a page that refused to open on re-check. What would settle it: one retailer lets us read twelve months of delivery records against their billing file and we count tanks at addresses with no rent line. A week, and the test IS the product.
THE OP — Leave the registration half to the thirty-year incumbent and sell only the tonnage half: compute what each state is actually owed from the company's own sales data instead of rounding up to stay safe, file it, and take a share of the over-payment we stop — plus the label-against-fifty-states check that fifteen firms currently sell as human hours at $300 a label. Who is there: fifteen named — Kelly Registration Systems (roughly thirty years old, claiming 40% of all US product registrations are submitted through it, and running several state agriculture departments' own portals), AgSTARS, the fertiliser trade association's own paid subscription, Fertton, PAW, plus ten consultancies selling people. Pet Food Compliance publishes an actual rate card: $200+ per state to register, $175+ to renew, $25+ per state per tonnage report, $300 per label audit, $325 an hour. 🔴 Where they stop: registration is genuinely closed and reporting is admitted-unfixed by the industry's own trade association, in its own words — members said "the varying frequencies, rates, and methods to reporting were overwhelming" and that they feared "paying more than they needed to because they did not want to worry about violating tax codes." That sentence is the whole opportunity: companies are deliberately over-paying state governments, every reporting period, on purpose, because working out the right number is harder than overpaying. The association's answer was to sell a subscription, and it is fertiliser-only. Everyone else sells this as people typing, priced per unit. Honest weakness: the arithmetic core is ordinary software that shipped decades ago, so this scores badly on AI being the lever, and the customer relationship is already owned. The genuinely AI-shaped slices are the label-versus-fifty-states check (sold by hand at $300 a label), matching each product to each state's own category names, and watching fifty states for rule changes. What would settle it: ask five feed or pet food companies filing in more than fifteen states whether they calculate the exact tonnage amount or round up to be safe — and by how much. The association says the practice exists; nobody has ever put a number on it.
8 opportunities · 3 strong
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 7.1 | Build the refrigerant leak file backwards from contractor invoices | Cold stores, food plants, rinks | 🟢 | 3·4·5·5 | going-forward data entry |
| 7.2 | Chase every waste manifest closed inside the legal clocks | Multi-site waste generators | 🟢 | 3·3·5·3 | quarterly reconcilers |
| 7.3 | Rule monthly on what the organisations you fund actually spent | Grant-funded agencies | 🟢 | 3·3·4·5 | front-of-lifecycle software |
| 7.4 | Keep the chemical register right all year, not each March | Small industrial plants | 🔵 | 2·3·4·4 | filing-only platforms |
| 7.5 | Keep the fuel tank monthly file for non-retail owners | School yards, marinas, plants | 🔵 | 2·3·4·2 | one enterprise vendor |
| 7.6 | Check the contractor's training paperwork at the quarry gate | Aggregate producers | 🔵 | 3·3·4·2 | own-employee systems |
| 7.7 | Read the paper coming off the boats and find what is missing | Small towing fleets | 🟡 | 1·2·4·4 | strong installed vendor |
| 7.8 | Rewrite the safety document the reviewer just rejected | Industrial contractors | 🟡 | 1·2·5·4 | funded AI-native at $69/mo |
THE OP — Sell the owner of a mid-size refrigerated site the one thing nobody sells them: go and get their refrigeration contractor's service invoices, read them, rebuild the legally required leak-rate file backwards from that paper, and keep it current every month — because the vendors in this space sell a going-forward data-entry system to an enterprise compliance officer, or a field app to the contractor, and neither will chase a stack of PDFs out of a contractor on behalf of the owner who is the party the government actually fines.
THE OP — Sell mid-size companies that ship waste from many sites a standing chase-and-close service on every shipment: watch each manifest against the regulator's own clocks, chase the disposal facility ourselves when the signed copy does not come back, judge the quantity and description mismatches, and file the reports electronically — replacing a waste broker who reconciles quarterly, which cannot meet a 45-day duty by arithmetic.
THE OP — Sell pass-through grant recipients — workforce boards, community action agencies, councils of governments, housing and transport authorities — an operated monthly review of the organisations they fund: read every sub-recipient invoice, receipt and payroll register against that grant's rules and the approved budget, and return an allowable / not allowable / needs-a-question ruling with the file already assembled for the auditor, because the grants platforms they already pay for collect these documents and never read them.
THE OP — Sell the small manufacturer with real chemicals and no environmental staff the year-round chemical register the filing software assumes he already has: read his purchase invoices and his suppliers' safety data sheets, keep the two matched, and hand the March filing over finished — and sell it now, because every safety data sheet in the country is being reissued between now and 2027 under a new hazard standard, so every facility's hazard categories have to be re-derived from paperwork that has not arrived yet. Who is there: moderately crowded at the FILING layer — Encamp (the most visible specialist, monitoring thresholds and submitting automatically), ERA Environmental, HazConnect, the government's own free tools, and a long tail of regional consultancies doing it as a once-a-year job. 🔴 Where they stop: at the form. Every one takes a chemical list and turns it into a correctly formatted filing, and every one assumes the customer arrives holding an accurate list. Building that list is the actual work, and the proof is a customer review of the market leader: "the least convenient part of this software is the amount of time that was spent doing things by hand, as new materials had to be entered by hand." The consultancies do assemble the list, but as an annual project by a person at consultancy rates — which proves the budget and sets the price to beat. 🔴 The honest weakness, and it is serious: the filing happens ONCE A YEAR, which fails the frequency test outright. Enforcement is thinner than the penalty numbers imply — one state reviewed only 29% of the reports it received and found problems in 4% of those. The business only works if what is sold is the continuous register and the filing is a by-product. What would settle it: take one manufacturer's twelve months of purchase invoices and its safety data sheet folder and see what share of line items match automatically to the right sheet and hazard category. Above 90% is a product with a person handling exceptions; 60% is a person with a helpful tool, which is the wrong shape.
THE OP — Sell the small non-retail fuel-tank owner — the school bus depot, the ready-mix plant, the city yard, the marina, the waste hauler with a fuelling island — a compliance file that keeps itself: read each site's gauge printout and paper walkthrough sheet every month, decide whether that month passes under that state's rules, chase the sites that sent nothing, and have the last year's file ready the day an inspector walks in. Who is there: one strong, well-funded vendor. Titan Cloud sells compliance software plus a managed service explicitly for "fuel operators managing compliance across multiple locations, including convenience retailers, fuel marketers, fleets, and other organizations with large fueling site networks", reporting over 8,000 mobile inspections a month, and claiming over 60% of global convenience stores. Underneath it, the gauge hardware already runs and prints the monthly test. 🔴 Where they stop: at a customer size and a business case, not a capability. Their pitch is a risk-and-profit story — shrinkage, wet-stock loss, alarm dispatch — that only pays back where fuel is the product. A city yard with four tanks does not lose money to shrinkage; it has a pure regulatory obligation and a clerk with a binder. Their managed service is also explicitly positioned as complementing their software, so it is not sold to somebody with no platform. The hardware vendors stop at the tank: the gauge prints the result; nobody reads the printout, decides whether the month passes under that state's rule, or notices that three sites have sent nothing for two months. 🔴 The honest weakness, stated rather than dressed up: the AI lever is the thinnest of the strong-shape members. A gauge printout is fixed-format text and the walkthrough is a checkbox form; most of this is careful automation with a service wrapper. The consequence is real though — an inspector can red-tag tanks and stop fuel deliveries, which stops the operation. There are 534,189 active tanks at about 190,224 facilities. What would settle it: count the non-retail multi-site owners in two states from the public facility register — a morning's work that decides whether this is a business or a rounding error. Then ask three of them who keeps the monthly file.
THE OP — Sell multi-pit aggregate producers the gate: before any contractor's crew sets foot on any pit, read whatever training certificate that contractor's office emailed over, decide whether it actually satisfies the rule, and keep the site's whole training file inspection-ready — because the training vendors sell courses and a system that tracks YOUR OWN employees, and the thing that shuts a pit is somebody else's employee arriving with somebody else's paperwork. Who is there: training content and learning-management vendors (HSI, KPA, Vector, Mine Safety Center, RedVector) selling courses, plan builders and completion tracking; generic health-and-safety platforms; consultancies writing plans; plus SyncMine and a fleet product that both track certificates and expiry dates. 🔴 Where they stop: inside your own company. A learning system tracks the training it delivered to people on its own roster — a closed loop that works. The rule does not respect that boundary: a withdrawal order is issued to the EMPLOYER of any untrained miner, including independent contractors, so the exposure walks through the gate every morning in a subcontractor's pickup carrying a certificate produced by a different company by a "competent person" the mine operator has never met and is nonetheless supposed to have judged competent. Nothing found joins those two sides. Why it is only worth a look: the record-keeping half fails the frequency test (annual refreshers, quarterly reports) and is mostly tracking and chasing, which is software rather than judgement. The AI is narrow but real: reading a stranger's certificate in an arbitrary format and deciding whether it covers the task he is here to do. Size: NOT SIZED per company; penalty anchors are small (roughly $9,000 and $15,000 in two published examples, both from a training vendor's blog) and the real cost is a stopped crew. What would settle it: sit at the gate of one multi-pit producer for a week and count contractor arrivals and how each one's paperwork is checked. Thirty crews a month makes it real; four a quarter kills it.
NO OP YET — and the reason is size plus a real incumbent, not a missing gap. The only honest op is "read the pile of paper coming off the boats and tell the shore office what is missing before the auditor does", and it sits directly next to a strong installed vendor. Who is there: Helm CONNECT, covering certifications, forms, document management, audit planning and tracking, corrective actions and dashboards — sold into this industry for years and explicitly positioned into this regulation as it came in, with a partner delivering customised implementations through it. Plus classification societies and consultancies selling the audits themselves. 🔴 Where they stop, fairly: Helm is a container. It stores, schedules, tracks and dashboards what has been entered. It does not read a stack of paper that came off a boat with bad signal and tell you that a drill record is dated outside its interval, that a crew member who joined forty days ago has no familiarisation record, or that a corrective action was signed off with a sentence that does not close it. But that gap is narrow and sits inside the incumbent's obvious roadmap rather than beside it. Why it stays weak: more than 5,000 US towing vessels must hold a certificate and close to 70% took the safety-system route, but the company count — the number that actually decides this — could not be established and is plausibly in the low hundreds. Half the work is also a person chasing boats. ⚠️ One correction: the corpus claimed the rules "explicitly permit contracted individuals to conduct internal audits as if they were employees"; the section says auditors must be trained and independent, with a small-organisation exception. The opening is still real; the reason is different. What would change it: one number — how many US towing companies operate under the safety-system option. Under about three hundred and it closes.
THE OP, narrow — Sell industrial contractors the one thing the new entrant explicitly refuses: read the reviewer's actual rejection note on your actual safety document and rewrite that document to pass, across all four vetting websites at once, priced per pass rather than per month. Who is there: PrequalPilot, an AI-native startup aimed at exactly this customer — "small to mid-size specialty contractors (5-100 employees)" — at $69, $149 and $299 a month, with AI-drafted answers to forty common safety questions and automatic certificate reading. Private alpha February 2026, shipping by April 2026. The vetting platforms themselves are shipping AI: one announced "the industry's first generative AI risk assistant" in September 2025, another shipped AI findings in June 2026. Six more AI players circle it, ten-plus human services sell the same work (one at a flat $149 per safety programme), and a comparison-shopping content layer ranks them. 🔴 Where they stop: PrequalPilot supports one platform fully, a second in beta on higher tiers, and lists two more as roadmap — so a contractor running four accounts is half served. It also states plainly what it will not do: on-site investigations, complex audits and custom policy writing — and custom policy writing is the actual failure point, because reviewers reject written programmes for missing citations, for saying "should" where they demand "must", and for missing a company name or date. PrequalPilot ships forty templates; it does not read your specific rejection and fix your specific document. Why it is weak: this is the one candidate where the incumbent test genuinely bites — a funded, shipping, AI-native product aimed at precisely our customer at $69 a month, both platforms shipping their own AI, and a price war starting. The remaining cracks are features inside somebody else's market. What would change it: how many contractor firms maintain three or more platform accounts at once? Mostly one and it is closed.
9 opportunities · none strong · four came back closed or weak, which is the warning
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 8.1 | Own the county and city layer of oversize load permits | Heavy haul, cranes | 🔵 | 4·2·3·4 | states are taken |
| 8.2 | File every state's monthly tobacco shipment and excise return | Traditional tobacco wholesalers | 🔵 | 3·2·5·3 | one unreachable vendor |
| 8.3 | Turn an abandoned vehicle into a saleable title, in any state | Collision, RV, marine groups | 🔵 | 3·3·4·3 | tow-industry software |
| 8.4 | File the union benefit money into every fund's own portal | Unionised contractors | 🔵 | 4·3·5·3 | payroll calculates only |
| 8.5 | Answer the court, not just deduct the pay | Multi-state hourly employers | 🟡 | 2·2·5·2 | payroll platforms |
| 8.6 | Chase the doctor who has not signed the death certificate | Funeral homes | 🟡 | 2·1·4·2 | one AI incumbent |
| 8.7 | Research and file a sign permit | Sign makers | ⚫ | 1·2·4·4 | three AI products |
| 8.8 | File backflow test reports to any water authority | Testers, property owners | ⚫ | 1·1·4·2 | eleven vendors, $29/mo |
| 8.9 | Report daily scrap purchases to police | Scrap yards, pawn shops | ⚫ | 1·1·3·1 | connector already ships |
THE OP — Sell heavy-haul carriers and crane firms the county, city and township half of permitting — the layer every state-level product and directory stops at — as a run-it-for-you service, because the national picture of what three thousand local jurisdictions each demand does not exist anywhere and reading it into existence is the product. Who is there: the STATE layer is thoroughly crowded and the previous kill was right about that — ProMiles (whose product already populates multiple state systems from a single application, and which was to be in 23 states handling ~57% of all US permits), oversize.io, GOTPermits, Bestpass, plus eight vendors selling to the states themselves, plus AI arrivals including the trade association's own assistant launched March 2026 and a bureau running managed AI agents. 🔴 Where they stop: at the state line, and that is the whole finding. Every directory covers states and stops — 50-state directories, 46-state services, nothing local. A search for a national database of LOCAL permit requirements returns nothing; the published guidance simply says you must get permission from each county, city or township on the route and points people at a permit service. The human bureaus that do cover local are people on phones, which proves the budget. And the layer underneath is genuinely awful, by the industry's own account: local offices have "multiple priorities, fewer staff, and less (or outdated) technology"; in one state only 15 of 53 counties joined a consolidated programme; local permits take three to five days on their own; one city's crane permits run 15–20 business days. The incumbent that just deployed AI published exactly what it could not hand to a machine: complex multi-state permits, superloads, pricing decisions, refunds, upset customers, legal interpretation, and all final approval. Two honest weaknesses: a good share of local counterparties only answer the phone or take cheques, so this cannot be machine-only; and the trade association has campaigned for harmonisation for decades without success. A quieter adjacent customer: crane and rigging firms, whose permitting is almost entirely municipal — street closures, crane permits, pavement occupancy — so the local layer is not their tail, it is their whole job. What would settle it: for one real 800-mile route, count how many local jurisdictions require their own permit and how many publish requirements online at all. Two days at a desk.
THE OP — Sell traditional wholesale tobacco and nicotine distributors a filing desk that builds and submits every state's monthly shipment report and every state's monthly excise return out of their own order data, in each state's own format, hitting every deadline — sold on the consequence rather than the labour, because getting it wrong puts you on a federal non-compliance list and carriers then refuse to ship for you at all. Who is there: thin and unresolved. Avior Tax publishes state-by-state guidance — I could not open their site, so what they actually sell is genuinely unknown, not cleared. Token of Trust sells compliance tooling but its own checklist is titled for vape retailers, i.e. aimed at the e-commerce population that is excluded here. 🔴 Where they stop: unresolved, and said so rather than guessed. What can be stated: the one incumbent whose material is readable writes for vape retailers, not wholesale distributors — two different businesses with two different systems. The previous kill said the affected population is mostly online direct-to-consumer sellers. That is HALF right: it holds for the shipment report, which bites hardest on people mailing to consumers. It does not hold for the state excise return, which every traditional wholesale distributor files regardless. The shape is close to ideal — forced by regulators, monthly, hard legal deadlines, the same order data reshaped into forty-plus incompatible formats, and a penalty that is not a fine but being cut off from carriers entirely. The problem is population and it is honestly unknown: the mid-market wholesale distributor count could not be established. A large share of this is also reformatting known fields, which is automation rather than AI. What would settle it: two cheap facts — the member count of the wholesale distributors' association, and whether Avior Tax actually FILES the returns or merely publishes rules and sells tax rates. Either answer moves this decisively, and both are a morning's work.
THE OP — Sell multi-location collision, RV, marine and heavy-truck repair groups the abandoned-vehicle title recovery nobody sells them — the whole chain from owner search through certified notices to the filed title application, across every state they operate in — on a per-vehicle fee out of the recovered asset. Who is there: Autura's TowLien is entrenched and genuinely good at its slice; Recovery Title Solutions is a manual service from New York and New Jersey that does serve collision facilities; plus a scatter of single-state bureaus, and one state that actually requires an approved third-party notification vendor. 🔴 Where they stop, read off the leader's own page: wrong customer — TowLien is sold to tow operators and impound lots and bolted into towing software, which a collision group with 40 shops does not run, so the product is not addressable to them at all. It stops before the hard part — their own FAQ says "TowLien does not issue titles. It helps you search for interested parties, send required notifications, and build the digital print packet to apply for title or salvage. The state issues the title." The searching and mailing is the mechanical half; the title application, fifty different forms and fifty evidence packs, is handed straight back. And geographic holes — approved in 39 states, so eleven are uncovered by the market leader. Why it is only worth a look: frequency. A 40-shop collision group might have a handful of abandoned vehicles a month, not forty a week, so this leans on value per event rather than count. The strength is that each event is a whole vehicle and the alternative is an asset sitting in a bay indefinitely. Size: the leader publishes no subscription, only pre-paid per-search and per-mailing rates varying by state. What would settle it: ask three multi-location groups how many customer-abandoned vehicles are on their lots right now, how long the oldest has been there, and who is handling the title. "Nobody, they just sit" makes it real.
THE OP — Sell unionised contractors the filing itself: take the numbers their payroll system already produces and put them into every individual trust fund's own form and portal, on every fund's own deadline, paid per fund per month, because the six payroll vendors all generate the report and then hand it back to a human to submit. Who is there: six named payroll products selling union fringe reporting to contractors — Miter, Points North, SmartBarrel, JOBPOWER, LumberFi, MyConstructionPayroll — plus third-party administrators running the funds and accounting firms selling compliance audits to fund trustees. 🔴 Where they stop: they COMPUTE the numbers, they do not FILE them. Miter's own page says it will "generate union fringe reports" and separately offers direct upload to state portals — but that upload claim attaches to certified payroll, not benefit fund remittance. On the fringe side the verb is "generate", and nothing claims submission into an individual fund's system or names how many funds are supported. Meanwhile the funds each run their own front door: one union's carpenters use their own remittance system, another runs its own employer portal, a third is adding self-service fund by fund, and one region's operating engineers require the joint-fund contribution and the union check-off to be remitted separately. Every one is a different destination with a different login and a different form, and that last mile is entirely human. 🔴 A second, unexploited side: the fund administrator receives hundreds of contractors' remittances in hundreds of shapes and must reconcile hours against contributions and flag delinquencies — and the industry's answer today is a periodic payroll audit by an accounting firm, after the fact, on a sample. Nobody reconciles it as it arrives, and that room is emptier. Honest weaknesses: AI is a middling lever — fund forms are mostly structured, so much of this is careful mapping and portal work, with the genuine AI at the edges (reading each fund's own form to build the mapping, and reading the union agreement that sets the rates). And a widely repeated "10-15 hours weekly" figure appears only on vendor pages with no study behind it. What would settle it: ask five contractors how many separate funds they file to each month and who physically submits each one. "Our system exports it and Karen keys it into eleven portals" means it is live.
NO OP YET, and the reason is structural rather than competitive: this job is identical in every industry, so there is no industry to niche into and nothing to learn — the opposite of what this hunt wants. The narrowest possible op, unconfirmed: sell multi-state hourly employers running payroll in-house the sworn answer to the court — the form due in ten to fourteen days that carries the liability — rather than the deduction, which their payroll system already does. Who is there: the payroll platforms, with named products — one sells "Garnishment Administration", another "Garnishment Managed Services", plus a legal-services company selling automation alongside process serving. 🔴 Where they stop: could not be established, and that is said rather than guessed — one provider's managed-services page returned nothing readable and none publishes what is inside versus outside the service. The distinction that would matter is between calculating and remitting the deduction (clearly inside every one) and preparing the sworn answer to the issuing court per state inside the response window — which is where the liability sits, because an employer that fails can be made to pay the whole underlying debt and in some places the payroll manager personally. Why it is weak: with no industry to learn there is no blue ocean to find, no trade association to read, no ancient software to work around. It is a horizontal compliance product in a market owned by payroll platforms. What would change it: get the actual scope document for one platform's garnishment service and find out whether the court answer is included. Excluded everywhere makes this a narrow deadline-driven op worth one more look.
NO OP on the original framing — the typing half is already sold by the incumbent and the filing half is blocked by fifty state governments. The one op left is a different job: sell funeral homes the doctor chase — an agent that works the physician or medical examiner who has not completed the medical half of the death certificate, across whatever mix of phone, fax, portal and email that hospital uses, until it is signed — because the vendors have automated the typing and nobody has automated the chasing, and the chasing is where the days go. Who is there: Passare, already shipping AI that reads handwritten intake sheets straight into the case; Halcyon, which announced in July 2025 the first working link between funeral software and a state death registration system — in one state; a vendor that has supported registration links since 2002 in another state; plus a roll-up of several products. 🔴 Where they stop: at the funeral home's own four walls. Passare's own published list of what its AI cannot do is the honest map — it cannot confirm information is legally correct, cannot guarantee a name matches official documentation, cannot reliably catch subtle errors. And nobody works the outside parties at all. Why it is weak — five things, any two of which would be enough: the average funeral home is a $1.06 million business (15,401 homes, $16.3bn between them, about 75% family-owned); 100–150 funerals a year is two or three a week, which fails the frequency test badly; the AI-shaped half is already shipped; the filing half is blocked by a wall a funded vendor already walked into (its own account describes meeting over two dozen state vital records departments and finding none had a way to connect or a budget to build one); and the whole job lives inside government vital records including social security numbers and a doctor's cause-of-death section, which is out of scope. The chasing shape survives all five and this industry does not.
GENUINELY CLOSED — and it is the fastest-moving example of this whole shape, which is the useful part. Who is there: three AI products doing this exact job for this exact niche — PermitPal (registered September 2024, claims 7,000+ cities and 40+ sign company customers, publishes overage pricing), ClearSign (registered March 2026, publishing $250 per location, $1,500 for ten, $2,000 a month for up to twenty), PermitMySign (registered July 2026) — plus a human service since 2001 and seven national permit-chasing firms. 🔴 Where they stop, narrowly: one covers 14 states out of fifty; one promises a ready-to-submit packet and says nothing about answering a rejection, a correction notice or a variance application, which is where the cost actually sits since a review round takes two to six weeks. Why it is closed: this space went from empty to three competing AI products in twenty-three months, they are differentiating against each other, and one publishes a price list. The only thing to carry forward is the lesson: a public form with a legal deadline attracts vendors early. What would change it: ask ten sign companies who writes the response when a town rejects a submittal. Uniformly "we do, and it takes two weeks" leaves a narrow feature to sell to the three incumbents' customers — still not a business of our own.
GENUINELY CLOSED, and unusually there is nothing left at all. Who is there: seven vendors on the water utility side, five on the tester side, four on the property-owner side. The strongest correction to the corpus: two sibling candidates claimed nobody sells to the tester — that is wrong. Syncta (owned by a large plumbing manufacturer since 2018, published from $29 a month) is built specifically around the problem those candidates called unserved: testers work with multiple water authorities, each with its own requirements, and reports get rejected when they do not match, so it generates the correct format for each one. Another product lets a tester "switch between all your registered water jurisdictions with one login". A third maintains "a library of hundreds of local water company backflow forms" and adds a missing one free. Why it is closed: all three parties already have vendors, the price is competed down to about $29 a month, and the buyers are either governments or one-truck plumbers. The genuinely thin sliver left — a multi-site property owner with devices across thirty water authorities — sits under eleven vendors and that price ceiling. The lesson, not the candidate, is what matters: three of the four government-form candidates in one researcher's slice came back occupied.
GENUINELY CLOSED — the bridge this candidate assumed was missing has already been built, and it is a push rather than a re-key. The yard's own point-of-sale system transmits transactions to the police database automatically. Who is there: LeadsOnline, the contracted destination in most places, holding over 1.2 billion transactions across all fifty states and integrating directly with the point-of-sale systems these trades already run, with published state-specific setup guides; plus several states running their own programmes. 🔴 Where they stop: two genuinely uncovered edges, neither big enough — yards on old software with no automatic transmission, and multi-state operators whose states demand different fields and different destinations. Neither could be sized, and the sizing is what would matter. Why it is not interesting, with a live factor making it worse: third-party electronic reporting is under legal challenge, so the requirement itself may move — building on a legal obligation a court may remove is a bad shape regardless of how empty the room is. AI is also near zero: the data is captured in a structured system and pushed to a structured destination. The lesson: this is the cleanest example of a public, legally-mandated reporting duty being bridged early by the software the industry already runs, because the vendor's customer would otherwise leave over a compliance failure.
2 opportunities of their own, both strong; four more members are listed elsewhere
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 9.1 | Read what the vendor's insurance ACTUALLY says, not the cover sheet | Builders, owners, hospitals | 🟢 | 4·3·4·5 | eight certificate trackers |
| 9.2 | Chase and judge made-in-America certificates down the chain | Highway, water, transit contractors | 🟢 | 2–4·3·4·4 | one European-facing vendor |
Arguably the single best entry in the catalogue.
THE OP — Sell general contractors, property owners, hospital and school facilities departments and manufacturers the reading of what their subcontractors' and vendors' insurance ACTUALLY says — the coverage documents themselves, not the certificate — against the insurance requirements of the contract that vendor signed, standing, every vendor, every renewal — because every certificate-tracking product on the market checks a one-page document that is legally worthless, and the customer only finds out at the claim.
THE OP — Sell heavy-civil, highway and water contractors — and the fabricators and material suppliers under them — the chase itself: run the domestic-content certification campaign down every material's supply chain on every federally funded job, read whatever comes back, judge whether it actually says the three things the current rules demand, and hand over an audit-proof file per project.
2 opportunities of their own; three more members listed elsewhere
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 10.1 | Judge every aircraft part certificate at the receiving desk | Parts distributors, repair shops | 🟢 | 4·3·5·5 | one generic extractor |
| 10.2 | Read the issued policy against the binder | Insurance agencies | ⚫ | 1·3·5·4 | two vendors at 95–99% |
THE OP — Sell independent aircraft parts distributors and repair stations a receiving desk that reads every airworthiness certificate arriving in the box, checks the issuing shop's approval actually exists and covers that part, flags the internally-inconsistent forgeries the industry started fearing after the 2023 counterfeit scandal, and hands the human inspector a pre-filled accept-or-reject with the reasons attached.
GENUINELY CLOSED on the job as written, and this is one of the rare cases where the original kill was right for the right reason. The job: reading the policy the insurer actually issued, line by line, against the quote, the binder and the expiring policy, to find every coverage, limit, exclusion or location that quietly changed. Who is there: two funded vendors claiming 95% and 99% accuracy across most of the commercial market, one covering nineteen commercial lines representing roughly 85% of premium volume, both with published links into the two major agency systems; plus a mature offshore done-for-you service layer; plus a trade-press feature article about the job, which is the clearest tell a space is visible. 🔴 Where they stop: at the retail agency, checking the insurer's document against the agency's own file — one triangle. Nobody checks a policy against a THIRD PARTY'S CONTRACT, which is the version where somebody outside insurance is the customer and the failure mode is a denied claim on a job they thought was covered. That is entry 9.1, and it is why this entry is kept: the reading engine underneath is valuable and points somewhere empty. What would change it: nothing on the original job.
6 opportunities of their own; five more members listed elsewhere
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 11.1 | Close the delivery route from the paper that comes back | Food and drink distributors | 🔵 | 2·3·4·4 | four full-system vendors |
| 11.2 | Run the estimating desk, do not sell estimating software | Machine shops | 🔵 | 1·3·4·4 | six funded vendors |
| 11.3 | Turn inspection findings into a quote | Fire protection contractors | 🟡 | 1·2·5·2 | leader shipped it May 2026 |
| 11.4 | Read whatever the dentist actually sent | Dental and optical labs | 🟡 | 2·2·4·4 | portal vendors, paper ignored |
| 11.5 | Price the deposition correctly | Court reporting agencies | 🟡 | 2·1·4·2 | forty-year-old system |
| 11.6 | Describe and grade a salvaged part from a photo | Auto recyclers | 🟡 | 1·2·5·3 | six vendors, $0–$95/mo |
THE OP — Sell mid-sized food and drink distributors a settlement closer that reads whatever actually comes back off the truck today — the driver's marked-up paper manifest, a phone photograph of a signed credit slip, the cash and cheque envelope — and closes the route against the accounting system they already own, so they get the found money without buying the whole new route system four vendors are trying to sell them. Who is there: LaceUp, which publishes an article titled "The Hidden Cost of Manual Driver Settlement", plus bMobile Route, Orderwerks and SimplyDepo, and a comparison-content layer ranking them — the usual sign of a real category. 🔴 Where they stop: they all solve it by replacing the whole system, not by reading what exists. Every one sells order entry, mobile invoicing, truck inventory, payments and electronic proof of delivery — a rip-and-replace of the entire order-to-cash flow, sold as a system decision with a training programme and a cutover date. That is a completely different purchase from "close my routes correctly on Friday." The distributor who has already bought a business system, has drivers who will not use a tablet, or runs independent operators who are not employees cannot buy any of the four without also buying a change programme. The paper is still coming back and nobody reads it. 🔴 A side nobody serves at all, named even though it fails the size test: the independent route operator receiving a weekly settlement statement from the parent bakery, full of deductions for route loan repayments, equipment, lost product, insurance, supplies and truck lease. Reading and arguing that statement is the same job from the other end, and the only people currently serving those operators are class-action law firms. They are one-truck businesses, so it is not a niche this hunt can use — but it is the clearest example of an opposite-party room being genuinely empty. Honest note: four vendors actively selling means competing on sales, not insight. What would settle it: ask five distributors with 20–100 routes what closes a route today and how long Friday takes — and whether they have looked at the vendors and said no, and why.
THE OP — Not a quoting product but an outsourced estimating desk: sell job shops the quotes they currently never send at all — we run the drawing-reading AI ourselves and hand back a priced quote within hours, paid per quote or on a share of work won — because every competitor sells software the shop has to operate, and the shop's actual constraint is that it has no estimator left to operate it. Who is there: the most crowded single space in the catalogue — Paperless Parts, CADDi Drawer, Machine Research, Spanflug, Xometry, aPriori, AMFG, all doing AI extraction from drawings and automated quoting. The trade's own forums carry threads asking who is using instant quoting, so the shops are already shopping. 🔴 Where they stop: at selling software the shop buys and runs. Paperless Parts describes itself as for any sized business from a five-person shop to a 500-person enterprise and publishes no price at all — you request a quote for the quoting software. Under this brief that is a different business: we install, operate and take a share. Second, the real constraint is not software, it is people — the estimator who would drive it has retired and not been replaced, with one trade source reporting 36% of manufacturers saying the right talent is harder to find than ever. And the number nobody tracks: the requests that never got quoted at all because the backlog was too deep and the deadline passed. That number is invisible and probably larger than the lost-bid number. Honest read — the weakest of the live entries here: the AI capability is well served by well-funded vendors who have been at it for years, and the only differentiator is the delivery model. A refused figure: a claim that faster responses give a $100 million manufacturer a $20.4 million revenue boost is vendor arithmetic about a company far larger than the target, and is not used. What would settle it: ask ten job shops how many requests they received last month and how many they actually quoted. The gap is the business — and they probably cannot name it, which is itself the sales problem.
NO OP on the job as written. The main half was shipped by the market leader in May 2026 — a named AI product generating ready-to-review deficiency quotes with the contractor's own pricing and a mandatory human review — and the filing half was priced by the incumbent at $5 a report in June 2026, with its own staff doing the typing. Four vendors, plus a public buyer's guide, which is the definition of a settled category. The one op left is on the opposite side: sell the BUILDING OWNER the checking of what the inspection report actually said against what the contractor then charged to fix it. That room is genuinely empty — every vendor sells to the contractor and building owners get a portal to VIEW status — but the buyer is a property manager, the money per building is small, and it lands next to a crowded facilities-compliance market. One residual worth knowing: the incumbent's done-for-you filing service only files into its own portal, covering 1,420 authorities out of 29,452 US fire departments — but that residual is data typing, the price is already set at $5, and it is actively closing. What would settle it: ask three multi-site property managers whether anyone checks the repair quote against the inspection report.
THE OP — Sell independent dental and optical laboratories the intake nobody sells them: read whatever the practice actually sends — fax, phone message, paper slip in the box with the impression, handwriting and hand-drawn diagram included — and hand back a clean case in their lab system. Who is there: EasyRx, describing itself as the only fully integrated platform connecting practices and labs with thousands of each running on it — and saying nothing anywhere on its front page about paper, fax or handwriting — plus five more dental products and three optical ordering portals. 🔴 Where they stop: where the dentist stops. Every product solves this by getting the practice to submit digitally, and the practice has no incentive to adopt and often several labs to deal with. That pitch is more than a decade old and the faxes have not stopped — labs still publish downloadable paper prescription forms for dentists to print. One vendor concedes it in the same breath as the sales pitch: cases entered online flow straight in requiring no re-entry, and labs still handle paper prescriptions which are difficult to manage and may result in missing essential information. So the portal half is taken and the paper half is admitted and abandoned. The lab is the party who suffers and the party with no leverage — a misread becomes a remake, which is the whole job done twice at zero revenue plus shipping. 🔴 Why it is weak, and the second reason is the serious one: the good half of the market is taken, and a dental or optical prescription carries a patient's name alongside clinical instructions, which brushes the regulated-data exclusion. It is not a medical record and a lab is not a hospital, but that is a straight legal answer somebody needs before anyone builds, not after. What would settle it: one lawyer's hour on the privacy question, and ask ten labs what share of cases still arrive by fax, paper or phone. Under about 15% and the residue is too thin.
NO OP — and this one dies on the size of the buyer, not the incumbent, which is worth recording because the file blamed the vendor. The job: turning a completed deposition into a correctly-priced invoice — page counts, turnaround, ordering versus copy party, appearance fees, exhibit scanning, rush surcharges, and each law firm's separately negotiated rates. The numbers that kill it: 2,996 US companies sharing about $3.2 billion, which averages barely $1.07 million per firm, with the top four already holding 29%. Even if the forty-year-old billing software left a gap, the firms are mostly too small to buy anything meaningful and the large ones are consolidators with their own technology. Who is there: the long-established business system, over 150 functions across 13 modules including rate and rush-charge application. The one gap, recorded for completeness: it prices an invoice from data somebody already typed — it does not read the job sheet and work out what happened — and nobody audits these invoices for the law firm paying them, though that lands in the established legal e-billing market. What would change it: count agencies above $5 million in revenue. Under a hundred settles it for good.
NO OP — and the reasons are structural rather than competitive. The job: standing next to a wrecked donor vehicle with a tablet and, part by part, working out how it was built, picking the compatibility number, judging damage against the industry's credit-card rule, and typing it into the yard's system. Who is there: six-plus named players — PartOutPro (published at $0 and $20 a month), a 35-year outsourced service, the dominant marketplace's own tool which has auto-filled descriptions since 2006 and decodes compatibility from the vehicle number, a second industry platform, plus a 2026 AI arrival with five named yard customers. 🔴 The only hole left is tiny: nobody assigns the specific three-character damage code from a photograph inside the professional systems. That is a feature, not a business. Why it is closed: the market has already priced this job publicly at between nothing and about $95 a month per yard, against an industry averaging $3.6 million per location (over 9,000 locations, $32 billion). Even winning it outright is not worth learning an industry for. There is also no headcount to remove — "parts cataloguer" does not exist as a job title because the task is smeared across the dismantler's day. And the trade body's own chief executive named image recognition and automated part descriptions as the industry's opportunity in a July 2026 interview. When the trade body's chief executive is naming your idea in a magazine, the room is not empty. ⚠️ One correction: the previous round cited a consumer marketplace as proof the AI product already ships. Opening it shows a site for people parting out a car in a driveway, with no link to the professional systems — it was not the incumbent it was described as.
2 opportunities of their own; two more members listed elsewhere
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 12.1 | Run the questionnaire desk for American food plants | Food, drink, ingredient makers | 🟢 | 4·3·4·5 | two tiny European firms |
| 12.2 | Read the insurer's answer back into the agency's system | Insurance agencies | 🔵 | 2·3·4·3 | one one-way plug-in |
THE OP — Run the questionnaire desk for American food and ingredient makers: take in every customer form, portal request and spec sheet that lands on their quality manager, answer it from a library built out of their own past answers, and hand back a finished document for one signature — which nobody in the United States currently sells, because the only two companies doing this job are a Dutch one and a Belgian one with about €600,000 between them and European data hosting as a selling point.
THE OP — Sell independent insurance agencies the RETURN leg nobody automates: reading the quote, declaration and endorsement documents the insurer sends back and writing them into the agency's own system — rather than the outbound typing, which one agency software vendor already fills with a browser plug-in for its own customers only. Who is there: heavily. HawkSoft ships a browser extension that auto-fills agency and policy data into insurer websites, with over 50 sites pre-mapped by crowd-sourcing and point-and-click mapping for anything else. Beyond that, insurance agency automation is one of the most heavily pursued areas in applied AI right now. 🔴 Where they stop, in three ways worth naming: the plug-in is ONE-WAY — it fills forms going out and does not read the insurer's answer coming back, which is the half needing judgement. It works only for that vendor's own customers, leaving every agency on a different system unserved. And it is dumb autofill, not reading — field mapping maintained by users, not a system that understands a document it has never seen. Why it is only worth a look: the return leg is a real unclaimed half of a real job, but this is the most contested room in the catalogue — the applied-AI crowd is pointed straight at insurance agencies right now, which cuts against the blue-ocean mandate more than anything else here. A good job in a full room. What would settle it: ask five agencies not on that vendor's system what happens to an insurer's quote PDF when it arrives, and how long retyping takes per policy. Then count how many of the funded insurance-AI entrants already claim it — that number decides this.
3 opportunities of their own, all closed; five more members listed elsewhere
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 13.1 | Reconcile a vending route | Vending operators | ⚫ | 1·1·4·1 | industry data standard |
| 13.2 | Handle cotton gin bale and receipt data | Cotton gins | ⚫ | 1·1·4·1 | one vendor owns all three layers |
| 13.3 | Settle grain tickets against contracts | Country grain elevators | ⚫ | 1·2·4·2 | automated at the weighbridge |
GENUINELY CLOSED — the only one in its slice, and correctly killed the first time. The data is not messy: an industry standard pulls structured records of sales, cash and product movement straight out of the machine, and both dominant platforms ship the variance report as a named standard feature, with a driver-card swipe bookmarking each fill so cash reconciles by driver. There is nothing here for AI to read or judge. The universe is also small — one incumbent's entire customer base is about 200 companies built over 25 years. One thing noticed but NOT verified, recorded so it is not lost: the operator's own commission statements to the host locations where the machines sit are a different job in the opposite direction, and there is no evidence either way about who serves it.
GENUINELY CLOSED, and it meets the high bar rather than the lazy version of it. One vendor occupies all three layers at once: it states that 99% of US cotton bales are issued using its system, it is the leading provider of gin software, and it acquired the main gin accounting product in 2004. Founded 1994; paper receipts were fully phased out of US cotton by 2002. The universe is roughly 520 gins, of which that vendor supplies about 370, leaving perhaps 150 mostly small gins. The obvious flip — serve the GROWER instead — fails on size rather than logic: individual cotton farms sit far below the band. AI is not the lever either; the industry digitised twenty-five years ago and the data is structured and registry-backed. This is what a genuinely closed candidate looks like.
GENUINELY CLOSED, and the vendors' own pages closed it. The established system "connects directly to scales and moisture meters", auto-applies loads to contracts, keeps balances live and settles by loads, lots or dollars. A newer cloud competitor says "shrink and discounts apply automatically, and tickets are printed or emailed before the truck leaves the pit", with tickets instantly updating contracts and bins and settlements posting in one click. Everything downstream of the grade call is automated and the ticket prints before the truck leaves. What is not automated is whether the grade call was right — and the party who would benefit from checking it is the farmer, who is too small and too numerous, while the party who can pay is the elevator, which has no reason to buy a tool that argues with its own inspector. The shape survives even though this instance is dead — timber and scrap are the same shape with the settlement NOT computed at the scale, which is exactly why they are open and this is not. The only thing worth reopening: ask five large row-crop farms whether anyone ever checks the discount schedule on their settlement statements against the contract.
No opportunities of its own — all four members are stronger from the other side of the table, and are listed under Groups 1, 7 and 13.
The one genuinely strong member, monthly monitoring of sub-recipients by grant-funded organisations, is entry 7.3. The other three — car auction arbitration, railcar repair billing, and workers' compensation premium audits — are all better opportunities pointed the other way, and appear as entries 1.5, 1.13 and 1.15.
1 opportunity of its own; three more members listed elsewhere
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 15.1 | Argue the survey against the dredge's own production record | Dredging contractors | 🟡 | 5·1·4·4 | nobody at all |
NO OP AT THIS NICHE — the search is genuinely empty and that is the market speaking, not an opening. The job: reconciling what the dredge says it moved against what the survey says came out of the hole — the daily quantity argument that decides how much the contractor gets paid. The number that kills it: the trade association's membership is twelve large companies, sixteen small companies and two associates — thirty organisations in total. This is one of the rare candidates that dies on population rather than competition. Who is there: nobody doing the reconciliation; survey firms sell the before-and-after surveys and the international dredging body publishes a paper on who pays for over-dredging, so the argument is recognised in the trade. 🔴 The shape is a real opportunity somewhere larger. Strip the water out and the job is: a machine keeps a continuous record of what it did, an independent third party measures the result, the two never agree, and the difference is the contractor's money. That exact argument runs in EARTHMOVING, where the population is orders of magnitude larger and a construction engineering firm writes publicly that contractors commonly raise concerns over pay quantities and that unit-price earthwork must be carefully structured to protect against quantity disputes. Nobody in this corpus has researched it. What would settle the bigger version: read state highway earthwork measurement manuals and ask three earthwork contractors how often the engineer's quantity and theirs disagree.
| # | The op, in a line | Who buys it | Status | Ocean·Size·Same·AI | Who is there |
|---|---|---|---|---|---|
| 16.1 | Catch what the payoff letter is about to leave off | Property management companies | 🟡 | 2·2·4·3 | fulfilment services |
THE OP, and it does not survive — sell property management companies a before-you-sign omission check: read the violation log, the board minutes and the community's own governing documents and tell them what they are about to leave off the payoff letter and therefore forfeit permanently, paid on a share of what it catches. Why this is singular: you must state on the record what somebody owes, under a clock, and an omission is money gone forever rather than a correctable error — one state's law says the association "waives the right to collect any moneys owed in excess of the amounts specified" from anyone relying on it in good faith. Who is there: a managed service selling exactly this job with preparers trained per jurisdiction; the largest management company in North America routes its orders through that marketplace rather than doing it in-house; plus long-established portals, and an AI service on the requesting side running 10,000 agent tasks a day. The core platforms are already AI-branded. 🔴 Where they stop: every vendor sells fulfilment and turnaround. Not one advertises catching the omission, which is where the money is. But the real stop is not competitive, it is structural, and it is why this fails: the price is capped by legislation state after state — $160, $299, $375, $400 — so doing the job faster or better cannot earn more money anywhere. And the party that loses the forfeited money is the association; the party who would buy the fix is the management company. A share-of-what-we-catch arrangement needs the loser and the buyer to be the same person, and here they are not. What would change it: ask three management companies whether an association they manage has ever been unable to collect because it was left off a payoff letter, and who ate it. "The management company gets billed" flips everything above.
The last round killed 31 of 39 candidates at the screen because somebody was already selling something nearby. Reopening them, the same answer came back over and over: the company already there was serving the other side of the same argument, or selling a system rather than doing the work. Here is each one that came back, with the specific gap that was found.
| Candidate | Why it was killed | What reopening it actually found |
|---|---|---|
| Utility damage defence | "Already crowded" | The crowd is on the utility's side. One firm alone invoices 81,000 claims a year against contractors. On the contractor's side: one part-time private investigator licensed in a single state. |
| Car auction arbitration | Vendors exist | Every vendor serves the buyer filing the claim, including a no-win-no-fee service advertising $300 average concessions. Nothing at all for the seller defending. |
| Produce rejection claims | Three organisations already do it | All three are referees — a credit bureau, a joint dispute body, the government. They are structurally forbidden from taking a side, and they only see the escalated cases; the informal ten-minute phone concession is untouched. |
| Milk cheque audit | Three named software vendors | All three sell to the co-operative writing the cheque. Three separate searches for anyone auditing a milk cheque for a farmer returned nothing. |
| Utility rebate filing | Three named competitors | Opening all three: one was a marketing blog post, one a generic landing page, one a general-purpose form filler. The real market is three funded startups covering heat pumps in two states. |
| Made-in-America certification | One named vendor | That vendor's own site names European packaging and substances rules; made-in-America appears only in its blog. (A later pass read it less favourably — see the weaknesses below.) |
| Timber scale settlement | Seven forestry software vendors | All seven stop at the mill's front gate. The leader's own competitor comparison confirms none of them dispute a mill's weight, grade or deductions. |
| Oil and gas part-owner audits | Industry too technical | The technical screen was applied to the operator, who is not the customer. The customer is a family partnership opening a PDF at a kitchen table — and their audit right silently expires every year. |
| Railway flagging charges | Railroads never concede | Chasing the wrong money. The railroads' own forms already promise that unused prepaid deposits "shall be returned" — an entitlement nobody collects, not a dispute. |
| Freight damage claims | Outsourced services exist | They serve the shipper. The carrier-side product is a filing cabinet — records, storage, workflows, no judgement — while the law gives carriers five specific defences nobody helps them build. |
| Cold storage unbilled work | Vendors market the fix | The vendor's own article names the failure as charges "never recorded" and its answer is capture inside its own software going forward. It makes no claim to recover a charge from a paper dock ticket. |
| Hazardous waste manifests | A rule change would kill it | The rule is only proposed, paper would end 24 months after a final rule that has not happened, and the December 2025 electronic reporting duties remain fully enforceable. The medium is changing, not the job. |
| Concrete billing | The dominant vendor ships it | True for producers who can afford a $50,000-a-plant system. And nobody at all audits the concrete bill for the contractor being billed. |
| Damaged goods at distributors | The industry giant owns it | That giant states on its own site that in January 2025 it "completed the full divestiture of all post-purchase and retail returns operations". |
| Aviation parts certificates | A vendor claims to read them | The industry's own integrity coalition lists software to validate those certificate fields as a medium-term action — a thing that does not exist yet. |
| Uniform and linen invoices | Cost-recovery consultants exist | They are people, working monthly or quarterly, for large accounts, on a share of savings for up to five years. The bills arrive weekly. |
And the honest half of the record: some candidates came back CLOSED on re-examination, which means the previous rounds were sometimes not thorough enough rather than too harsh. Two sibling candidates claimed nobody sells backflow filing to testers — a product built for exactly that has existed since 2018 at $29 a month. A salvage-yard candidate cited a consumer marketplace as proof the AI product already ships; opening it shows a site for people parting out a car in a driveway. A snow-services candidate said nothing assembles weather and service records into a defensible record; a vendor's homepage says it does, by name. And import duty refunds turned out to have an AI-native, contingency-priced competitor running our exact business model on our exact job, five months old, with a formidable founding team.
Read this before trusting any single number in the catalogue.
1. This round did no new hunting. The shape analysis is synthesis over what the research already contained. Every fact inherits the grade its original researcher gave it, and several entries flag their own numbers as vendor marketing or unverified.
2. Nine candidates were researched twice, under identical names, by different rounds. They have been merged here, later verdict winning, and five of them disagreed: aircraft-part certificates (worth a look → strong), fuel tank files (weak → worth a look), mine training records (weak → worth a look), towing vessel evidence (worth a look → weak), refrigerant leak files (worth a look → strong). Two more entries were the same job under two different names — utility damage defence — and have been merged into one. The count is 78 after that merge; the underlying research file still holds both copies and should be de-duplicated before anyone counts entries again.
3. Two entries carry an unresolved factual contradiction that would embarrass us in front of a customer. The made-in-America domestic content threshold is recorded as 65% in one place and 55% in another; the 55% comes from the regulator's own page and the 65% belongs to a different regime, and both primary sources blocked the fetch. And the two passes on that same entry read its main competitor completely differently — one found a European-facing product, the other a $4M-funded company marketing on exactly this language. That single question decides whether the entry is quiet or contested.
4. Several searches were never run because the budget ran out, and their absence is not evidence. Nobody swept the big construction software marketplaces for a domestic-content module — the check most likely to weaken entry 9.2. Nobody counted mid-market tobacco distributors, US towing companies, US ship agencies, or the population of pass-through grant recipients. Four entries rest on a population figure that does not exist yet.
5. Numbers. Most entries are honestly NOT SIZED, which is correct at this stage. Where figures do appear, several are vendor marketing about a problem that vendor sells the cure for, and they are labelled as such — the 3–15% cold-storage leakage range, the 20–40% uniform savings claims, the $11–15 billion of unclaimed duty refunds, the 10–15 hours a week of union payroll. Two outright fabrications were caught and discarded during the work: a demurrage error-rate statistic attributed to a guide that does not contain it, and a "56% uncaptured charges" figure absent from the paper credited with it. Others may survive.
6. Two entries brush the out-of-scope line and need a legal answer before anything is built. Grant sub-recipient monitoring (7.3) touches organisations that hold sensitive client records, and dental and optical lab intake (11.4) handles a patient's name alongside clinical instructions. Both must be scoped to money and documents only, and if they cannot be, they die. One more, power pole attachments (1.10), sells to cable and internet companies — the closest anything here comes to the no-technology-industries rule, and that is a call for Nick rather than a researcher.
7. Some strong-looking entries fail the frequency test and it is stated inside them rather than hidden. Chemical inventory filing is annual. Workers' compensation premium audits are annual. Abandoned vehicle titles, lumber grade claims, snow claim defence and payoff letters are all episodic. Frequency is where the money is, and these are the entries most likely to look better on paper than in a year of operating.
8. Several sites refused to be read. Figures for a trade association's membership, a scrap-metal recycling body, a dredging association's member list, an agency software vendor's product page and a healthcare distribution white paper all rest on search summaries rather than an opened page. Each is flagged in its entry; none is load-bearing on its own, but they should be re-opened before anyone quotes them outward.