Scene 01
The money is already yours
A global customer reaches your checkout, card in hand, intent locked. At this moment the revenue already belongs to you — it just hasn't moved yet.
Track 03 · AI Revenue Recovery
FrictionGuard intercepts international payment failures, diagnoses them with sandboxed AI, and recovers the sale — but only after a deterministic Policy Engine approves the move.
0.0%
Net recovery rate
$0
Recovered TPV
0.0%
Override rate
Reference results from the supplied project brief · 50-case evaluation
Lightweight visual mode (reduced motion / no WebGL)
initializing recovery pipeline
Scene 01
A global customer reaches your checkout, card in hand, intent locked. At this moment the revenue already belongs to you — it just hasn't moved yet.
Scene 02
A forced 3DS challenge, an issuer soft decline, a generic error. International checkouts fail in ways domestic ones never do — and most systems turn a temporary failure into a permanent goodbye.
Scene 03
Before the customer sees a dead end, FrictionGuard intercepts the failure, normalizes it into a structured signal, and starts the recovery pipeline — in seconds.
Scene 04
Groq's openai/gpt-oss-20b classifies the root cause, proposes a bounded recovery action, and drafts customer microcopy — structured JSON, validated by Zod, with a calibrated confidence score.
Scene 05
The AI can diagnose. It cannot create payments, move money, change amounts, or approve itself. Its proposal is just that — a proposal.
Scene 06
A deterministic Policy Engine checks every proposal: confidence floor, amount ceiling, retry caps, allowed actions, duplicate prevention, session validity. APPROVE, OVERRIDE, or BLOCK — nothing in between.
Scene 07
Approved recoveries execute safely: a bounded retry, a lighter alternative method, or an async Payment Link that waits for the customer to come back.
Scene 08
Failed → intercepted → diagnosed → approved → recovered. The sale that was gone comes back and completes — verified server-side, recorded in Turso, visible on the dashboard.
Scene 09
A built-in 50-case evaluation replays deterministic international failures through the real Policy Engine: 74% net recovery, $4,250 recovered TPV, 6% override rate — evidence the gate actively catches unsafe proposals.
Enter a real Razorpay Test Mode checkout, break it on purpose, and watch FrictionGuard diagnose, decide, and recover — then verify everything in the dashboard and the 50-case evaluation.
AI PROPOSES → POLICY DECIDES → SYSTEM EXECUTES SAFELY