Chargeback Representment & Dispute Recovery Engine

01 Thesis

Card disputes have become a structural tax on online commerce. Global chargeback volume is heading from roughly $33.8B in 2025 toward $41.7B by 2028, the large majority of it "friendly fraud" — legitimate customers disputing legitimate purchases — and most merchants either don't fight back or fight badly because the work is deadline-bound, reason-code-specific evidence assembly that nobody on staff has time to master. The same merchants now face Visa's consolidated monitoring program (VAMP), whose merchant threshold tightens to 1.5% on April 1, 2026, with real fines and account-termination risk above it.

Representment is a near-perfect AI-native production problem: pull the dispute and its reason code, gather the order, device, AVS/CVV, fulfillment/tracking, prior-transaction and login data, and compose the specific rebuttal the network rules require — fast, before a window that for Visa is now as short as ~9 days. Merchants who fight well win materially more than those who don't, and specialist platforms already demonstrate the gap. We do this as a managed outcome: connect to the merchant's processor, run the engine, keep a human analyst at the judgment chokepoint (which disputes are winnable, which evidence is honest and sufficient, which to concede), and bill only a share of what we recover.

02 Discovery rationale

This run searched recovery-and-compliance terrain where (a) money is being lost on a recurring, high-frequency cadence, (b) the work is data-and-document bound with a hard external deadline, (c) a regulator-or-network rule creates additional pull beyond the recovered dollars, and (d) pricing can be cleanly outcome-based. Card-dispute recovery surfaced as the strongest currently-active candidate: multiple independent 2025–2026 sources quantify a tens-of-billions loss pool growing double-digits, a documented majority-friendly-fraud root cause, an established contingency-fee service category (validating budget and willingness to pay), and a fresh, dated regulatory driver — Visa's VAMP consolidation and threshold tightening through 2025–2026 — that converts "nice to have" recovery into "keep my account open" urgency.

It also screened out a tempting alternative. WOTC (Work Opportunity Tax Credit) point-of-hire capture initially looked attractive (clean per-hire contingency, 28-day deadline, billions unclaimed) but the program's authority lapsed on January 1, 2026 and remains in legislative hiatus as of this run — an unresolved regulatory blocker that would put the entire revenue base on a coin-flip reauthorization. Card-dispute recovery has the opposite property: the driver is live, dated, and tightening. It is also clearly distinct from all prior blueprints in the manifest (no prior touches card networks, PSP integrations, or chargeback representment).

03 Candidate comparison

Six candidates were considered this run and scored on demand evidence, distinctiveness vs priors, margin/automation potential, outcome-pricing fit, and regulatory/licensing safety (1–5).

The WOTC contrast is the instructive one: identical structural appeal (deadline-bound, contingency-priced, billions unclaimed) but the regulatory ground gave way under it this year. Selecting on a live, tightening network rule rather than a lapsed statute is the deciding move.

04 Hard disqualifier check

Each fatal disqualifier from the operating standard, tested against this candidate.

No fatal disqualifier triggers. The only open question is the margin/automation kill-metric, addressed explicitly in scaling and carried as a pilot stop-condition.

05 Rubric scorecard — the six gates

Composite ≈ 4.3/5. Gate 4 and Gate 6 are the soft spots: the "moat" is network-rule complexity and integration breadth rather than a licensing monopoly, and the same model improvements that help us also help PSPs and networks fold representment in-house. Both are addressed in the moat and anti-commoditization section, and both are reasons the compliance/monitoring outcome (not just recovery) is the durable wedge.

06 CODE validation

07 Opportunity

Figures are drawn from vendor and analyst reports (chargeback specialists, market-research firms, network fact sheets) and verified in two or more places where possible. Vendor-published win rates (e.g. "up to 80% with automation") are marketing-adjacent and treated as Unverified ceilings, not planning numbers; the 54% average-when-fought figure is the planning anchor.

08 Evidence & source-claim matrix

Every load-bearing claim, its label, source, type, date/access, confidence, and where it is used. Access date for all web sources: 2026-06-30.

09 Why now

• Visa pulled the threshold tighter, on a clock. VAMP consolidated three legacy programs in 2025; Excessive-level enforcement began Oct 1 2025; the merchant ratio drops to 1.5% on Apr 1 2026. Merchants who were "fine" at 2.2% can be non-compliant overnight — a dated, external forcing function. V
• Friendly fraud is structurally rising. Dispute rates climbed ~53% across 2025 in at least one large index; CNP/subscription/digital-goods mix amplifies it. The loss pool grows every quarter. V
• The win is mostly mechanical and now AI-tractable. CE 3.0 turned "compelling evidence" into a structured, rules-defined data exchange (Order Insight) — exactly what an AI engine plus a thin review layer can assemble at scale and at low marginal cost. V
• Windows shrank. Visa's response window is now as short as ~9 days in the US/Canada — punishing manual, human-only desks and rewarding an always-on automated intake-to-submission engine. V
• Budget already exists and is contingency-shaped. Merchants already pay 20–25% of recovery to specialists; we are redirecting an existing budget line, not creating one. V

10 Customer & product-market fit

Ideal customer profile (ICP)

Jobs-to-be-Done

• Functional: "Recover the revenue customers wrongly clawed back, without my team touching it, before the deadline."
• Functional (compliance): "Keep my fraud-plus-dispute ratio under VAMP so my processor doesn't fine or fire me."
• Emotional: "Stop dreading the dispute queue and the acquirer's threatening emails."
• Social: "Show my board/investors that payment risk is handled by specialists."

11 The outcome we sell

A done-for-you guarantee that every winnable dispute is fought with the right reason-code-specific evidence and filed before the network deadline, recovered revenue is returned to the merchant, and the merchant's fraud-plus-dispute ratio is monitored and managed to stay under Visa VAMP and equivalent Mastercard programs — delivered as a managed service with a dispute analyst as the named owner, not a dashboard the merchant operates.

We are explicit that we are not selling "fewer chargebacks" by magic, nor advising merchants to misrepresent transactions. We sell recovered dollars on genuinely defensible disputes plus a managed ratio. Where a dispute is legitimate (true fraud, a real service failure), the honest move — concede and route to prevention/refund — protects the merchant's ratio better than a doomed fight, and we say so.

12 First one-feature MVP wedge

13 Internal AI engine architecture

Designed as an internal operating engine. The merchant's interface is the analyst and the recovery report — never a tool they must drive.

14 AI-vs-human operations pipeline

Roughly 8 of 10 steps are AI/rule/ops; the two human steps are the fight/concede judgment and the honesty/sufficiency sign-off — the exact places where a wrong call costs money or violates network rules.

15 Dynasty translation layer

16 Operations as product

• SOPs per reason code — a documented evidence checklist and rebuttal template for each Visa/Mastercard/Amex reason code, versioned as rules change.
• Structured intake — a required-evidence manifest per dispute type; automated completeness checks block submission of thin packets.
• Exception queues — low-confidence, high-value, or novel reason codes route to senior analysts; everything else flows on the standard path.
• Reviewer assignment logic — by confidence × disputed amount × reason-code risk; sampling rate tightens after any loss-pattern spike.
• Confidence & ROI scoring — every dispute carries a win-probability and net-recovery score driving the fight/concede decision.
• Audit trail & version control — immutable log of evidence used, rule version, analyst sign-off, and submission confirmation for every case.
• Gold-standard library — exemplar winning packets per reason code; red-team checks for fabricated/misleading evidence.
• Root-cause & postmortem loop — every loss is categorized (bad evidence, wrong code, missed deadline, genuinely unwinnable) and feeds SOP/rule/prompt fixes.

17 No-holes quality engine

Quality here has two failure surfaces: losing winnable disputes (revenue holes) and submitting misleading or non-compliant evidence (existential network-rule holes). The engine guards both.

18 What the human expert actually does

No professional license is legally required to act as a merchant's agent in representment. The "expertise" is network-rule fluency and judgment, not a credential — which is why training and SOPs, not licensing, are the scaling constraint. Target: total analyst minutes per dispute compress from ~22 at launch toward <10 by day 90.

19 Pricing & unit economics

Pricing model (outcome-based, never hourly)

• Core: 20–25% contingency of dollars actually recovered. Pay-for-performance — no recovery, no fee — matching the established market norm (Chargeflow 25%, Justt "pay when you win"). V
• Add-on: VAMP/ratio-monitoring subscription — a modest flat monthly for continuous fraud-plus-dispute ratio monitoring, breach early-warning, and prevention-alert routing (RDR/CDRN/Ethoca). Recurring base that smooths contingency lumpiness.
• Add-on: prevention alerts priced on prevention — charge only for a chargeback actually stopped (mirrors Chargeflow's prevented-only model), never per-alert.
• Never: hourly, per-seat, or cost-plus.

Illustrative unit economics (per recovered dispute, modeled — Inferred)

Targets: revenue/FTE $0.5–1.0M; blended gross margin 50–60% at scale; CAC payback < 6 months on a contingency base; rework/loss-investigation < 8% of cases. All except the verified pricing norm are Inferred and carried as pilot kill-criteria.

20 Nonlinear scaling & the central kill-metric

Revenue scales with dispute volume across merchants, which the engine processes at near-zero marginal model cost; headcount scales only with the residual judgment minutes per dispute. The whole thesis lives or dies on one number: analyst minutes per dispute, and whether they compress cohort-over-cohort.

Build-before-scale checkpoints: After 5 pilot merchants, harden intake + reason-code SOPs + the anti-fabrication gate. After 10, harden exception queues, reviewer checklists, and per-vertical evidence templates. After 20, pause new pilots until COGS, review-minutes, win-rate-by-code, missed-deadline rate, and loss-investigation rate are measured and the margin trend is confirmed. If review-minutes are not compressing toward target by the 20-merchant gate, the model is a labor-arbitrage shop, not an AI-native business — stop and re-architect.

21 Moat, Sam Altman test & anti-commoditization

Moat

• Outcomes data — proprietary win/loss by reason code × issuer × evidence-type × PSP becomes the highest-signal asset for evidence selection and fight/concede calls; it compounds with volume and can't be bought.
• Network-rule operating system — a maintained, versioned library of reason-code SOPs and CE 3.0 data-element logic that tracks constant Visa/Mastercard changes; painful to keep current, valuable precisely because it's painful.
• Dual outcome — recovery plus VAMP-ratio management; the compliance outcome raises switching costs and reframes us from "an app" to "the team that keeps the account alive."
• Integration breadth — coverage across PSPs/acquirers and the prevention networks (RDR/CDRN, Order Insight, Ethoca) is a real, sequenced build that late entrants must replicate.

Sam Altman test — does it get stronger as models improve? Yes, 4.0

Better models assemble more persuasive, more complete evidence with fewer human edits, lift win rates, and shrink the review-minute kill-metric — directly expanding margin and the addressable set of disputes worth fighting. Our durable assets (rules library, outcomes data) are model-portable, so model gains accrue to us.

Anti-commoditization — what if general models make this self-serve?

The honest threat is a pincer: (1) PSPs (Stripe/Adyen/Shopify) and the networks themselves (Verifi/Ethoca, Order Insight) bundle "auto-fight your disputes" into the checkout stack; (2) cheap general models let any merchant self-assemble evidence. Mitigations: we sell the managed compliance outcome and accountability (a named team that keeps you under VAMP), not the keystroke; we go where PSP-native tooling is weakest (high-risk multi-PSP merchants the big platforms under-serve or de-risk away from); we own the cross-PSP outcomes data no single platform has; and we offer to white-label the engine to acquirers/ISOs — turning the most likely commoditizers into channel partners rather than competitors. This is the single biggest strategic risk and is tracked as such in the kill section.

22 Go-to-market & distribution

Distribution proof table

Founder/expert-led content plan

Teach the buyer to understand VAMP math, CE 3.0 qualification, reason-code evidence, the cost of conceding everything, and the cost of fighting badly. Repurpose the best organic pieces into paid-ad creative once they prove out.

Warm GTM & targeted outbound

Convert scan/lead-magnet users with a personalized "Recovery + Ratio" memo (their numbers, their threshold gap). Outbound leads with a diagnosis ("your vertical's dispute rate vs yours"), never a generic demo ask.

23 First 30 days of content

10 educational posts

1. "VAMP 1.5%: the April 2026 threshold that can get your merchant account closed."
2. "Friendly fraud is ~75% of your chargebacks — here's how to prove it."
3. "Compelling Evidence 3.0, explained: when it wins and when it doesn't count."
4. "You have ~9 days: the new Visa response window and why manual desks miss it."
5. "Why conceding the right disputes protects your ratio better than fighting all of them."
6. "Reason code 10.4 vs 13.1: the evidence each actually requires."
7. "RDR, CDRN, Order Insight, Ethoca — the pre-dispute tools that keep cases off your ratio."
8. "Subscription merchants: the three evidence elements that win recurring-billing disputes."
9. "The true cost of a chargeback isn't the chargeback — it's $4.61 per fraud dollar."
10. "How to read your processor's VAMP warning email before it becomes an offboarding."

3 diagnostic teardown formats

• "Live teardown: a losing representment packet and the 4 fixes that would have won it."
• "Ratio teardown: this merchant was at 2.0% — here's the 60-day path under 1.5%."
• "Reason-code teardown: 20 disputes, which to fight, which to concede, and why."

2 lead-magnet angles

• Free "VAMP Exposure & Recovery Scan" (their ratio vs threshold + recoverable-dollar estimate).
• "Reason-Code Evidence Checklist" PDF — the required evidence per top-15 codes.

1 webinar / live review

"VAMP Survival Clinic: bring your dispute numbers, leave with a ratio plan." (Live scan of 3 volunteer merchants.)

1 outbound diagnosis template

"[Name], merchants in [vertical] average ~[X]% disputes; at your volume that's ~$[Y]k/yr in recoverable revenue and a ratio of ~[Z]% vs the 1.5% VAMP line. Want a free scan of your actual numbers?"

24 Pilot design, feedback flywheel & checkpoints

25 Competitive landscape & budget validation

Budget validation: Merchants already pay 20–25% of recovery to these vendors; a $0.9B+ NA services market growing ~13% confirms the budget line exists. "No competitors" is not the situation — the situation is a crowded but fragmented market where the high-risk mid-market done-for-you + VAMP-compliance niche is under-served. Our risk is commoditization from above (PSPs/networks), addressed in the moat and kill sections.

26 Regulation, compliance & licensing boundary

The real compliance line: the business must never become a "win at all costs" representment mill. Submitting misleading evidence or fighting legitimate disputes risks network sanctions, merchant offboarding, and reputational ruin. The honesty gate and fight/concede discipline are not nice-to-haves — they are the license to operate.

27 Exhaustive risk register

28 Tech stack & build plan

29 Metrics & KPIs

30 What could kill this

• Commoditization from above — PSPs/networks bundle representment so deeply that done-for-you loses its reason to exist. Leading indicator: Stripe/Adyen ship credible auto-representment with ratio management; hedge: white-label channel + high-risk niche.
• Margin never crosses 50% — review-minutes stay high on messy data. Indicator: flat minutes/dispute by the 20-merchant gate.
• An honesty scandal — one fabricated-evidence incident triggers network sanction and reputational collapse. Hedge: the integrity gate is non-negotiable.
• PSP ToS lockout — processors restrict third-party submission. Hedge: partner status + merchant agency + multi-PSP diversification.

31 7 / 30 / 90-day launch plan

First 7 days

• Stand up the "VAMP Exposure & Recovery Scan" lead magnet + landing page.
• Publish 3 cornerstone posts (VAMP 1.5%, CE 3.0, the ~9-day window) + reason-code checklist PDF.
• Build the Stripe dispute-webhook ingester + a manual analyst case desk.
• Line up 8 pilot prospects from 2–3 high-risk verticals.

First 30 days

• Onboard 3–5 pilots; run representment manually-heavy; instrument minutes/dispute from case one.
• Ship reason-code evidence templates + completeness lint + anti-fabrication gate.
• Launch outbound diagnosis sequence to the two priority verticals.
• Stand up weekly case-review + loss-postmortem cadence.

First 90 days

• Reach 8 pilots; add CE 3.0 / Order Insight submission + ROI/win scoring + auto-assembly.
• Hit the build-before-scale gate: confirm minute-compression, win-rate-by-code, missed-deadline=0 before adding merchants.
• Launch the flat ratio-monitoring subscription as the recurring base.
• Sign the first acquirer/ISO white-label conversation; productize per-vertical evidence packs.

32 Sources

Accessed 2026-06-30. Vendor/market-research figures are marketing-adjacent and labeled accordingly in the source-claim matrix; network fact sheets and government pages are primary.

• Chargeflow — Chargeback Statistics 2026 (volume/cost projections)
• Chargeback.io — Chargeback Statistics 2026
• Chargeflow — Friendly Fraud: the $132B threat
• Chargebacks911 — Chargeback Stats 2026
• Kount — Chargeback Representment Guide (win rates & evidence)
• Market.us — Chargeback Management Market (size, CAGR, incumbents)
• Visa — Acquirer Monitoring Program (VAMP) Fact Sheet 2025 (primary)
• Merchant Risk Council — Stricter VAMP thresholds now in effect
• Chargeflow — Visa VAMP Rules 2025 guide
• Forter — Visa's updated VAMP program (CE 3.0 exclusions)
• Ravelin — VAMP changes to dispute thresholds (May '25 update)
• ChargebackHelp — 2025 chargeback time limits
• Chargeblast — Mastercard chargeback timeframe changes 2025
• Chargeflow — Pricing (25% success fee, pay-for-performance)
• Justt — FAQ (success-based pricing)
• Sift — Q4 2025 Digital Trust Index (dispute-rate trend)
• Swell — High-risk eCommerce statistics 2025
• OpenPR — Chargeback management software market & leaders
• IRS — WOTC available until end of 2025 (lapse context, primary)
• Experian — WOTC renewal & 2025 expiration (candidate-rejection evidence)