ReVAT

1. One-liner

Turns a shoebox of foreign invoices into a complete 8th Directive refund claim before the September deadline kills it.

2. Trend signal — why now?

Three things moved in the last twelve months, and they all push the same direction: the paperwork bar for cross-border VAT refunds went up, and it went up hardest for the people with the smallest claims.

The documentation rule changed on 1 January 2026. Germany’s Bundeszentralamt für Steuern now requires invoices and import documents to be uploaded to its online portal whenever the individual document exceeds €250. That is not a “please retain for inspection” rule — it is an upload-at-submission rule, and €250 is a low bar. A single trade-fair stand invoice or a week of Munich hotel nights clears it easily.

The Netherlands did the same and added a hard cutoff. From 1 January 2026, EU claimants must attach invoices where the claim per invoice exceeds €1,000, or €250 for fuel. From 1 April 2026, paper submissions are gone entirely for non-EU claimants and claims without the required attachments are not processed. Not “delayed” — not processed.

The courts are actively litigating whether one missing field kills an entire claim. In case T-407/26, Wagner GmbH & Co. KG SPORT SIGNAGE had its whole Hungarian refund application rejected without examination of its merits — treated as never submitted — because it omitted the VAT identification number for a single invoice. The Budapest High Court referred the question to the CJEU on 5 June 2026, asking whether wholesale rejection over one formal omission is even proportionate. That referral exists because tax administrations really do behave this way.

Now stack that on the European Commission’s own study of the system. Tax administrations processed 669,000 claims in 2016 and approved 629,000 — a 94% approval rate, which sounds fine until you notice what the 6% die of: invalid or insufficient documentation is the third most common rejection ground cited by administrations, and 55% of tax administrations reject a claim automatically if the taxpayer fails to supply requested additional information inside the Article 20 deadline. Of 217 businesses surveyed who process claims in-house, 92% have had to handle an additional-information request, and they described those requirements as particularly burdensome. Language was the single most common problem, at ~21% of responses — five Member States accept applications and additional information only in their national language.

And the Commission recorded the punchline itself: VAT refund agents told the study that translation costs alone can make a claim uneconomic, because agents work on commission and tight margins, so “even small additional costs could make submitting a claim uneconomic for them.”

That is a regulator writing down, in plain text, that the professional market structurally abandons small claims. Vendor marketing puts the annual unclaimed pile around €20 billion — I treat that as an unsourced vendor estimate, not a statistic, and the argument does not need it.

Provenance:

• Signal 1 (demand): EC study — 669,000 claims processed in 2016, 94% approved; invalid/insufficient documentation a top-3 rejection ground; 55% of administrations auto-reject on missed Article 20 information deadlines; 92% of 217 in-house filers have faced additional-information requests; language the #1 reported issue (~21%); agents state small claims go uneconomic — https://taxation-customs.ec.europa.eu/system/files/2019-07/20190620_final_report_vat_refunds.pdf — accessed 2026-08-28
• Signal 2 (regulatory/feasibility trigger): Germany BZSt requires invoice/import document upload for documents over €250 from 1 January 2026 — https://www.bzst.de/EN/Businesses/VAT/VAT_Refund/Businesses_Non_EU_Member_States/businesses_from_non_eu_member_states_node.html — accessed 2026-08-28; Netherlands requires attachments over €1,000 (€250 fuel) and stops processing unattached claims from 1 April 2026 — https://www.pwc.nl/en/insights-and-publications/tax-news/vat/new-dutch-vat-refund-rules-for-non-eu-entrepreneurs.html and https://dutchtaxcalculators.com/vat-refund-netherlands-2026/ — accessed 2026-08-28
• Signal 3 (economic): CJEU referral T-407/26 (Wagner Sport Signage), referred 5 June 2026 — entire refund application rejected as never submitted over a missing VAT ID on one invoice, now testing proportionality — https://www.vatupdate.com/2026/08/25/new-general-court-vat-case-t-435-26-t-407-26-wagner-sport-signage-no-details-known/ — accessed 2026-08-28; commission-based recovery agents (VAT IT, VAT4U, FastVAT, VATcube) all gate pricing behind contact forms and run success-fee models — https://vatit.com/ , https://vat4u.com/en/vat-refund-hotels-travel-expenses-eu/ , https://fastvat.com/claiming-vat-refund-europeanunion/ — accessed 2026-08-28 Category: Regulatory arbitrage

3. The opportunity

The cross-border VAT refund market has a structural hole in it, and the hole is shaped exactly like a small business.

Recovery agents — VAT IT, VAT4U, FastVAT, Global Tax Reclaim, VATcube — all run the same commercial model: success fee, a percentage of what they recover, no public pricing, “request a demo.” I checked. Not one of them publishes a price. That model is rational for them and fatal for the small claimant: if an agent takes 25–30% of a €900 claim, that is €225–270 of revenue against a claim that needs invoice chasing, possible translation, portal submission, and a live response to an additional-information request four months later. The Commission’s own study captured agents saying exactly this — small extra costs make a claim uneconomic for the agent. So they don’t chase it. They aim at corporates with six-figure travel programmes and ERP integrations.

Meanwhile the statutory floor for claiming is astonishingly low. Article 17 of Directive 2008/9/EC sets the minimum at €400 for a period of at least three months but under a year, and €50 for a full-calendar-year claim. The law says a €400 claim is valid. The market says a €400 claim is not worth a professional’s time. Everything between those two lines is unserved — and it’s unserved by design, not by oversight.

What actually blocks the small claimant is not the arithmetic. It’s four mechanical chores that a commission model can’t afford and a busy founder won’t do: (1) finding which of last year’s foreign invoices carry recoverable VAT at all, given eligibility for meals, accommodation and entertainment varies by Member State of Refund; (2) reading each invoice well enough to pull supplier VAT number, invoice number, date, taxable base, VAT amount, and the correct expense sub-code; (3) getting attachments over the new €250/€1,000 upload thresholds into the portal in the right form; and (4) answering an additional-information request, in the refund state’s national language, before the clock runs out — the thing 55% of administrations auto-reject you for missing.

That is a document-extraction, classification, translation and deadline problem. All four are now cheap. That’s the whole thesis.

The incumbent weakness is not bad software — it’s a pricing model that cannot bend. A flat-fee tool doesn’t care whether the claim is €400 or €40,000, so it can serve a band the commission agents must refuse. Memory of this catalog says it plainly: hunt segments incumbents publicly refuse on unit economics. The Commission published the refusal.

4. Target market

• Primary customer: Owner-managers and their bookkeepers at EU-established companies with 3–50 staff that spend abroad inside the EU but are not VAT-registered in the countries they spend in. Concretely: exhibitors and stand-builders who work the German and French trade-fair circuit; specialist consultancies and agencies with recurring client travel; hauliers and coach operators buying foreign diesel; construction and industrial-services subcontractors buying local materials, tool hire and accommodation on cross-border jobs; conference-heavy software and biotech firms. Typical annual recoverable foreign VAT of €800–€15,000 — above the Article 17 floor, below what makes an agent’s phone ring.
• Why they buy: Because the money is theirs, it is provably reclaimable, and the current alternative is a bookkeeper spending an evening they will not enjoy. Businesses surveyed by the Commission take two to five hours to prepare and submit claims, and that’s before an additional-information request lands. The pain has a date attached: 30 September of the year following the refund period. Miss it and the money is simply gone — not deferred, gone.
• Rough TAM reasoning: The system processed roughly 669,000 claims a year at the time of the Commission’s study, filed by the businesses already inside the process. The addressable pool is bigger than that, because it includes the firms that stopped filing or never started — the Commission explicitly notes language problems “can ultimately lead to businesses refraining from exercising their right to a VAT refund.” I don’t need a precise count. If a single-digit percentage of the existing claim population, plus lapsed claimants, will pay €150–500 a year to stop losing claims on formalities, the €1–5M band is reachable. I’d rather be honest that this number is soft than invent one.
• Why now for them: Two of their most-used refund states raised the documentation bar in the same January, one of them stops processing non-compliant claims outright from April, and the CJEU is being asked whether a single missing field can void a whole claim — which tells them the downside is total, not partial.

5. Product sketch (MVP)

• Drop in a year of foreign invoices — PDFs, phone photos of fuel and hotel receipts, or a mailbox forward — and get back a per-country ledger of what is recoverable, what isn’t, and why.
• Country eligibility screen that flags the expense types Member States actually fight over — food and restaurant services, accommodation, entertainment — so you don’t file a claim that gets struck for non-refundable expenses, the most common rejection ground administrations cite.
• Automatic extraction of the fields a claim lives or dies on: supplier VAT identification number, invoice number and date, taxable amount, VAT amount, and expense sub-code — with a hard “this invoice is missing its VAT ID” warning before submission, which is precisely how Wagner Sport Signage lost an entire application.
• Threshold-aware attachment packaging: knows Germany’s €250 rule and the Dutch €1,000/€250-fuel rule, and tells you exactly which invoice images must be uploaded rather than merely retained.
• Claim-value check against the Article 17 floors (€400 sub-annual, €50 annual) so you file the right period and don’t get rejected for being under threshold — a documented rejection driver in Romania.
• A deadline clock for 30 September, and — the part that matters — an additional-information watch: when the refund state comes back with a query, it drafts the response, translates it into the required national language, and pushes you until it’s sent inside the Article 20 window.
• Supplier chase-up letters in the supplier’s language for invoices that are missing a VAT number or were issued incorrectly.
• Year-over-year archive so the next claim starts from last year’s supplier map instead of a blank page.

6. AI angle — what’s load-bearing

Strip the AI out and this collapses into a spreadsheet template nobody fills in. Four jobs are doing real work:

Extraction across chaos. The input is not a clean data feed — it is a German fuel receipt, a French hotel folio, an Italian stand-rental invoice and a phone photo of a taxi slip. Pulling supplier VAT ID, invoice number, taxable base and VAT amount out of that mess, across languages and layouts, is the task modern vision-language models are genuinely good at and rules-based OCR was genuinely bad at. This is why the tool can exist in 2026 at a price point that didn’t work in 2019.

Eligibility classification per Member State. “Is this recoverable?” is not one rule, it’s twenty-seven, and it differs most on exactly the expense types small firms spend on — meals, hotels, entertainment. Classifying a line item into the right expense sub-code for the right refund state is judgment work at a scale a human won’t repeat 200 times.

Translation as a deadline instrument. Five Member States accept applications and additional information only in their national language; two more strongly prefer it. The Commission found language the single most-cited problem and named translation cost as the thing that makes claims uneconomic for agents. Cheap, competent machine translation is the specific input cost that collapsed — it converts the agent’s margin-killer into a rounding error.

Drafting the response to an information request. This is where 55% of administrations auto-reject. Turning “the authority wants proof of business purpose for invoice 14” into a drafted, translated reply with the right attachments, inside the window, is the highest-value thing in the product.

If you removed the AI you would have a deadline reminder. That is not a business.

7. Localization angle

This is a localization play in the truest sense — the moat is per-country procedural detail, not a UI language toggle.

Launch order is Germany and the Netherlands, because those two moved their documentation rules in January 2026 and the pain is freshest, and because Germany is where businesses reported the most trouble: ~21% of claims to Germany hit language/translation problems and ~26% hit difficulties with additional-information requests, per the Commission survey. Germany is simultaneously the biggest source of claims and the biggest source of friction. That’s the wedge.

The national-language requirement is a feature, not an obstacle. Czech Republic, Poland, Romania, Slovenia and Spain accept applications and additional information only in their own languages; Austria and France want theirs with English tolerated where necessary. A product that answers a Spanish authority in Spanish and a Polish one in Polish is doing something the claimant genuinely cannot do at 9pm on a Thursday.

Pricing localizes too: a €39/month tool is a rounding error to a Dutch consultancy and a real decision for a Bulgarian haulier, so the entry tier has to be per-claim, not per-seat.

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

• Pricing: Two doors. Per-claim: €99 per country-claim filed (covers extraction, eligibility screen, attachment packaging, submission-ready output). Subscription: €49/month (€588/yr) for unlimited claims, multi-country, the additional-information watch and the year-over-year archive. Accountants and bookkeepers get a €149/month multi-client tier. Published prices on the website — which by itself distinguishes it from every incumbent I checked.
• Why not commission: Deliberate. Success fee is what forces incumbents to abandon the small claim. Flat fee is the whole strategic point, and it also keeps the product out of regulated payment-handling — the customer’s refund goes from the tax authority to the customer, never through us.
• ACV: Blended ~€600. Subscribers land near €588; per-claim users who file two or three country-claims a year land €200–300; accountant seats land €1,788.
• Rough math to $1M ARR: ~1,550 subscribers at €588, or a realistic blend — 900 direct subscribers (€529K) + 250 accountant seats (€447K) ≈ €976K. Call it 1,150 paying accounts.
• Rough math to $5M ARR: Needs the accountant channel to carry it. ~1,200 accountant/bookkeeper seats at €1,788 (€2.15M) plus ~4,800 direct subscribers (€2.82M). That means being the default tool inside small accounting practices in 3–4 countries, not a direct-to-SMB product. Realistic in year 3–4, not year 2.
• Expansion path: More refund states per customer, then adjacent recovery — domestic input VAT on foreign supplier invoices, then 13th Directive claims for non-EU clients of EU accountants (the Dutch April 2026 digital-only switch creates the same pain one layer out), then per-claim insurance-style guarantees once rejection data accumulates.

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

The deadline does the selling. 30 September is a real date that costs real money, and it recurs annually.

• Trade-fair exhibitor lists, timed to the deadline. Messe Frankfurt, Messe München, Koelnmesse and RAI Amsterdam publish exhibitor directories per show. Foreign exhibitors at a German fair paid German VAT on stand rental, services and hotels almost by definition, and most are not German-VAT-registered. Pull exhibitor lists from the 2025 shows, filter to non-German-established companies of the right size, and send a country-specific note in August–September: “You exhibited at [show] in 2025. Your deadline to reclaim the German VAT on that stand is 30 September. Here’s what it’s probably worth.” That is a named list, a provable spend event, and a countdown — the strongest cold-outreach setup I’ve seen in a while.
• Bookkeepers and small accounting practices, via the thing they hate. Small practices get asked about foreign VAT once a year and dread it because it’s out-of-domain, multilingual and low-fee. Approach through national bookkeeper associations and their forums (NL: NOAB / SRA member firms; DE: Steuerberater networks and DATEV-adjacent communities) with a “we do the foreign one, you keep the client” pitch. One practice converts to a multi-client seat and brings 5–15 claims. This is the channel that reaches $5M; the direct channel alone does not.
• Haulage and coach operators on foreign diesel. Fuel is the cleanest possible case: high VAT, unambiguous eligibility, explicit €250 fuel threshold in the Dutch rules, and operators who already know the refund exists but find the paperwork not worth it. National road-transport associations and their trade press are a concentrated, reachable list.
• Rejection-triggered outreach. The CJEU referral in T-407/26 gives a content hook that the target audience will actually read: “one missing VAT number cost this company its entire claim.” Publish a plain-language breakdown per refund country, aimed at the search terms a panicking bookkeeper types in September.
• Pre-sell before building. Take the 2025 exhibitor lists, offer a free “what’s your claim worth” estimate from photographed invoices, and charge €99 to file for anyone who likes the number. Revenue before the product is finished.

10. Build complexity — justification

Medium. The extraction, classification and translation layers are off-the-shelf model calls — no custom training, no proprietary dataset needed. The genuine work is procedural: encoding per-Member-State eligibility rules, expense sub-codes, thresholds, language requirements and portal formats, and getting the submission path right for each launch country. That’s research and careful implementation, not R&D.

Two honest frictions. First, submission is not a clean API story — claims route through the claimant’s own Member State portal, so v1 realistically produces a submission-ready package and walks the user through filing rather than filing headlessly; full automation comes country by country. Second, correctness matters more than in most SaaS, because a wrong eligibility call costs the customer the claim. That argues for conservative flagging and human confirmation on anything ambiguous.

Two people, 3–4 months to a Germany + Netherlands v1. Add roughly a month per additional refund state.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Preparing and packaging a claimant’s own refund application. No regulated tax-advice licence needed to extract data and format a filing; positioned as software, with the customer submitting. Care needed on how advice is worded per country.
Ethical — no harm / dark patterns	✅	Helps businesses collect money they are statutorily owed. Flat fee, published pricing, no commission skim, no handling of customer funds.
Market exists (evidence above)	✅	669,000 claims/year processed, an entire commission-based agent industry, and a documented segment those agents refuse on economics.
1–5 person team can build this	✅	Two people, 3–4 months for a two-country v1.
Launchable with <$50K / ₹40L	✅	Model inference, hosting, and per-country rule research. Well under $50K.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	15/20	Real money with a hard annual expiry, and total loss on formal defects. Docked because it’s felt once or twice a year, not daily — the Commission’s own figure of 2–5 hours per claim is painful but survivable, which is exactly why people tolerate it.
Demand evidence	15	12/15	Unusually strong for this catalog: a regulator’s own quantitative study, statutory thresholds, two 2026 rule changes, a live CJEU referral, and a whole agent industry charging for it. Docked because the direct evidence that small claimants will pay a flat fee is inferential — I have the agents’ refusal documented, not the SMB’s yes. No verbatim SMB quotes found; I won’t invent them.
Build feasibility	15	11/15	Off-the-shelf models do the hard cognitive work. Docked for per-country procedural encoding and the messy submission path.
Distribution clarity	15	11/15	Exhibitor directories are a genuinely named, timed, provable-spend list. Docked because the accountant channel — the one that reaches $5M — is slower and unproven.
Revenue mechanics	15	12/15	Flat pricing against a commission-priced market, clear ACV, sane customer counts for $1M. Docked because $5M depends on channel partners behaving.
Time to first revenue	10	8/10	Pre-sellable against the 30 September deadline before the product is complete; the estimate-then-file motion produces cash early.
Defensibility	10	7/10	Accumulating per-country rejection and eligibility data compounds — knowing which expense codes actually clear in which state is worth more each year. Rules are copyable; the workflow lock-in and the archive are stickier.
Total	100	76/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

You need someone who can wrangle document extraction pipelines and someone who will happily read Directive 2008/9/EC and twenty-seven national implementations. A VAT practitioner as co-founder or first hire is close to mandatory — this is a domain where being 90% right is a product defect.

Key assumptions to validate (3–5)

1. Assumption: Small claimants (€800–€15,000 annual recoverable VAT) will pay a flat €49/mo or €99/claim rather than do nothing. How to test: Take 200 non-German exhibitors from 2025 German trade-fair directories, offer a free claim-value estimate, and count how many pay €99 to proceed. Need ≥8% of estimate-takers to convert.
2. Assumption: Extraction is accurate enough on real-world receipts that the eligibility output is trustworthy. How to test: Run 300 genuine foreign invoices — including phone photos of fuel and hotel receipts in four languages — and measure field-level accuracy on supplier VAT ID, VAT amount and date. Below ~97% on VAT ID, the product is dangerous.
3. Assumption: Commission agents genuinely decline this segment. How to test: Mystery-shop six agents with a €1,200 claim and record who quotes, who declines, and at what fee.
4. Assumption: Bookkeeping practices will adopt a multi-client seat. How to test: 20 conversations with small NL/DE practices; look for ≥5 willing to pilot at €149/mo.
5. Assumption: A submission-ready package (not headless filing) is acceptable to customers. How to test: Watch 10 users attempt an actual portal submission with the package in hand; measure completion without support.

Risk flags

1. Regulatory risk (two-sided): The CJEU ruling in T-407/26 could go the claimant’s way and require authorities to request missing information rather than reject wholesale. That softens the fear that drives urgency — though it would also legitimise the additional-information-response feature, which is the stickiest part of the product. Watch it closely.
2. Correctness liability: If the tool says “recoverable” and the authority disagrees, the customer loses money and blames you. Mitigate with conservative flagging, explicit confidence levels, and never auto-asserting eligibility on the contested categories (meals, entertainment, accommodation).
3. Incumbent drop-down: Nothing stops VAT IT or VAT4U launching a €49 self-serve tier. They’ve had years and haven’t, because it cannibalises commission revenue and their sales motion is enterprise — but a well-funded entrant with no legacy model is the real threat.
4. Seasonality: Demand concentrates hard around the 30 September deadline. Cash flow and churn will be lumpy; the subscription tier and the additional-information watch (which fires months after filing) are the antidotes.
5. Evidence gap: My demand case rests on regulator and vendor sources, not on small-business voices. I looked for verbatim SMB complaints and did not find them at the quality bar. That is the single biggest reason confidence is Medium and not High.

14. Structured verdict

Score:                  76/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder who can build document-extraction pipelines,
                        paired with an EU VAT practitioner who has actually filed
                        8th Directive claims
Time to revenue:        6–10 weeks (pre-sell against the 30 September deadline)
Capital to launch:      €15–25K
Top 3 assumptions to validate first:
  1. Flat-fee willingness to pay among €800–€15K claimants — 200 exhibitor
     outreach, need ≥8% of estimate-takers to pay €99
  2. Extraction accuracy on real receipts — 300 invoices, ≥97% on supplier VAT ID
  3. Agents genuinely decline small claims — mystery-shop 6 agents with a €1,200 claim
Kill criteria:
  - Abandon if <5% of 200 targeted exhibitors accept a free claim-value estimate
  - Abandon if extraction accuracy on supplier VAT ID stays below 95% after tuning
  - Abandon if two or more major recovery agents launch published sub-€100
    self-serve tiers before v1 ships
  - Abandon if fewer than 3 of 20 bookkeeping practices will pilot the multi-client tier

15. Next step — 1-week validation sprint

• Day 1–2: Pull exhibitor directories from three 2025 German and Dutch trade fairs. Build a list of 200 non-domestic exhibitors, 3–50 staff. Simultaneously mystery-shop six recovery agents with a realistic €1,200 claim and record every quote or decline.
• Day 3–4: Collect 300 real foreign invoices (own network, bookkeeper contacts, purchased sample sets) across four languages including phone photos. Run extraction. Measure field-level accuracy on supplier VAT ID, VAT amount, invoice date. This is the technical go/no-go.
• Day 5: Send the 200 exhibitors a country-specific note offering a free claim-value estimate before the 30 September deadline. Measure reply rate, then convert estimate-takers to a €99 paid filing commitment.

Falsifiable outcome: ≥5% of the 200 exhibitors request an estimate, ≥8% of those pay €99 to proceed, extraction hits ≥97% on supplier VAT ID, and at least 3 of 6 agents decline or price the €1,200 claim unattractively. Miss the extraction bar or the payment bar and this is a no-go regardless of how good the regulatory story reads.