AI revenue recovery · Razorpay AI Buildathon

Recover revenue.
With receipts.

RecoverFlow diagnoses why a payment failed, recommends one bounded action, and leaves a tamper-evident trail. The AI proposes; policy decides; a human approves.

R run · T theme
EngineREADY
Webhook receiverCHECKING
Audit chain—
Synthetic dataset · no real payments are moved

Revenue at risk

₹0across 0 payment events

Recovered this sweep

₹00% of recoverable value

Awaiting approval

0actions gated by a human

Diagnosis confidence

0%average on eligible cases

Live workflow

The recovery engine

Idle
01DetectFailure and intent signals0
02DiagnoseEvidence-grounded cause0
03BoundPolicy and stop rules0
04RecoverApproved, measured action0

AI judgment, not AI autopilot. The agent classifies the cause and drafts the action. Deterministic policy is the final authority, and every decision is logged with its evidence.

Loss drivers

Run a sweep

Computed

Run a sweep to rank what is costing you the most, by value at risk.

Evidence coverage0%

Priority queue

Interventions that need a decision

CustomerAt riskAI diagnosisConfidenceRecommended recoverySafety

Razorpay webhooks

Live feed

Waiting

Signed payment.failed events from Razorpay Test Mode appear here. The HMAC signature is verified before anything is accepted, and each event is processed once.

/webhooks/razorpay
  1. Deploy and set RAZORPAY_WEBHOOK_SECRET.
  2. Add the URL above in Dashboard → Webhooks (Test Mode).
  3. Trigger a failed Test Mode payment.

No webhook events yet. When one arrives it is verified, diagnosed, policy-checked and shown here.

Interactive proof

Decision lab

Edit the payment event, send it to the live diagnosis and policy service, and inspect exactly why it was allowed or held.

  • Signed events in production
  • No automatic charge, ever
  • Fully explainable result
POST /api/diagnose
Choose a scenario and evaluate it.

Batch outcome simulator

Measure recovery without hiding risk.

An illustrative model comparing generic retry-everything recovery to RecoverFlow’s gated approach. Assumptions are shown with the result.

Generic recovery

Retry every failure, then chase manually
Recovered
—
Unsafe or wasted actions
—
Human reviews
—

RecoverFlow

Diagnose, bound, approve, measure
Recovered
—
Protected by policy
—
Human reviews
—

Run the simulation to see transparent, reproducible numbers.

Explainability

Immutable decision trail

Local session

No sweep has been run. Every decision will appear here with its evidence, policy status and outcome.

Non-negotiable guardrails

What the agent will never do

One retry per 24 hours

A second failure is held for a human. Repeated retries erode trust and trigger issuer blocks.

retry_limit_reached

No contact after opt-out

Consent is checked before any message is drafted. No consent, no outreach.

NO_CONSENT

No unapproved money action

Retries, refunds and payments always wait for explicit human approval.

HUMAN_GATE

No hidden exceptions

Duplicates, disputes and low-confidence cases stay visible in the queue.

EXCEPTION_QUEUE

Human in the loop

Approve recovery action?

Recovery case file

Payment case