New startup ideas · Money rails for agentic commerce · Disputes, refunds and reconciliation for agent purchases
startup concept
Verdikt
Issuer-side triage for disputes on agent-initiated transactions
Verdikt gives issuing banks and card programs a decision layer for disputes where the transaction came from an AI agent: it parses Trusted Agent Protocol identities, consent records and token scopes to determine whether the agent acted within its mandate, then recommends accept, represent or write off.
- Software subscription
- Enterprise
- Visa and Mastercard agent rails
7
similar startups, last 2 years (35 all-time)
no
no public money matching the concept's terms
Test it before you build it
$1,000 · 5 weeks · 20 prospects
For about $1,000 and 5 weeks, this proves whether fintech card programs will pay $5,000 now for agent-dispute triage instead of waiting for Visa and Mastercard to ship network rules.
Riskiest assumption · A dispute lead at a US fintech card program with live agent-transaction volume will pay for mandate adjudication this quarter rather than wait for Visa and Mastercard to ship network-level agent dispute rules.
1Focus group: who and where
Head of disputes or risk operations at a US fintech card program or BaaS sponsor bank (10-500 employees) whose chargeback queue now contains agent-initiated transactions carrying Visa Trusted Agent Protocol or Mastercard Agent Pay consent records that nobody on the team has tooling to read.
where to find 20 · The This Week in Fintech Slack community, where card-program risk leads are active; Fintech Business Weekly's running coverage of BaaS sponsor banks (Evolve, Cross River, Lead, Sutton, Celtic and peers) used as a call list; warm introductions through Lithic and Marqeta ecosystem contacts and fintech seed investors; Money20/20 USA in Las Vegas in late October if the pipeline needs topping up.
2Sell first, build later
A 30-day design-partner pilot starting October 5, 2026: Verdikt scores the program's last 90 days of disputed agent-initiated transactions against their recorded TAP or Agent Pay consent scopes and delivers drafted accept, represent or write-off decisions in the program's own dispute format, plus a written adjudication rulebook the team keeps.
the ask · $5,000 per pilot, invoiced 100% up front, three cohort slots; converts to a $40,000 per year subscription after the pilot.
a real yes · Money moved on the pilot invoice, or a sponsor-bank LOI signed by an authorized officer with a dated start and a named data scope. Compliments, unpaid data-sharing offers and 'send the deck to procurement' do not count.
3Small experiments
The first one attacks the riskiest assumption; each ends with a number that says whether to run the next.
1. Agent-dispute queue census
$300 · 12 days
Book 20-minute calls with 15 dispute leads sourced through warm intros and the This Week in Fintech Slack. Ask three counts: agent-initiated disputes in the last 90 days, how each was adjudicated, and whether they can afford to wait for network-level rules. One founder runs all calls off a shared script and logs answers in a sheet.
keep going if · 8 of 15 report agent-initiated disputes in their queue and at least 5 of 15 say they must decide these cases now, network rules or not
2. Consent-record memo inbound test
$100 · 7 days
Write a 2-page memo, 'What the TAP consent record does and does not settle in a dispute,' with worked examples from the published Trusted Agent Protocol spec. Post it in the This Week in Fintech Slack and send it to Fintech Business Weekly and two other fintech newsletter authors. Count qualified dispute leads who book a call from it.
keep going if · 4 dispute or risk leads at card programs book calls from the memo within 7 days
3. Paid pilot teardown pitch
$600 · 18 days
Build a 6-page sample teardown showing five disputed agent transactions scored against their consent scopes with accept, represent or write-off recommendations and rationale. Pitch a $5,000 30-day design-partner pilot (their last 90 days of agent disputes, scored and drafted in their own dispute format) to every program from experiments 1 and 2 that showed volume.
keep going if · 2 of 10 programs pitched sign the pilot order and pay the $5,000 invoice, or 1 pays and 2 sponsor banks sign LOIs with dated starts
4Collect a deposit up front
Tesla took $1,000 refundable reservations for the Model 3 and $100 for the Cybertruck before building either: the deposit is the measurement, not the revenue.
$5,000
per prospect, refundable
how · A one-page pilot order form signed by the head of disputes or their VP with the $5,000 invoice paid at signature; fintech card programs buy sub-$10k tooling on a director's signature without a procurement cycle. Sponsor banks that cannot pay before vendor risk review instead sign an LOI with a dated start and named pilot users. set up: Stripe Invoicing ↗
what it reserves · One of three October pilot slots, the $5,000 design-partner price locked against the $40,000 annual list, and their dispute format built into the first version of the rulebook.
refund · Refunded in full if Verdikt does not deliver scored decisions on the agreed dispute file within 30 days of receiving the data.
target · 2 paid pilots, or 1 paid pilot plus 2 signed sponsor-bank LOIs, from 20 conversations within 35 days
before taking money · Card programs and sponsor banks are regulated and their dispute files contain cardholder data, so run the pilot only on redacted or tokenized exports and expect an information security review before any production data moves.
Go: build it if
8 of 15 census calls confirm agent disputes already in queue, and 2 programs pay $5,000 (or 1 pays and 2 sponsor banks sign dated LOIs) within 5 weeks.
Kill: stop if
Fewer than 5 of 15 programs have any agent-initiated disputes in their queues yet, or 0 paid pilots and 0 signed LOIs after 20 completed conversations - that is the 'we will wait for the networks' answer in numbers.
5 Scripts to run itoutreach message, landing copy, deposit terms · click to open
outreach message
You run disputes at a card program, which means agent-initiated transactions are starting to land in your chargeback queue with TAP consent records your tooling was never built to read. I am building Verdikt: we score each disputed agent transaction against its recorded consent scope and draft the accept, represent or write-off decision. I am running paid 30-day pilots with three programs this fall. Can I get 20 minutes this week to hear how you are handling these cases today?
landing page
Decide agent-transaction disputes against the consent record, not guesswork. $5,000 design-partner pilot: your last 90 days of agent disputes scored and drafted within 30 days, three slots. Book a scoping call and reserve your October slot.
deposit terms
Your $5,000 pilot fee is invoiced when you sign the one-page pilot order and reserves one of three October design-partner slots. It is refunded in full if we do not deliver scored decisions on your agreed dispute file within 30 days of receiving the data. Pilot pricing is credited against the $40,000 annual subscription if you convert.
Would you run this test?
One tap. The yes-share feeds the Demand pillar of this idea's score; nobody sees who answered.
Budgets are out-of-pocket estimates for a team of one to three, US market. Size the deposit to the deal, and check the terms before taking money in a regulated line.
Scorecard
Ranked against every idea in the catalog: trend, demand and 100x potential from the corpus, competition relative to the other ideas. A generated concept has no judges or swipes yet, so its pillars use the data signals only.
55
Idea Score, 0-100 · raw 33.6 x 1.64
Active
competition: more crowded than 69% of ideas · headwind x0.65
+0.0
government priorities, secondary (0 matching grants)
Trend
46
Is the wave forming now? 2025-26 entrants vs 2023-24, rounds since 2025, the sector's live-batch direction, the 2026 trend analyst.
- Entrants 2025-26 vs 2023-24 (similar companies)55
- Rounds announced 2025+ in the sector33
- Sector direction (live batch)50
Demand
65
Does anyone want it? YC's current RFS, companies already paid for something similar, the operator judge, founders' yes-rate in decks, readers who would run the test.
- YC asks for it (current RFS: idea / sector)30
- Someone already pays (similar companies, recent / all-time)100
100x potential
45
Can it return a fund? The venture judge (double weight), market-size and moat axes, neighbours still alive, the technologist judge.
- Neighbours still alive45
Score = 100 x cbrt(Trend x Demand x 100x) x (1 - 0.5 x crowding) + government bonus (max 5), calibrated so the 95th-percentile idea scores 90 (order never changes). A geometric mean: a weak pillar cannot be papered over. Percentiles are among the 382 ideas in the catalog; the terms matched were issuer-side, triage, disputes, agent-initiated, transactions, issuing, banks, card.
The concept in full
- What
- Verdikt gives issuing banks and card programs a decision layer for disputes where the transaction came from an AI agent: it parses Trusted Agent Protocol identities, consent records and token scopes to determine whether the agent acted within its mandate, then recommends accept, represent or write off. It replaces clickstream forensics that do not exist for agent purchases.
- Grounded in (2025-2026 signals)
- Visa's Trusted Agent Protocol (September 2025) with issuer-signed consent records, expanded to Visa Intelligent Commerce in 2026; Mastercard Agent Pay (April 2025) with consent policies bound to Agentic Tokens; both networks joining Google's AP2 in September 2025; Rapidfolio (yc S26) building AI for fintechs and banks signals issuers are buying AI tooling now.
- What it rides
- Visa and Mastercard agent rails: both networks now attach verified agent identity and consent policy to transactions, and Verdikt is the software that lets an issuer actually read and rule on those artifacts at dispute time.
- Why now
- Both networks attached consent records to agent transactions between April and September 2025 and consolidated integrations in 2026, so 2026 is the first year issuers hold cryptographic mandate evidence with no tooling to adjudicate against it; the YC Fall 2026 RFS 'Proving You're Human' points at the same distinction between human and agent action.
- Wedge: first customer and entry point
- One fintech card program or sponsor bank with early agent transaction volume; deliver a rules-plus-model service that scores each disputed agent transaction against its recorded consent scope and drafts the chargeback decision.
- Closest real companies, as the generator saw them
- Rapidfolio sells AI to fintechs and banks generally; MouseCat fights financial crime and Socratix AI serves merchant-side fraud teams. None tracked adjudicating agent mandate on the issuer side.
- Main risk
- Issuers wait for Visa and Mastercard to ship network-level dispute rules for agent transactions and buy nothing in the interim.
Similar startups in the directory
Companies whose pitch matches most of the concept's terms (issuer-side, triage, disputes, agent-initiated, transactions, issuing, banks, card).
AI powered reconciliation for high-volume payments companies
Foundation models on enterprise transaction data.
Recepta turns receipts into a powerful marketing tool, using data to understand customer behavior.
Financial infrastructure for the agent economy
Computer Agents for Finance tasks like reconciliation, in your system!
Endl - Global team spend, done right
AI that eliminates manual workflows, enabling businesses to serve more customers faster.
Paytweak offers innovative and smart payment services anywhere, at any time, on any device.
Card issuer processor and banking stack for businesses in Sub-…
Budgeting app with built-in banking
Modern finance platform for the Middle East
SaaS Intelligence for Banks to Boost their Profits on Card Transactions by 20%
Public money in this direction
US federal grants and open opportunities matched to the concept's terms.
No matching grants or programs tracked; 2 startup-relevant grants exist in Fintech overall.
Other concepts in this collection
- QuadrailOne endpoint that speaks ACP, UCP, AP2 and both card network agent protocols
- CartproofContinuous conformance testing for merchant agent checkout endpoints
- TokentraceReconciliation built around scoped payment tokens for PSPs handling agent orders
- CountersignDispute defense for agent purchases using issuer-signed consent records
- DutylineLanded cost inside the agent checkout response for cross-border orders
- ConsentryVerification gateway that tells merchants which agent orders to trust
- KeyfleetCredential lifecycle for every AI agent a company lets spend money
- MintgatePolicy engine deciding when an agent gets a scoped payment token
- TrailproofAn audit agent that reconstructs consent for every dollar your agents spent
- HoldpointHuman approvals that become signed consent records for agent purchases
- MeterlinePer-agent stablecoin allowances and kill switches for machine-to-machine payments
- VetgateOne decision point for banks verifying agent credentials across every rail
- AttestaChargeback evidence packets built from agent consent records
- ReversioRefund rails for purchases made with expired scoped tokens
- TallybridgeReconciles agent-originated orders against ERPs and the general ledger
- RefluentRefund and reconciliation layer for stablecoin agent payments
- DutybackRecovers duties and tax when cross-border agent orders come back
- SpendfenceBudgets, scoped tokens and approval policy for every agent a company runs
- NetformNetting and clearing for millions of tiny machine-to-machine payments
- CovenorYour AP agent negotiates terms and settles invoices with their AR agent
- FloatworksTreasury operations for the stablecoin float that funds agent budgets
- Recourse LabsDispute and refund rails for payments agents make on their own
- CharterlineLicensing and exam readiness software for stablecoin issuers under the GENIUS Act
- RelayproofTravel rule data for stablecoin transfers that an agent initiated
- SievewireSanctions and AML screening built for machine-speed agent payments
- AttestlyOne evidence vault for agent checkout consent records and scoped tokens
- AmletAML program operations run by agents for seed-stage stablecoin PSPs
- YieldgateProof that stablecoin rewards are activity linked, not prohibited interest
- ClearquoteLanded cost inside the agent's checkout quote, duties and taxes priced before the token
- DrawpointAn agent that files duty refunds when cross-border agent orders come back
- PegwiseStablecoin settlement between the agent's currency and the merchant's, with locked FX
- DutygraphThe customs knowledge base agents query before buying across a border
- PortruleSpend policy for purchasing agents that buy from foreign suppliers
- AttestoRegulator-ready evidence for every cross-border stablecoin settlement an agent triggers
- ParflowRedemption operations API for stablecoin issuer programs
- StablebooksTreasury policy and reconciliation layer for companies holding stablecoins
- CorridorlinePayout orchestration across stablecoin rails with compliance evidence built in
- RegimarkAgent service that builds dual GENIUS and MiCA compliance evidence for issuers
- RewardrailCompliant activity-linked rewards engine for stablecoin issuers and platforms
- FeedstoneAn agent that builds and maintains your UCP and ACP catalog endpoints
- TracelightAttribution and revenue reporting for orders placed by AI agents
- GatewickThe traffic gate that tells buying agents from scraping bots
- ClausemarkPublish return, shipping and warranty policies agents can read and bind to
- CounterproofChargeback evidence built from agent consent records and scoped tokens
- StockproofReal-time inventory and price truth for agent catalog queries
Fictional concept generated 2026-08-26 by claude-fable-5 from the collection's brief and MarkosWeb data. Treat it as a research prompt, not a plan.