DisputeIQ is an AI-native intake layer for credit card dispute operations. Multi-agent classification, adaptive evidence collection, and real-time completeness scoring — built for modern card issuers.
| Cardholder | Merchant | Amount | Category | Score | Status |
|---|---|---|---|---|---|
| J. Mitchell | Netflix Inc. | $15.99 | Cancelled Sub | 92 | Complete |
| A. Patel | Amazon MKTP | $248.00 | Not Delivered | 71 | Review |
| K. Torres | Uber Technologies | $94.50 | Duplicate | 55 | In Progress |
| S. Williams | Marriott Hotels | $340.00 | Hotel No-Show | 88 | Complete |
DisputeIQ replaces brittle static forms with a conversational AI agent that knows exactly what evidence each dispute category requires.
Analyzes transaction context and cardholder language to assign one of 15 dispute categories with a calibrated confidence score. Can create the case record in Supabase on first contact.
Agent 1 →Determines the single most important missing field and asks a short, customer-friendly follow-up — fact-aware, never repeating what's already known, capped at 12 turns per case.
Agent 2 →Reviews the completed intake and recommends 2–5 high-value supporting documents specific to the dispute category — cancellation emails, receipts, shipping confirmations, and more.
Agent 3 →Scores every turn across 5 dimensions: category fit, intake completeness, response consistency, question relevance, and investigation readiness. Routes to the correct next agent.
Agent 4 · Orchestrator + Evaluator →Generates a structured 3-section brief — Reason for Dispute, Key Details Collected, and Current Status — from the full Q&A history. Ready for back-office review with no rewriting.
Agent 5 →Coordinates all five agents in sequence every turn, maintains stateful conversation sessions, and routes outputs between agents — all in under a second per message.
Express · ESM · Sub-agent SDK →DisputeIQ guides cardholders through a structured intake flow — and every agent handoff happens invisibly in under a second.
User taps Dispute on a transaction. Metadata is passed to the orchestrator instantly.
Assigns one of 8 dispute categories with a confidence score. Re-runs every turn as context grows.
Picks the most critical missing field and asks a natural follow-up — never repeating what's already known.
Structures each answer into a formatted evidence record mapped to the correct chargeback field.
Confidence & completeness evaluated 0–100. Submit unlocks at 80+.
Summarization agent generates a structured dispute brief — ready for bank review or chargeback filing.
A cardholder disputes a duplicate charge. Watch confidence classify it, completeness score build, and evidence collect — turn by turn.
Maps complaint + transaction to one of 15 dispute categories. Creates the case record in Supabase on first contact via tool call.
Asks the next best follow-up — fact-aware, never re-asking answered questions. Max 12 turns, computes dispute amount where inferable.
Recommends 2–5 high-value supporting documents specific to the dispute category and conversation collected.
Scores 5 dimensions per turn, determines routing action (CLASSIFY → CONTINUE → EVIDENCE → SUBMIT → HUMAN_REVIEW), flags human review below threshold.
Generates a structured 3-section case brief — reason, key facts, current status — from the full conversation history.
n8n orchestrates the workflow. Supabase persists case state, turn history, scores, routing decisions, and the final intake summary.
Instead of a single monolithic prompt, DisputeIQ uses five focused sub-agents — each an expert in exactly one job — coordinated by a stateful orchestrator. The result is higher accuracy, lower latency, and full auditability at every step.
Built with
Whether you're a card issuer looking to automate intake, a fintech exploring AI-powered chargeback tooling, or a product team evaluating this architecture — reach out. Happy to demo, consult, or collaborate.
Tell me about your use case and I'll get back to you shortly.
DisputeIQ is built by Ravi Appana — product builder and AI engineer focused on financial services automation. Available for demos, integrations, and consulting.
Walk through a full dispute intake session — end to end in under 3 minutes.
Drop the REST endpoints into your existing issuer portal or agent desktop.
Need more reason codes, custom scoring, or workflow triggers? Let's scope it.