ClaimBack

1. One-liner

Finds every tariff surcharge you paid a supplier and builds the demand letter that gets it credited back.

2. Trend signal — why now?

The IEEPA tariffs got struck down and the money is flowing back — to the wrong people.

CBP had processed $20.6 billion in certified refunds with interest as of 22 May 2026, against roughly $85 billion in accepted claims. By 31 July 2026 that number had crossed $100 billion in refunds processed. Over 40 S&P 500 companies have booked $9.6 billion in tariff refunds, with about $2.1 billion already in hand.

Here is the structural problem. The refund goes to exactly one party: the importer of record — “typically the manufacturer, brand owner, master distributor, or customs broker.” Everyone downstream who actually absorbed the cost gets nothing automatically. Distribution Strategy Group put it plainly: distributors, retailers and end customers who paid higher prices as tariff costs moved through the supply chain “have no direct claim to the refunds,” and whether they recover “depends on contract language and commercial leverage rather than government policy.”

And a very large number of those downstream buyers were billed the tariff as a separately stated line item — which is the single fact that converts a closed AP transaction into a live receivable. Crestron Electronics applied a 5% tariff surcharge to the gross hardware total on all US invoices from 1 May 2025, later cut to 4% by 29 May 2026. Ubiquiti is facing a class action over its tariff surcharge. Washington State’s Department of Revenue has published formal guidance on how tariff surcharges are treated. This wasn’t a fringe practice — it was the standard billing convention across electronics, specialty chemicals, polymers, building products and equipment distribution for eighteen months.

The legal community has already turned. Carter Ledyard’s read: consumer suits against importers like Nintendo and Campbell’s face “significant challenges,” but direct commercial purchasers of importers and distributors represent the emerging litigation frontier, particularly in specialty chemicals, electronics and polymers. The theories are unjust enrichment, money had and received, breach of contract and implied covenant, and restitution. Trade attorneys are warning sellers that keeping a government refund while keeping the tariff-driven price increase invites double-recovery allegations — “particularly when customers can document separate tariff surcharges.”

That last clause is the whole business. When customers can document. Most of them can’t, because the evidence is 18 months of PDF invoices scattered across an AP inbox and a QuickBooks file.

The demand side is already audible. From Marketplace’s reporting: an HVAC equipment supplier explained that manufacturers and importers passed on tariff charges but his company can’t apply to the government because “it has to be the importer that has to apply for the credit,” so asking suppliers is all he can do — and when he asked a major manufacturer about filing, the answer was “We made an executive decision, we met, and the answer is no, we’re not going to file for it.” Restaurant supply CEO Spiro Pappadopoulos was more resigned: the supply chains are so long that asking suppliers for tariff money back likely wouldn’t work — “They’re probably just going to say, ‘Hey, we paid tariffs, too.’”

Both men are describing the same failure: they are negotiating without evidence. Pappadopoulos assumes he’ll lose the argument because he has no idea what he actually paid in surcharges. He is probably wrong about that, and he’ll never find out.

Provenance:

• Signal 1 (Demand): Downstream B2B buyers who paid itemised tariff surcharges have no direct refund claim and must pursue suppliers on contract and unjust-enrichment theories; direct commercial purchasers are the “emerging litigation frontier” — https://www.clm.com/tariff-recovery-litigation-the-focus-shifts-to-downstream-claims-and-b2b-customers/ — August 2026
• Signal 2 (Feasibility): “A closed accounts-payable transaction can become the evidence for a new receivable” — recovery turns entirely on separately itemised surcharges, duty clauses and supplier communications buried in historical invoices — https://www.pymnts.com/news/b2b-payments/2026/old-b2b-invoices-could-decide-100-billion-in-tariff-refunds/ — 2026
• Signal 3 (Economic): $20.6B in certified CBP refunds with interest by 22 May 2026 against ~$85B in accepted claims; distributors and customers “have no direct claim” and recovery depends on “contract language and commercial leverage” — https://distributionstrategy.com/2026/06/tariff-refunds-spark-new-pricing-fight-across-distribution-channels/ — June 2026
• Signal 4 (Corroborating customer voice): HVAC supplier and restaurant-supply CEO on Marketplace describing the inability to pursue suppliers, including a manufacturer refusing to file — https://www.marketplace.org/story/2026/05/07/business-owners-must-decide-what-to-do-about-tariff-refunds-after-supreme-court-ruling — 7 May 2026 Category: Workflow automation (a one-time, high-value evidence-assembly job across 18 months of unstructured AP records) + Underserved niche (contingency recovery-audit firms serve enterprise; the $5M–$150M-revenue buyer has no self-serve option)

3. The opportunity

Two products already exist around this money. Neither is this one.

The first is the importer-side refund tool. Calculators, ES-003 analyzers, liquidation trackers — everything aimed at the party who legally owns the claim. That party is a small population and it is now well served. My own catalog has an entry there.

The second is the AP recovery audit. apexanalytix and the Big-4-adjacent recovery firms have run this play for thirty years: point analysts at a client’s payables, find duplicate payments and missed discounts, take 25–40% of what you recover. Recovery audits are explicitly “evolving to include enhanced customs and trade compliance expertise” in response to tariff volatility. That’s real. It’s also priced and structured for a client spending nine figures on procurement. Nobody at apexanalytix is running an engagement to recover $180,000 for a 40-person electrical contractor.

So the gap is a specific, currently-unserved buyer: the company that paid the surcharge but didn’t import anything. They are too small for a contingency recovery firm, too unsophisticated to run the analysis internally, and completely unrepresented by the importer-side tooling because they are on the wrong side of the transaction.

The 10× is evidence assembly. The reason Pappadopoulos won’t ask his suppliers is not that he lacks a claim — it’s that “asking” without a number is a conversation he loses in fifteen seconds. What changes the dynamic is walking into the call with: across 23 May 2025 to 12 Feb 2026 we paid you $214,880 in separately-stated tariff surcharge line items across 341 invoices; here they are itemised; your published surcharge policy described these as pass-through duties; the duty is no longer payable; we’re requesting a credit. That is not a negotiation, it’s a reconciliation. And the supplier’s own lawyers have already told them that refusing a documented surcharge claim while pocketing a CBP refund is the fact pattern that generates a double-recovery suit.

The load-bearing work is finding the line items. A tariff surcharge appears in the wild as “TARIFF SURCHARGE,” “IEEPA DUTY,” “Sec 232 Adj,” “TRF ADJ 15%,” “Duty Recovery Fee,” “Import Cost Adjustment,” a footnote in the terms block, or an unlabelled percentage of the subtotal. It sits inside PDFs from four hundred different vendor billing systems. There is no field for it in QuickBooks or NetSuite. It was never coded as anything but part of the invoice total. A human reviewing 18 months of AP finds it in about three weeks of tedium. A vision-and-extraction pipeline finds it in an afternoon for the price of lunch.

The second thing incumbents don’t do: the claim doesn’t stop at one supplier. A mid-market manufacturer buys from 30–80 vendors. Every one of them needs its own tally, its own contract-clause read, its own letter. That’s a portfolio job, and portfolios are what software is for.

4. Target market

Primary customer: The controller, CFO or owner of a US company with $5M–$150M revenue that buys physical goods domestically but does not import. Concretely: electrical and mechanical contractors buying through wholesale distribution, equipment dealers, industrial and machine-shop manufacturers buying components, food-service and restaurant-supply operators, MSPs and AV integrators buying hardware, injection molders and fabricators buying resin and metal. 20–400 employees. They run QuickBooks Online, NetSuite, Sage Intacct or Xero. They have one to four people in AP. They have no trade counsel and no customs broker, because they never touch customs.

Why they buy: Because there is a five- or six-figure number sitting in their 2025–26 payables that they can plausibly recover, and they currently have no way to even estimate it. In their words — the HVAC supplier’s version: “it has to be the importer that has to apply for the credit,” so asking suppliers “is really all they can do.” And Pappadopoulos’s version of why he won’t bother: “They’re probably just going to say, ‘Hey, we paid tariffs, too.’” The product’s job is to replace that shrug with an itemised number.

The secondary buyer, and possibly the better one commercially, is the distributor sitting in the middle — who needs the mirror-image analysis. They must answer inbound customer demands with defensible numbers and simultaneously chase their own upstream importers. Distributors are already “managing the challenge that refunds flow only to importers of record, while they must respond to customers who paid tariff-related surcharges.” Same engine, both directions.

Rough TAM reasoning: I’m not going to pretend to size this from the $100B headline, because most of that sits with large importers. The relevant denominator is: US firms in goods-buying industries with $5M–$150M revenue that bought through distribution during the surcharge window. US Census data puts the count of employer firms in construction specialty trades, durable-goods wholesale, and small-to-mid manufacturing in the hundreds of thousands; the subset with material surcharge exposure and enough sophistication to act is realistically 20,000–60,000 companies. At a $2,500–$9,000 engagement value that’s a $50M–$500M addressable pool, most of which will never be captured. I need 400 of them. That is a rounding error on the denominator, which is the correct shape for a bootstrapped play.

Why now for them: Three clocks. First, suppliers are receiving CBP refunds right now — a supplier who has already banked and spent the refund is a much harder target than one processing it this quarter. Second, statutes of limitation on contract and unjust-enrichment claims run from the payment date, and the earliest surcharges were billed in Q2 2025. Third, and most practically: this window closes socially. Once the industry norm settles into “nobody gave credits back,” asking becomes awkward and late. The next nine months is when this is a normal commercial conversation.

5. Product sketch (MVP)

• Connect your books. Read-only OAuth into QuickBooks Online, Xero, NetSuite or Sage Intacct. Pull every AP bill and attached document from Feb 2025 forward. Fallback: drop a folder of PDFs, or forward a mailbox.
• Find the surcharges. Read every invoice and identify tariff-attributable line items regardless of what the vendor called them — separate lines, percentage adjustments, terms-block footnotes, blended increases flagged in a cover email.
• The exposure tally. Per supplier: total surcharge paid, invoice count, date range, the percentage rates applied and when they changed, and a confidence grade on each item. This report is the thing the customer cannot currently produce at any price they’d pay.
• Claim strength grading. Rank each supplier relationship by how recoverable it looks — separately itemised beats blended; an explicit pass-through clause or a surcharge-announcement email beats silence; a supplier who is a known importer of record beats one who isn’t.
• Evidence pack per supplier. A single PDF: the itemised schedule, copies of the source invoices, and any supplier communication describing the charge as a pass-through duty.
• Demand letter drafter. A commercially-toned credit request — not a lawsuit threat — with the number, the schedule attached, and the ask (credit memo, ACH, or applied to open balance). Escalation templates behind it.
• Response tracker. Who was sent what, who replied, who agreed to what, what’s been credited. Because this is 40 conversations, not one.
• The mirror mode. For distributors: run the same analysis on your outbound invoices, so you know your total downstream exposure before your customers do.

6. AI angle — what’s load-bearing

Remove the AI and this is a data-entry firm.

The core task is extracting one semantically-defined concept — a charge that exists because of tariffs — from tens of thousands of visually inconsistent PDF invoices generated by hundreds of unrelated billing systems, where the concept has no standard label, no standard position, and sometimes no label at all. This is not template OCR. A rule that greps for “tariff” misses “IEEPA DUTY,” “232 ADJ,” “Import Cost Recovery,” and the very common case of a 15% line item labelled only with a SKU-like code that the vendor’s cover letter explained once, in April 2025.

Two further pieces of real model work. Contract and correspondence reading: scanning supplier agreements, T&Cs and email threads for the language that determines claim strength — “duties actually incurred,” “surcharge applies while the duty remains payable,” “pass-through,” any promise of reconciliation or refund-sharing. Those phrases are the difference between a strong claim and a weak one, and they’re buried in prose. Blended-increase detection: where a supplier didn’t itemise, inferring the tariff component from a price-series break that correlates with the tariff schedule — flagged as inference, never as fact.

The economics only work because of the price collapse in extraction. Running 30,000 invoice pages through a model costs low tens of dollars. A bookkeeper doing the same job at $35/hour costs more than the engagement fee. That inversion is roughly 18 months old.

7. Localization angle (if any)

N/A — this is a US play, tightly. The entire opportunity is the specific fact pattern of IEEPA tariffs being invalidated by the US Supreme Court and refunded by CBP to importers of record. There is no analogue elsewhere, and the moment you leave US jurisdiction the legal theories and the refund mechanics both evaporate. The only geographic extension worth noting is Canadian and Mexican buyers who purchased from US distributors carrying US surcharges — a real but small tail.

8. Business model — path to $1M–$5M ARR

This is deliberately not priced as pure SaaS, because the primary job is one-time. Pretending otherwise is how you get 90% churn and call it a subscription business.

• Pricing — Scan: $499 flat for the exposure tally. Connect books, get the number, per supplier, with confidence grades. Deliberately cheap; it’s the qualifier.
• Pricing — Recovery Pack: $2,500–$7,500 based on identified exposure — evidence packs, letters and tracking for every supplier. Tiered: under $50K exposure $2,500; $50K–$250K $4,500; over $250K $7,500.
• Pricing — Distributor mirror: $12,000–$30,000 annually. A distributor with 800 downstream customers needs the outbound analysis, a defensible position paper, and a way to process inbound claims consistently. That’s a real operational system, and it recurs while the wave lasts.
• No contingency. Tempting, and wrong for a solo operator: collection risk, no control over whether the customer actually sends the letter, and it drags you toward being a law firm. Sell the evidence, not the outcome.

ACV: ~$3,800 blended across the buyer side; ~$18,000 on the distributor side.

Math to $1M: 220 Recovery Packs at $4,000 average = $880K, plus 8 distributor accounts at $18K = $144K. $1.02M. Reachable inside 12 months if the outbound engine works.

Math to $5M: needs roughly 900 buyer engagements plus 60 distributor accounts, which means either a small sales team or a channel — most likely accounting firms and industry associations reselling. Honestly, $5M is a stretch for this specific wave; $1.5–2.5M is the realistic ceiling on tariffs alone.

Expansion path — and this is the part that decides whether it’s a business or a project. The engine is find money hidden inside unstructured AP documents. Tariff surcharges are the wedge because they’re topical and enormous. The same connected books and the same extraction pipeline then run: fuel surcharges billed after the index dropped, freight accessorials that were never validated against contract, price increases applied before their effective date, rebate thresholds crossed but never claimed, duplicate payments. That converts a $499 one-time scan into a $199–$499/month AP leak monitor for a business that has now seen you produce a real number. If that conversion lands at 25%+, this is a durable company. If it doesn’t, it’s a very profitable 18-month campaign — which, at these margins, I’d still take.

9. Go-to-market wedge — first 100 customers

• Publish the surcharge registry, then mine the reactions. Build a free public index of every US distributor and manufacturer known to have applied an itemised tariff surcharge, with the rate and the dates — starting from the documented ones (Crestron’s 5% from 1 May 2025 cut to 4%, Ubiquiti’s, and whatever else is findable in published price-increase letters and terms pages). Every buyer of those companies is a qualified lead, and the page is a permanent inbound magnet with zero competition for the keyword. Announce it in the trade press that’s already covering the fight.
• Go through the trade associations, not around them. NECA, PHCC, MCAA, PMMI, NAW’s member distributors, state restaurant associations. Every one of these has a members’ newsletter and a 2026 agenda dominated by tariff whiplash. Offer a free “what did tariffs actually cost your members” webinar with a live scan of a volunteer member’s books. Associations will run this because it’s a member benefit that costs them nothing. Ten associations × one webinar × 200 attendees × 3% = 60 scans.
• The class-action adjacency. The Ubiquiti tariff-surcharge class action and the Crestron investigation have plaintiff firms actively soliciting affected purchasers. Those firms need documented per-purchaser damages and don’t want to build the tooling. Partner rather than compete: they refer commercial purchasers whose claims are too small or too individual for the class, and I hand them clean schedules.
• Cold outbound with the number pre-computed — partially. For any target whose supplier appears in the registry, the email writes itself: “You buy from [Distributor]. They billed a documented 5% tariff surcharge from May 2025. On $2M of annual purchasing that’s roughly $100K you may be able to claim back. We’ll tell you the exact figure for $499.” Specific, checkable, and unlike most cold email it contains a fact the recipient can verify in ten seconds. Target 3,000 sends, expect 2–4% to buy the scan.
• Accountants and fractional CFOs as the channel. They already have read access to the books, they’re already being asked about tariff refunds by clients, and they have no answer. A $150 referral fee plus co-branded reports gets a fractional CFO with 12 clients to run 12 scans in a week.

10. Build complexity — justification

Low. Off-the-shelf accounting OAuth integrations (QBO, Xero, NetSuite, Intacct all have documented APIs and existing SDKs), document extraction on commodity models, PDF generation, a straightforward tracker UI. No novel infrastructure, no real-time anything, no ML training.

The genuinely hard part is not engineering — it’s the extraction taxonomy: assembling and continuously curating the list of how tariff charges are actually labelled in the wild, and the contract-language patterns that grade claim strength. That’s domain work done by looking at a few thousand real invoices, and it improves with every customer. Two people, 8–10 weeks to a v1 that can process a real customer’s books end to end. First revenue can precede the integrations entirely — the drop-a-folder-of-PDFs path is a two-week build and is how the first twenty customers should be served.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Assembling a customer’s own invoice data and drafting a commercial credit request is not legal practice. Must stay firmly on the evidence side of the line and never render an opinion on the merits of a claim.
Ethical — no harm / dark patterns	✅	Helping buyers recover money they demonstrably paid for a duty that no longer exists. The inference mode for blended increases must be labelled as inference, never presented as documented fact.
Market exists (evidence above)	✅	$100B+ refunded, active downstream litigation, named surcharge programs, sourced customer voice.
1–5 person team can build this	✅	Two people, 8–10 weeks.
Launchable with <$50K / ₹40L	✅	Realistically $8–15K: API credits, integration developer accounts, the registry site.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Five to six figures of recoverable cash per customer, with a closing window. Not a nice-to-have. Docked because it’s episodic rather than daily — the pain is acute but has a specific season.
Demand evidence	15	13/15	Multiple independent hard signals: $100B in processed refunds, active downstream litigation frontier, named surcharge programs with documented rates, sourced quotes from two affected buyers. Docked only because nobody has yet proven buyers will pay a fee rather than shrug.
Build feasibility	15	12/15	Standard stack, well-documented integrations, commodity extraction. The taxonomy curation is real work but it’s grind, not risk.
Distribution clarity	15	12/15	The registry play plus association webinars plus a cold email containing a verifiable fact is a genuinely strong opening. Docked because association calendars move slowly and I’m assuming a cold-outbound conversion rate I haven’t tested.
Revenue mechanics	15	12/15	Pricing is benchmarked against what recovery audits and trade counsel cost, and the value is self-evident relative to a five-figure recovery. Docked because the one-time shape means $1M is a treadmill until the AP-monitor conversion is proven.
Time to first revenue	10	8/10	The manual version — take a folder of PDFs, do the analysis, deliver a PDF — is sellable in week three. Genuine pre-sales are realistic before any code ships.
Defensibility	10	3/10	Almost none. The extraction taxonomy and the supplier registry are a modest accumulating advantage; the workflow is copyable in a quarter. This is an execution-and-speed play against a closing window, and I’m scoring it honestly.
Total	100	77/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · sales-heavy

Technical to build the extraction pipeline and the integrations. Sales-heavy because the buyer is a controller who has never bought software like this, the purchase is episodic, and the association and accountant channels are relationship work. A pure builder will stall at 30 customers.

Key assumptions to validate

1. Assumption: Separately-itemised tariff surcharges are common enough in mid-market AP to make the average scan worth running — I’m assuming a material fraction of invoices in the target industries carry them. How to test: Get read access to 10 real companies’ 2025–26 AP files across five industries. Count. If the median company’s identifiable surcharge exposure is under $15,000, the fee can’t be justified and the idea dies.
2. Assumption: Buyers will pay for evidence rather than only for outcome — i.e. they’ll take a $4,500 fixed fee over a contingency. How to test: Offer both structures to the first 20 qualified prospects and see which they choose. If more than 70% insist on contingency, the business model is wrong even if the product isn’t.
3. Assumption: Suppliers actually pay some of these claims. The product sells evidence, not outcomes, but if the realised recovery rate is near zero, word travels and sales stop by month five. How to test: Track outcomes obsessively on the first 30 engagements. Need at least 30% of demands to produce some credit.
4. Assumption: The AP-leak-monitor upsell converts. How to test: Offer it to the first 50 completed engagements at $299/mo. Need 25%+.

Risk flags

1. Market timing — the window is the whole thesis. This opportunity has a natural half-life. Refunds are being disbursed now; social permission to ask expires; limitation periods run. Twelve months late is worthless. That cuts both ways — it deters slower competitors, but it means a slow build kills the business outright.
2. Legal boundary risk. Drafting demand letters for a fee, in volume, sits uncomfortably close to unauthorised practice of law in some states. Mitigation is strict: the product produces a commercial credit request and an evidence schedule, never legal advice or an assessment of claim merits, with a plaintiff-firm referral path for anything contentious. This needs an actual lawyer’s sign-off before launch, not a founder’s judgement.
3. Relationship friction. Some buyers won’t pursue a supplier they depend on, at any evidence quality. Pappadopoulos’s resignation is partly economic and partly relational. Expect a meaningful chunk of scans that never convert to a recovery pack because the customer decides not to pick the fight.
4. Appeal risk. Elements of the tariff picture remain in litigation — Section 122’s rejection was under appeal at the time of writing, and Section 232 tariffs were never invalidated. Misclassifying a still-valid duty as recoverable produces a customer who sends a demand letter that gets destroyed. The taxonomy has to distinguish authorities, not just spot charges.
5. Single-event dependency. If the AP-monitor expansion doesn’t convert, this is a campaign with an end date, not a company. Underwrite it as such.

14. Structured verdict

Score:                  77/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical operator comfortable with document-extraction pipelines,
                        paired with someone who can work trade associations and accounting
                        channels. Prior AP / finance-ops exposure is a real advantage.
Time to revenue:        3–6 weeks (manual delivery precedes the product)
Capital to launch:      $8,000–$15,000
Top 3 assumptions to validate first:
  1. Median identifiable surcharge exposure exceeds $15K per target company —
     test by analysing 10 real companies' 2025–26 AP files by hand
  2. Buyers accept a fixed fee over contingency — offer both to 20 prospects, count
  3. At least 30% of demand letters produce some credit — track the first 30 engagements
Kill criteria:
  - Abandon if median identifiable surcharge exposure across 10 hand-analysed AP files
    is under $15,000
  - Abandon if fewer than 5 of the first 40 paid scans convert to a Recovery Pack
  - Abandon if under 20% of the first 30 demand letters produce any credit within 90 days
  - Abandon if a well-capitalised AP-recovery incumbent launches a self-serve
    sub-$1,000 tariff scan before month 4

15. Next step — 1-week validation sprint

• Day 1–2: Get five real companies to share their 2025–26 AP invoice files — a contractor, a machine shop, an AV integrator, a restaurant-supply operator, an equipment dealer. Offer the analysis free in exchange for the data and a candid conversation. Manually count separately-stated tariff surcharges in each.
• Day 3: Build the surcharge registry v0 from public sources — every distributor and manufacturer price-increase letter, terms page and surcharge announcement I can find. This is a day of searching, and it’s the distribution asset.
• Day 4: Call 20 controllers in the target industries, cold. Lead with a specific number from the registry relevant to their likely suppliers. Ask two questions: would you pay $499 to find out exactly what you paid, and would you actually send the letter?
• Day 5: Decide.

The falsifiable outcome: across the five hand-analysed AP files, the median identifiable surcharge exposure must exceed $15,000, and at least 6 of 20 cold-called controllers must say they’d pay $499 for the scan and would send the demand. Miss either bar and this is a thesis about money that exists but nobody will chase — which is a PASS, not a GO.