Recovery IntelligenceRecovery firms keep 20 to 30% of what they recover [1]. Our agents build the proof for each claim and a specialist signs it. Twenty of the 198 largest US companies make this a $12M a year business. Pre-revenue: it starts with a paid blind test.
Companies pay invoices twice, pay above contract prices and miss rebates. A supplier repays only when someone ties the payment to the contract in force that day. Our hypothesis is that this proof limits how many claims get pursued. The blind test checks it.
Agents match the contract to invoices and payments and assemble a claim with every number linked to its source. We audit a company's supplier payments and are paid only from what comes back.
Language models now read contracts and match invoices. Most of the remaining cost is specialist review time.
We serve the largest US companies, where one whole audit earns us $750K a year. $12M a year takes whole audits on $160B of supplier payments a year: 16 companies paying all year. The plan signs 20 by month 36, because each new client's cash arrives 5 months after its claims.
US companies with $20B or more of revenue on the 2025 Fortune 500 [2]. If half of revenue goes to suppliers (our assumption), each pays suppliers $10B or more a year.
Our estimate for the whole Fortune 500: half of $19.9T revenue paid to suppliers [3], claims of 0.04% to 0.1% of spend [4] [1], 75% collected as in our plan, 20 to 30% fees [1]. Outside estimates are lower: $260M to $1.28B [5].
Our fees at plan inputs. Our ceiling, not a forecast.
From 20 clients, 10% of the 198.
Proof the category pays: in healthcare payment recovery, Performant reported $123M of revenue in 2024 [6] and sold for about $670M in 2025 [7], and HMS reported $673M of revenue in 2020 [8].
One client in the plan: a whole audit at 25% contingency, the rate buyers already pay.
Audited spend, claims found and the collection rate are our assumptions; change them under Financials. Claims found (0.04% of spend) is the input with the least evidence. Texas expects 0.04% from a whole audit [4]. The fee is the commercial rate buyers already pay [1].
Year 3 revenue (year 2: $3.75M)
Revenue in month 36 alone
Every month profitable from here (first profitable month: 17)
Peak cash needed
Cash break-even. High case: month 20. Low case: not by month 36.
A share of every dollar recovered, paid when the cash posts. The commercial recovery audit default.
Year 1 revenue
Year 2 revenue
Swipe the table sideways for more columns.
| Structure | Year 3 revenue | Peak cash needed | Cash break-even | First profitable month | Clients by month 36 | Year 1 | Year 2 |
|---|---|---|---|---|---|---|---|
| $9.75M | $1.95M | Month 27 | Month 17 | 20 | $62.5K | $3.75M | |
| $5.46M | $2.45M | Not by month 36 | Month 23 | 20 | $35K | $2.10M | |
| $3.75M | $5.99M | Not by month 36 | Not by month 36 | 30 | $12.5K | $1.40M | |
| $5.85M | $3.04M | Not by month 36 | Month 26 | 30 | $25K | $2.25M | |
| $5.23M | $1.62M | Month 36 | Month 20 | 20 | $373K | $2.67M | |
| $7.62M | $1.72M | Month 29 | Month 17 | 20 | $240K | $3.30M | |
| $2.25M | $2.77M | Not by month 36 | Month 34 | 15 | $180K | $1.17M |
Click a structure to select it. Structures that win clients faster than 1x reach the client ceiling sooner. Raise the ceiling to see how much volume a discount or a partner channel would have to deliver.
Drag any slider; every number and chart on the page follows. All of these are our assumptions until customers and the pilot replace them.
We go straight to the largest companies, not to recovery firms. A firm would pay us a slice of its fee for a second look; a company pays the full fee for the whole audit.
The plan finds claims at the 0.04% of spend Texas expects from a whole audit [4], on $10B of supplier payments per client. Two blind tests measure it before we sell: a batch a recovery firm has already worked shows what we add, and a fresh batch shows what a whole audit finds.
Recovery is sold to them on contingency, 20 to 30% of what comes back [1], so no budget line is needed. Ritchie leads sales and hires two account executives in year 1 (our plan). Each client costs $40K to win, our assumption; a contested whole audit could cost far more.
Founders, an engineer, the account executives, an audit lead and reviewers. Review time per claim sits in the per-claim cost.
The same plan (two clients a quarter, $120K a month team) at a slower sales pace, different finding rates and company sizes, and the business if we stayed a second look inside recovery firms.
| Case | Run rate at month 36 | Year 3 revenue | Cash break-even | Peak cash |
|---|---|---|---|---|
| Plan: Texas rate on $10B of spend | $12.7M | $9.75M | Month 27 | $1.95M |
| One new client a quarter | $6.8M | $5.13M | Month 33 | $1.90M |
| Texas rate on $5B of spend | $6.4M | $4.88M | Not by month 36 | $2.43M |
| High-case rate (0.06%) on $10B | $19.1M | $14.63M | Month 23 | $1.80M |
| Second look only: 0.03% on $5B, one client a quarter, $60K team | $2.5M | $2.02M | Not by month 36 | $1.31M |
What decides it: claims found per dollar audited, and how large the companies are. Staying a second look inside recovery firms, with smaller buyers, one client a quarter and half the team, earns $2.02M in year 3: $281K a client a year instead of $750K. That is why we sell the whole audit.
| Who | What they report | What they are to us |
|---|---|---|
| PRGX | Accounts payable recovery audits; says it returns $2B a year to clients [9] | Main rival for whole audits |
| apexanalytix | Supplier data and recovery audits; $8.4B a year prevented and recovered [10] | Rival, or a partner for supplier data |
| Cotiviti, Performant | Healthcare payment integrity; Cotiviti reports over $10B of claim errors fixed in 2025 [11] | Out of scope at first |
| Prevention software | Tools that flag duplicate payments before they go out | Different job |
Co-founder and CEO of Now City, a real estate team that built its own AI underwriting tools. Owns product development, go-to-market strategy and marketing.
Co-founder and COO of Now City. Owns sales and account management with the largest companies.
Sets the acceptance rules and signs claims. We recruit before the blind tests; none runs without one.
Not yet recruitedThe engineer builds secure record access, the claim pipeline and the review queue. The account executives sell whole audits with Ritchie.
Not yet recruitedWe are not raising yet. After the blind tests we will size a raise from the measured results. The plan needs about $1.95M before revenue covers costs; the low case needs $3.54M by month 36 and has not broken even.
A batch of past supplier payments for a blind test. We return the claims we can prove and what each is worth, and you decide on a whole audit.
A closed batch you have already worked, for a paid blind test. We measure the extra claims your reviewers accept and the minutes each takes.
Open any section. The research memo, team roles and system costs are in the top navigation.
Recovery audit firms review years of past payments for duplicates, price errors, missed credits and unearned discounts. Texas plans on recovering 0.04% of vendor payments [4]; a vendor benchmark is about 0.1% [1]. Finding an anomaly is cheap. Proving it to a supplier is not, and our hypothesis is that the proof is the bottleneck agents can remove. Recovery firms come first only as blind-test partners; the revenue is in whole audits for large companies.
Most recovery work is paid on contingency: nothing upfront, a percentage of what is recovered. Here is what buyers report paying.
Public buyers pay roughly half the commercial rate. Our plan uses 25% for direct sales; public work would be nearer 14%.
A recovery audit paired with prevention software sold separately [18].
A flat license; the client finds and recovers its own errors [18].
Government pays from documented savings, nothing if there are none [19].
Starting terms for the model. Each one changes under Financials, in Explore the model.
Our profit per client a year, after review and service costs, before fixed costs. The retainer leads because part of its revenue arrives monthly instead of waiting for collection; its terms are our assumption, and customer question 2 tests whether buyers would pay one. We lead with contingency because it is what buyers already pay.
clients cover the team's fixed costs at 25% contingency
clients won by month 36 in the plan, two a quarter
US companies with $20B or more of revenue [2]
a year if every one were a client, at plan inputs: our ceiling, not a forecast
If agents prepare claims in minutes, we can clear a backlog faster than a manual process. The tempting move is to pass that on as a fee five or ten times lower. A lower rate cuts what every recovered dollar earns by the same factor.
Profit per client a year at each rate, all other inputs equal. Red bars lose money on every client.
Offer a low-priced blind test on a closed batch. Prove the lift once, then sell at market rates.
Claims filed months earlier mean cash months earlier. In the model, that is a shorter cash lag at the same rate.
If small claims sit below an incumbent's working threshold, a lower rate there is new money. Discovery must confirm the tail exists.
The plan assumes a first paying client in month 7 and two more each quarter after that. These milestones are the proof that assumption needs.
Ten conversations with controllers at large companies and with recovery firms; a recovery audit lead signed
Proof to move onTwo batches with data rights: one a firm has worked, one fresh; a pricing structure buyers will pay
Claims on both batches, judged by the batch owner's reviewers
Proof to move onClaims found per dollar audited near 0.04% on the fresh batch, and extra accepted claims on the worked one
Four clients by month 12, in one claim class
Proof to move onCash posted; measured claims and collection replace the model's guesses
Two new clients a quarter on signed terms; twelve by month 24
Proof to move onA known cost to win a client and a 12-month cash plan; raise sized from the results
Twenty clients; a second claim class or the public-sector track
Proof to move onMargin after review holds at volume
Each answer moves an input in the model. Ask them in this order, and record the numbers, not the impressions.