Track 03 • AI Revenue Recovery & B2B Receivables

An AI agent for Indian B2B receivables that decides — per debtor, not per invoice — whether to wait, collect, negotiate, reconcile, or escalate.

Empirically measured recovery with a tamper-evident audit trail · Built for Razorpay AI Buildathon 2026 (Track 03). Grounded in MSMED statutory timelines, Section 43B(h) tax disallowance, and Form 26AS TDS withholding compliance.

Core Architecture Differentiators

Intelligent Restraint

The agent treats WAIT as a first-class strategy. Account-level historical payment reliability and active opt-outs trigger immediate suppression, preventing relationship damage.

India Statutory Grounding

Form 26AS TDS reconciliation, Section 43B(h) non-adversarial tax cues, and MSMED Section 15/16 eligibility gates. The agent refuses statutory threats when merchants are ineligible traders.

Counterfactual Evaluation

Head-to-head empirical evaluation against a rigid calendar dunning baseline across an identical seeded portfolio. Measured recovery delta, lower touch noise, and zero wrongful escalations.

End-to-End Decision Architecture

Autonomous five-stage execution lifecycle per commercial debtor account

Stage 01

Seeded Ledger

Ingests multi-invoice debt, UTR records, and payment history per debtor.
Stage 02

Policy Envelope

Deterministic rules bar unlawful citations, enforce opt-outs, and bound asks.
Stage 03

AI Strategist

Gemini reasons across tenure and aging to select 1 of 7 calibrated actions.
Stage 04

Channel Dispatch

Dispatches polite bilingual notices via WhatsApp & Email with Razorpay links.
Stage 05

Settlement Suppression

Webhook instantly halts active recovery the moment payment lands on-rail.

Application Surfaces

Direct access to operator, benchmark, and debtor portals

Curated Demo Invoices

Live resolvable receivable items with verified live balances

Invoice ID Debtor Account Outstanding Balance Aging Status Action
INV-4295 Coastal Marine Supplies โ‚น2,63,200.00 79d Overdue OVERDUE Resolve