5 August 2026 · internal · the adversarial pass
Six independent researchers, each on a different industry. A skeptic behind each one whose job was to kill their findings and personally open every cited source. Then this.
It went against us. The theory does not hold in the broad form it was written, and most of the numbers in the earlier report are now dead — including several that turned out to be fabricated by vendors and laundered into apparent consensus by search engines.
What survived is narrower and better. Restoration is still standing, for a real structural reason, but with almost nothing publishable behind it today. And a legal finding surfaced that changes the pricing model in the lead vertical — see section 6.
The earlier report is still up at work done, never billed with a warning banner on it. Nothing here is a decision.
Mostly no — not in the form it was written. The theory says money is lost in the gap between doing work and documenting it, so the work never becomes an invoice and is therefore invisible from inside the company. Six lanes of research were pointed at that sentence and almost every solid number that came back measures something else: work that was recorded and then discounted by a partner, work that was invoiced and then refused by a customer, or work that was billed and paid and then clawed back because the paperwork didn't support it. The one place the theory's exact mechanism was found in audited black and white is heavy civil construction — a $5.5bn contractor carrying $402m of work performed with no agreed price — and that is a megaproject claims business, not the mid-sized service companies HyperProfits is aiming at. Among the three genuine specialty trade contractors we checked, the same balance ranges from "immaterial" to 2.1% of revenue. Meanwhile 94.5% of trucking fleets do bill for the disputed work; law firms do record the time and then choose to write it down; and healthcare's national documentation gap runs overwhelmingly in the opposite direction — money the payer wants back, not money the provider forgot to ask for. The theory survives as a narrow proposition: in industries where the price is genuinely settled after the work is done, and where a third party controls whether you get paid, there is real money moving on documentation quality. That is restoration and insurance-claim contracting, and it is heavy construction claims. It is not a general law of service businesses, and the version of the theory that says "the loss is invisible from inside the company" is contradicted by our best source in the lane that supports the theory most: Tutor Perini's $402m is disclosed, audited, discussed with Deloitte and flagged as a critical audit matter. They know exactly where it is.
These are the only things allowed in front of a prospect, and each one has to travel with its caveat.
1. A US government auditor states plainly that contingency recovery does not work below a certain client size, and names the reason. Seven US states told the federal government they could not run a contingency-fee recovery programme because the money to be found was too small to pay a contractor. Several put the work out to tender and got no bids. One cancelled its procurement outright. The reason is in footnote 17: "A RAC incurs significant implementation costs, including systems integrations, data access and manipulation, provider outreach and education, and appeals support costs. States with small Medicaid populations, or some claim types — especially low-dollar, high-volume claims — generally do not lend themselves to contingency fee-based audits." MEASURED. US Government Accountability Office, GAO-23-106025, June 2023. https://www.gao.gov/assets/gao-23-106025.pdf, pages 10 and 13. This is evidence against us, and it is the strongest single fact in the whole programme. It is listed first for that reason.
2. Florida's insurance regulator counted every closed residential property claim in the state and found homeowners come back for more money on roughly half of water and wind claims. 43.7% of accidental-water claims, 48.5% of other water, 51.0% of wind and hail, out of 732,340 claims closed by 180 insurers. MEASURED. Florida Office of Insurance Regulation, Residential Property Claims and Litigation Report, January 2024, covering calendar year 2022. https://floir.gov/docs-sf/default-source/property-and-casualty/other-property-casualty-reports/january-2024-pclr.pdf Three caveats that must be said out loud every time: a reopening is a REQUEST for more money and the report never says how often the request succeeded or for how much; it is one state in the Hurricane Ian year, before Florida's insurance reforms, and the regulator says so itself; and it has never been published again — the word "reopen" appears zero times in the 2025 and 2026 editions. This is a photograph, not a live number.
3. The US government measures the gap between doing work and documenting it at 6.55% of Medicare fee-for-service spending — $28.83bn — and says explicitly it is not fraud. Down from 7.66% and $31.70bn the year before. About 53% of it is insufficient documentation and a further 12% is no documentation. MEASURED. CMS Fiscal Year 2025 Improper Payments Fact Sheet, published 15 January 2026. https://www.cms.gov/newsroom/fact-sheets/fiscal-year-2025-improper-payments-fact-sheet The caveat is fatal if you get it wrong: this is mostly money providers have to give BACK, not money they failed to bill. Use it to prove documentation gaps are real and enormous. Never use it to prove anyone is owed money. If a buyer catches you blurring those, you lose the room.
4. A $5.5bn contractor's audited accounts show $402 million of completed work with no agreed price. $402.1m at 31 Dec 2025, $393.8m the year before, plus $324.7m of claims, against $5.54bn of revenue. The filing says these amounts "are billable upon the agreement and resolution between the contractual parties." MEASURED. Tutor Perini 2025 annual filing. https://www.sec.gov/Archives/edgar/data/77543/000007754326000028/tpc-20251231.htm Two caveats. It is a balance, not an annual loss — it barely moved year on year, so it revolves rather than accumulates. And Tutor Perini is heavy civil megaprojects, the most dispute-heavy corner of construction. The three real specialty trade contractors we pulled carry between "immaterial" and 2.1%: Comfort Systems ($9.1bn revenue) says immaterial three years running, EMCOR 0.15% (0.84% counting disputed receivables), Limbach 2.10%.
5. Thomson Reuters, on the reported financials of 184 US law firms, says the loss happens inside the firm before the bill goes out. "Clients can do little to impact a bill between the time the work is done and when that work is billed... law firms continue to do more to negatively impact realization and collections on their own than what clients are doing to them." The gap between billed and collected — the client-pushback part — has "hovered around 2% since the start of 2022." Overall collection against the negotiated rate sits at 90.3%. MEASURED. Thomson Reuters Institute, Law Firm Rates in 2024 (September 2024) and 2026 State of the US Legal Market (January 2026). https://www.thomsonreuters.com/en-us/posts/wp-content/uploads/sites/20/2024/09/Law-Firm-Rates-Report-2024.pdf The caveat that removes this from our theory entirely: the rate is agreed first, the time IS recorded, and a partner then chooses to discount it. Not one dollar of undocumented work is inside that 9.7% gap. This is a pricing-discipline problem, not a documentation problem.
6. The industry's own published recovery yield is about one tenth of one percent of spend. Two competitors publish the same figure independently: apexanalytix says "typical recovery rates around 0.1 percent of total spend" with fees of 20-30%, and its worked example assumes $5 billion of annual spend. PRGX says roughly $1m recovered per $1bn of spend. MEASURED (two independent vendor benchmarks agreeing). apexanalytix, Antoine Pindrys, 10 April 2025. https://www.apexanalytix.com/resources/blog/recovery-audit-cost/ Do the arithmetic on a mid-sized company: 0.1% of $8m of spend is $8,000 found, and a 25% cut is $2,000. Less than a week of one person's time. This is the cost side of the ledger, not the revenue side HyperProfits is chasing — but it is why the incumbent industry never came down-market.
7. Texas ran a real contingency recovery audit across twelve state agencies and the auditor earned about $17,000 for two years' work. $181,696 identified, $121,784 recovered, 14% fee. The contract had two renewal options; only one was exercised and it ended in 2014. MEASURED. Texas Comptroller Recovery Audit Report 2014. https://fmx.cpa.texas.gov/fm/audit/recovery/rpt2014/Recovery_Audit_2014.pdf Caveats that soften it: the payments were three to four years cold, the agencies were already good at catching duplicates, and some funds legally couldn't pay a consultant. It is a cautionary data point, not a fair test. It is also the only publicly measured contingency engagement at anything like this client size, and both sides walked away.
8. The market leader in recovery audit moved UP-market over fifteen years, and tells shareholders the pool shrinks as clients improve. 2004: "principally large and mid-sized businesses." 2019: "typically Fortune 1000 companies... with multi-billion dollars of purchase transactions." And, in the growth-strategy section rather than boilerplate risk: "Over time, our clients tend to resolve recurring transaction processing deficiencies." MEASURED. PRGX Global annual filings. https://www.sec.gov/Archives/edgar/data/1007330/000100733020000003/a201910kprgx.htm The industry did not overlook the mid-market. It served it and left.
9. There is a hard contractual deadline that can extinguish the right to be paid for extra work. The standard American construction contract requires written notice of a claim for extra money "within twenty-one (21) days after occurrence of the event giving rise to such Claim." MEASURED, and it is a legal fact rather than a statistic. AIA A201-2017, Section 15.1.3.1. https://www.wisconsin.edu/procurement/download/construction_contracts/A201-2017---220617.pdf Three precisions. It applies to disputed claims, not signed change orders. The brutal "waiver of any and all claims" language sits in an owner's addition to this particular contract, not in the standard form. And it governs owner-to-general-contractor; subcontractors usually face SHORTER deadlines, so quoting 21 days to a subcontractor may understate their urgency.
10. The restoration industry's own body says getting paid is its members' number one ranked problem. 72% cite challenges getting paid and payment timeliness; 67% adjusters dictating charges; 43% carrier denial of overhead and profit. MEASURED, but weakly. Restoration Industry Association Advocacy and Government Affairs issues survey. http://www.restorationindustry.org/restoration-blog/exploring-findings-ria-aga-issues-survey-us-canada This is an open-link, self-selected internet poll run by a lobbying function, with no disclosed sample size, measuring feelings rather than dollars. Use it only as evidence that restorers BELIEVE this is their biggest problem. It cannot support any claim about how much money is lost.
11. Restoration companies wait an average of 60 days to get paid, and receivables are 47% of the average firm's assets. Under $1M revenue: 46 days. $1-10M: 54-57 days. $10-15M: 91 days. $50M+: 80 days. Receivables median 16.3% of annual gross revenue. MEASURED. RIA 2025 Cost of Doing Business Report, FY2024 data. https://2815486.fs1.hubspotusercontent-na1.net/hubfs/2815486/2025%20RIA%20Cost%20of%20Doing%20Business%20Report.pdf This is SLOW payment, not UNDER-payment. A real adjacent pain, and a different product. Do not blur them.
12. Trucking: 94.5% of fleets already bill for waiting time, and are paid on fewer than half of those invoices. MEASURED-ish (survey by an interested party). American Transportation Research Institute, 10 September 2024, 2023 data. https://truckingresearch.org/2024/09/new-research-documents-substantial-financial-and-safety-impacts-from-truck-driver-detention/ Included because it argues against us. The premise that companies fail to bill is false for the overwhelming majority of this market.
This happened repeatedly, across four different lanes, and it is the single most important operational finding of the whole exercise.
"Average net profit before taxes is 3.8% for restoration companies." DEAD, and this was the foundation of the entire commercial argument in the lead vertical. The article Nick was shown hyperlinks to a specific document — the 2024 RIA Cost of Doing Business Report. We downloaded that exact 80-page file and searched it. "3.8" appears twice, both times inside "$30M-$50M." "38%" appears zero times. We then downloaded the next edition in case it came from there: zero again. Worse than absent, it is contradicted — the RIA's real published distribution shows 23% of restoration companies earning over 20% net income, which is arithmetically impossible in a population averaging 3.8%. The report also states its own method: "We report medians, not averages (mean)" — so the RIA does not even publish the kind of figure being quoted.
"One dollar recovered is worth $26.32 of new sales." DEAD. It is nothing but 1 divided by 0.038. On the RIA's real numbers the multiplier is roughly 10x to 12.5x, less than half. It also assumes a recovered dollar is pure profit, which is only true when the work was already done and paid for out of pocket.
"More than one-third of restoration companies are breaking even or losing money." DEAD. The phrase "breaking even" appears zero times in either report. The real figure is 7% reporting negative net income, described as "flat at 7%."
"Companies that consistently document overhead-and-profit entitlement collect 22-28% more per job." DEAD for the third time. It exists on exactly one page on the internet — a marketing agency that sells SEO and websites to restoration companies — credited to a "2024 Xactware industry analysis" it does not link. We downloaded Verisk's actual Overhead and Profit white paper, the document the claim is pinned to, and extracted all five pages: the only numbers in the entire document are page numbers, a phone number, a street address, the year 1986 and the 2023 copyright. It is a definitional explainer telling software users where to type a percentage. It could not contain this statistic even in principle.
Three more fake statistics from the same marketing agency, all falsely credited to the Restoration Industry Association. "Additional damage on 65-75% of projects," "O&P disputed in ~30% of supplements," "experienced contractors recover 15-25% more per claim" — all attributed parenthetically to "RIA, 2024," none of them in any RIA document we read end to end. This is more dangerous than an obviously fake number, because the body is real and a busy prospect assumes the citation is too.
"18% of gross revenue lost to poor supplement processes" and "initial carrier offers fell from 87% to 59% of net invoice." Both dead, and here is the finding: they come from the same single page — an offshore virtual-assistant vendor selling $8/hour labour to write supplements. Two independently suspicious numbers collapsing into one uncited vendor page means restoration's statistics ecosystem has a small number of fabrication sources that search engines launder into apparent consensus.
"The average law firm loses 15-30% of its collectible billable hours." DEAD. We opened the page: a bare assertion in the summary box, no study, no dataset, no citation, published by a company that sells AI billing automation to law firms. The page cites real sources for its other claims and names nothing for this one. That is the tell.
"Bill fast or lose it — 10% lost after a day, 25% after two, 50-70% after a week." DEAD. Attributed to work "compiled" for a bar association blog, with no study title, no year, no sample. The vendor sells legal billing software. And the range was inflated in the retelling — the page says 10/25/50, not 50-70.
The accounts-receivable collectability curve (95-97% at 30 days falling to under 50% at 180). DEAD. No study, no author, no date. And its own published numbers rise twice as debt ages — 30-40% at 91-120 days becoming 40-50% at 121-180 days. A curve where old debt becomes more collectable than new debt is not a measurement of anything.
Circulating construction figures: "30% of project revenue lost to unbilled change orders," "8-12% of completed work unbilled," "10-20% routinely." No primary source exists for any of them. The 30% was handed over by a search-engine summary as though it came from a named research report, where it does not appear.
"Veterinary practices lose 5-10% of revenue, 17% of lab tests unbilled." DEAD for client-facing use. Every result is a secondary citation, mostly practice-management software vendors quoting each other. The article carrying it gives no citation for either figure, and its own companion claim that this costs practices "almost $1.1 million a year" is not credible for a clinic grossing one to two million.
"Dental audits find gaps on 10-15% of encounter days." DEAD. The only source is a dental billing company's marketing page with undated underlying data.
The healthcare 1% and 3-5% leakage figures. Both appear on the trade association's own website. Neither is a study. The 1% is asserted by a co-author who was VP of Strategy at the company that sold the client the fix; the 3-5% carries a footnote marker that points to a government payment-systems document saying nothing about leakage. The sector's most-quoted numbers are folklore, and a prospect's finance chief can dismantle them in one search.
The Novant Health $7.5m recovery, described as "the strongest evidence that the money is genuinely findable." It is a vendor case study printed in a trade magazine — co-authored by the software company's VP of Strategy, and marketed by that company's successor as its own customer result. And the wording is "identified," never "collected." When you divide it by Novant's own revenue you get 0.14% — roughly one tenth of the range the industry markets.
"Underbilling is three times more common than overbilling." Called the single best evidence in the whole programme. It is not a pooled result across 18 studies; it is one included study of 100 notes written by internal medicine residents — doctors in training who do not own the practice and are not paid on collections. The same study says the effect reverses with experience: underbilling falls as residents advance while overbilling rises. The "5,754 records" are papers screened in a literature search, not patient encounters.
Litigated insurance claims paid 2-5x more than non-litigated ones. The arithmetic was right and the source document disproves the interpretation on a page nobody read. Hold claim duration constant and the gap reverses on the core restoration perils: for claims open over a year, water damage pays $51,658 litigated against $57,479 non-litigated. On four of eight perils, including both water perils, long-running non-litigated claims pay MORE.
"Nobody credible publishes anything on billed-versus-collected for restoration." Overstated — a negative claim made without opening the one document that would hold the answer. The RIA reports are not paywalled; both are publicly downloadable. And both surveys explicitly ask members "Enter the total percentage of bad debt or write-offs." The number has been collected twice and simply is not printed. That is a publication gap, not a measurement gap. It is a phone call, not a research project.
"22 of 35 states couldn't find a contractor." The table entry is right, the framing isn't. States could give multiple reasons; the top reason was something else (25 states cited existing programme-integrity work); and the body text breaks the 22 down — only seven tied it to insufficient revenue. Seven is still a good number. Twenty-two is not defensible.
"There is no fat margin at enterprise scale." Refuted by the same filing it came from. The loss year contains a $10m one-off write-down, the two prior years are profitable, and the recovery-audit segments themselves earned $37.6m of EBIT on $164m of revenue — about 23%. The consolidated loss comes from an unprofitable side business and head-office overhead.
"Contingency fees are banned on tax work." Wrong on current law. A federal court permanently enjoined enforcement of that rule for ordinary refund claims in 2014, the IRS did not appeal, and the Treasury's own 2024 proposed rule concedes the point and proposes removing the section entirely. Any tax lawyer in the room would correct this.
"Contingency recovery breaks down below very large client size." Too broad, and disproved by two facts inside the same research. Pennsylvania licenses and caps unclaimed-property finders who work on contingency for individual people. Florida and Texas license public adjusters who work on a percentage for a single household. Both operate profitably three orders of magnitude below Fortune 1000. What has a size floor is accounts-payable recovery audit specifically, because its yield is a tenth of a percent of spend.
Ranked on evidence alone, for the theory as written — work done, never documented, never invoiced.
1. Heavy civil and large commercial construction. Only place the mechanism appears in audited numbers at scale. $402m of unpriced completed work at one company, plus a 21-day contractual cliff that a prospect can find in his own contract drawer. Weakness: these are megaproject builders, not the target customer, the balance revolves rather than accumulates, and the money is fully visible to management.
2. Restoration and insurance-claim contracting. Second, not first. It has one genuinely excellent measurement — roughly half of Florida water and wind claims reopened at the homeowner's request — and it has the clearest structural reason for the theory to be true: the work is done before the price is agreed, and a third party decides what you get. It also has the only reachable missing number in the whole programme: the RIA has collected restoration's write-off percentage twice and not published it. Weakness, and it is severe: everything else circulating in this industry is fabricated or vendor-manufactured, so there is almost nothing to build a pitch on today, and the two strongest-sounding figures both trace to a single uncited page.
3. Healthcare. The largest measured documentation gap anywhere — 6.55% of Medicare spending, $28.83bn — from a neutral government source. But it points the wrong way: mostly money the payer wants back. And the most damaging fact in the entire programme sits here. Recovery auditors ARE paid to find underpayments, the statute requires them to report on it, and they still find fourteen times more overbilling than underbilling: $352.5m against $25.9m. In the most heavily coded, most heavily audited billing environment that exists, systematic underbilling is small.
4. Professional services — law, accounting, agencies. The best-measured lane in the programme, and the least relevant. Realization is computed as money collected divided by recorded hours times the agreed rate, so by construction not one dollar of undocumented work is inside the gap. The rate is agreed first, the time is recorded, and a partner voluntarily discounts it. It is a real problem worth about a tenth of agreed value — and it is a pricing-discipline business, not a documentation business.
5. Trucking and freight. The evidence actively refutes the premise. 94.5% of fleets already bill for detention. The failure is that shippers refuse — over missing signatures and disputed arrival times — and that evidence cannot be recreated after the fact. You cannot go back and obtain a signature a shipper's representative declined to give in 2024. There is no historical pot to audit. What the evidence supports is a forward-looking proof-capture product: a software subscription with a slow ramp, not a finder's fee.
6. Field service — HVAC, plumbing, electrical. No measured evidence exists at all. The most-circulated figures were written by a software CEO about his own product category, and his headline recovery number is his own arithmetic from an assumed 30 minutes per technician per day. Do not build a pitch here on numbers, because there are none.
Restoration is not top on evidence quality. It is second, and it is the only vertical where the theory's mechanism is structurally sound and the missing evidence is one phone call away rather than one research programme away. That combination is a reasonable argument for keeping it as the lead vertical. Building the pitch on any restoration statistic currently in circulation is not.
This is the section that matters. Nine reasons, hardest first.
One. The incumbent industry served the mid-market and left it. PRGX told shareholders in 2004 that it served "large and mid-sized businesses." By 2019 it told them it serves "typically Fortune 1000 companies with multi-billion dollars of purchase transactions." That is not an oversight to be exploited. It is a retreat by the people who know this business best, over fifteen years, in a document where being wrong is expensive. Any plan has to say specifically what changed.
Two. The arithmetic of getting inside a client does not shrink when the client does. The government auditor names the cost drivers: systems integration, getting access to and cleaning the data, educating the client's people, and supporting the disputes. Those are roughly fixed per client. Seven states looked at their own numbers and concluded the pool would not fund a contractor. Several tendered and got no bids at all. One cancelled outright. That is a market clearing at zero for exactly the reason HyperProfits has to overcome.
Three. The revenue decays, by design, and the incumbent says so. "Over time, our clients tend to resolve recurring transaction processing deficiencies." Year one finds the most. Every year after finds less, because a competent client fixes the leak the moment you show them where it is. That makes this a treadmill of new logos rather than a compounding book — a completely different investment case, and one to price in now rather than discover in year two.
Four. When the behaviour is the problem, the behaviour reverts. In early 2020 law firms improved realization and collections sharply under existential pressure. Thomson Reuters: "the improvements proved fleeting; and when the pandemic-induced urgency abated, firms resumed their previous billing and collection practices." That is a documented natural experiment in precisely the intervention being sold, and it regressed. So either the fee is taken fast on a one-off basis, or the engagement never ends. A sophisticated buyer will raise this, and the answer has to exist before the first pitch.
Five. The percentage collapses the moment the buyer can run a process. Commercial rate 20-40%. Texas, running a competitive tender, got 14%. Medicare, tendering by region, got 9.0-12.5%. Federal law caps state Medicaid recovery fees at 12.5%. Pennsylvania caps unclaimed-property finders at 15%. Every mature version of this market is a lowest-bidder market. Model the percentage falling, not holding.
Six. In the most measurable environment on earth, the money mostly runs the other way. Recovery auditors are contractually paid to find underpayments and are statutorily required to report on it. They found $352.5m of overpayments and $25.9m of underpayments. Fourteen to one. The comfortable explanation — nobody is paid to look — is closed off by the report itself. The uncomfortable reading is that systematic underbilling is genuinely small relative to overbilling wherever anyone has actually measured it.
Seven. In one of the target verticals the money cannot be recovered retrospectively at all. Trucking's failure mode is a shipper's representative who declined to sign the arrival time. You cannot go back and get that signature. There is no historical pot. The same logic threatens restoration: if the loss is caused by photographs not taken and moisture readings not logged at the time, then it is not findable later either — and a business that can only prevent future losses is a software subscription, not a finder's fee.
Eight. Attribution is a fight on every invoice. A contingency fee requires proving the money would not have arrived anyway. On a supplement that would have been submitted regardless, on a change order the client was going to negotiate, on a lawyer's hour that would have been billed at the same rate — the counterfactual is genuinely contestable, and the client has every incentive to contest it after the money lands. The Novant example is instructive: the number was "identified," never "collected." Fee on identified, and the client feels robbed. Fee on collected, and the cash arrives twelve months late.
Nine. The obvious fix is already installed and it did not work. 85% of law firms have electronic time and billing software available, 95% at larger firms — and realization has sat around 90% for fifteen years anyway. Selling visibility to someone who already owns three dashboards is a much harder sale than it looks on a slide.
Partly, and less than the plan probably assumes.
What AI genuinely changes: reading unstructured material at near-zero marginal cost. The reason recovery audit needed billions in spend was that a human had to look at every transaction to find one worth a tenth of a percent. Reading photographs, job notes, moisture logs, technician texts and estimate revisions is now cheap. That is real and it is new.
What AI does not change, and this is the part to be honest about. Of the four cost drivers the government auditor names — systems integration, data access, client education and dispute support — AI meaningfully attacks one. Getting a $20m restoration firm to give you access to its job management system, its accounting system and its photo archive is a negotiation, not a computation. Educating the owner and his project managers to change what they capture in the field is human work with a human failure rate. And supporting a disputed supplement against a carrier's adjuster is a relationship and an argument, not an analysis.
Then the deeper problem. Healthcare's revenue-cycle industry exists because 98% of medical claims move in one federally mandated electronic format across 3.1 billion transactions a year. Uniformity is what made the machinery worth building. Restoration has no equivalent — no mandated format, no shared schema, every firm's job files shaped differently. So AI helps least on integration exactly where standardisation is worst, which is precisely the market being targeted. The cheap-reading advantage is real; the expensive-plumbing problem is untouched, and the plumbing is what the government auditor said kills these deals.
There is one honest counterweight, and it should be stated because it cuts our way. The "size floor" argument is narrower than it first appears. It is specific to accounts-payable recovery audit, whose yield is a tenth of a percent of spend. Contingency recovery demonstrably works at household scale — public adjusters take a percentage on a single home, and Pennsylvania regulates finders because they operate profitably on individual claims. So the question is not "does contingency work small," it is "is the pool per client, times our cut, bigger than the cost of getting inside that client." That is a testable question with a number, and nobody has produced the number yet.
The hinge is how you are paid, not what you do.
In Florida, a contractor may not act as an unlicensed public adjuster — but the statute contains an express carve-out: the contractor "may discuss or explain a bid for construction or repair of covered property with the residential property owner or the insurer of such property, if the contractor is doing so for the usual and customary fees applicable to the work to be performed as stated in the contract." So talking to the insurer about the estimate is lawful when you are paid your ordinary fee for the work. The moment payment becomes a share of the insurance proceeds, that protection is gone. The restoration trade association tells its own members the same thing: "Try to avoid contracts that base your compensation on a percentage of insurance proceeds. Courts may view this as an indicator of public adjusting." A statute and the industry body pointing at the same hinge.
Texas is harder and is the state to design against. "A contractor may not act as a public adjuster or advertise to adjust claims for any property for which the contractor is providing or may provide contracting services, regardless of whether the contractor: (1) holds a license under this chapter; or (2) is authorized to act on behalf of the insured under a power of attorney." Getting licensed does not fix it. In Florida a contractor can hold both licences; in Texas holding the licence is explicitly no defence. The two states are not parallel and must not be described as such. One open item: this text came from a statute-republishing site because the official Texas Legislature server returned only a navigation shell. Someone should pull the official text before this goes near a lawyer.
Drop Oregon from the sentence. The only thing found is a bare licensing requirement using the word "person," and the exemptions include people providing estimates or reports without compensation and people performing repair services. Whether it bites a contractor is unresolved. Saying "Florida, Texas and Oregon" overstates what we know.
Contingency fees on tax refund claims are NOT banned. A federal court permanently enjoined the IRS from enforcing that rule for ordinary refund claims, the IRS did not appeal, and Treasury's own 2024 proposal concedes the point and would delete the section. This lane is open, not closed.
Where this model has matured, the cut is capped and the finder is licensed. 12.5% federal cap on state Medicaid recovery. 9.0-12.5% for Medicare contractors. 15% cap plus mandatory registration for Pennsylvania unclaimed-property finders. Florida caps public adjuster fees. Expect the same shape to arrive if this works.
The shape this forces on the business. Anything touching a property insurance claim in Florida or Texas cannot be paid as a percentage of the insurance proceeds. Three viable structures: a flat or subscription fee for the operating layer, paid regardless of what it finds; an hourly or fixed engagement fee at customary rates; or a genuine partnership with a licensed public adjuster who takes the percentage on the claim side while HyperProfits is paid for the system. A fourth option is to take the percentage only on money that is not insurance proceeds — unbilled labour, missed materials, un-invoiced change orders on non-insurance work — which is a much smaller pool. This is not a footnote. It rules out the most natural pricing model in the lead vertical, in the two biggest restoration states in the country.
Cheapest first. These cost a phone call or a few hundred dollars.
1. Call the Restoration Industry Association and ask for the write-off number. They have asked their members "Enter the total percentage of bad debt or write-offs" in two consecutive annual surveys and published neither answer. This is the single highest-value action available and it is a phone call. If the answer is 2%, the restoration thesis is in trouble. If it is 12%, everything changes. 2. Register for and download the actual ATRI detention report. It is behind a form, not a paywall. It resolves a direct contradiction: ATRI's press release says fewer than 50% of detention invoices are paid, the trade press says 55%, and the "75% invoiced" figure that the whole trucking argument rested on cannot be checked against either. 3. Buy the Rosenberg 2025 accounting survey to replace a four-year-old realization benchmark, and to source or drop the underpricing claim that was attributed to it and is not on the page. 4. Buy the 2005 Ibbs paper that the "late changes double the damage" claim actually comes from. It is currently an unverified reference inside a verified paper. 5. Get Clio's Legal Trends Report properly — the primary source returns an access error to everyone, and there is an unreconciled arithmetic problem between its headline "$910 per $1,000" and its own separately-reported realization and collection rates. 6. Ask the American Dental Association's Health Policy Institute whether they hold anything on charge capture. Dental currently rests entirely on one billing company's marketing page. 7. Pull the Texas public adjuster statute from the official legislature server and have a Florida and a Texas insurance lawyer read both statutes against three specific proposed fee structures. Two or three hours of legal time answers a question that shapes the whole pricing model.
Then the things desk research can never answer. These need a real company's real books.
8. The pool per client. Nobody knows it. Not one source in six lanes measures how much money a $5M-$50M service business fails to invoice in a year. Every number we have measures a different thing: discounts on recorded time, refused invoices, clawbacks, or work-in-progress balances at megaproject builders. The only way to find out is to take a single restoration firm and reconcile, job by job for twelve months, what was scoped against what was performed against what was invoiced against what was collected. If the answer is under 2% of revenue, the contingency model cannot fund itself at this client size and the whole thing should be repriced as software. 9. Whether the loss is actually invisible from inside. The theory says reports built from invoices cannot see an unbilled job. Tutor Perini disclosed theirs to the penny. The test is simple and it needs a real owner: show him the reconciliation and watch whether he is surprised. If he already knew, the pitch is not "we found money you couldn't see," it is "we will do the chasing you never get round to" — a completely different sale. 10. Whether the found money stays found. Run the same reconciliation twelve months after the intervention. The law firm evidence says the behaviour reverts. If it reverts in restoration too, the business is a permanent engagement, not a finder's fee. 11. Whether an owner will sign a contingency agreement, and at what percentage. Desk research cannot answer this and it determines the entire revenue model. Ten conversations with owners in the $5M-$50M band would. 12. What it actually costs to get inside one company. Integration, data access, training, dispute support — the four things the government auditor named as the reason this fails at small scale. Measure them on the first client before signing the second. That single number decides whether the business exists.
One last thing worth saying plainly, because it applies to how this research gets used rather than to any single number. Two of the six lanes had errors introduced by our own researchers, not by the sources — a sample size misstated by a factor of six, a risk-warning chart reframed as an opportunity, a claim graded MEASURED that was eight people's opinion in an interview. Those were caught only because a second pass went looking for them. The failure mode this whole exercise is designed to prevent — being wrong in Nick's favour — showed up inside the process as well as outside it.