One graph, two wins: How AI-powered payment reconciliation connects chargeback evidence to fee-overcharge recovery

When your payment stack spans cards, wallets, UPI, RTP/FedNow, and multiple PSPs, two headaches dominate finance ops: chargebacks that eat margin and fee overcharges that hide inside blended pricing. They look like separate problems. They’re not. With a single event-lineage graph—auth → capture → shipment → settlement → bank deposit → dispute—you can auto-assemble evidence packs for representment and recompute fees line-by-line to expose markups and contract drift. One graph, two wins.

This matters at scale. U.S. merchants paid $172.05B in processing fees in 2023; a few basis points of drift is seven-figure money on nine-figure volumes. Pair that with rising dispute noise—fraudulent chargebacks are projected to cost businesses $15B in 2025, with roughly 45% first-party misuse—and the cost of operating without lineage becomes obvious.

The core idea: unify payments into a living evidence & economics graph

On Optimus Payment Reconciliation, we standardize every source (PSPs, acquirers, gateways, real-time rails, bank files, ERPs) into a canonical schema and stitch entities into a causal graph:

• Entities: payment, authorization, capture(s), refund(s), chargeback(s), settlement leg(s), bank deposit, fee lines.
• Edges: references (PSP Txn ID, ARN/UTR), time windows, fuzzy matches with explainable confidence.
• Artifacts: contracts parsed into a versioned rate engine; shipping/fulfillment proofs; risk decisions.

With this structure, the system can (a) assemble dispute evidence in seconds, and (b) recompute “what you should have paid” versus “what you were invoiced.” The same graph powers both.

Part 1 — Chargebacks: from mystery write-offs to evidence-ready win rates

Problem. In omnichannel flows, data is fragmented across PSPs, order systems, shippers, and banks. Without lineage, disputes become manual scavenger hunts and win rates stagnate (many merchants report sub-20% on fraud-coded chargebacks).

Solution with the graph.

1. Event stitching at T+0/T+1. As soon as an authorization occurs, the graph tracks the intended lineage: auth → capture(s) → settle(s) → deposit. When a dispute arrives, the system already knows the most probable order/shipment and fee context.

2. Auto-built evidence packs. The workbench compiles: proof of delivery or service, device/IP and 3-DS data, AVS/CVV results, customer communications, and the financial trail (auth→capture→settlement→deposit). Analysts only add nuance; no spreadsheet archeology.

3. Exception taxonomies and SLAs. Disputes get routed by cause (true fraud vs. first-party misuse vs. merchant error). Your KPIs become predictable: dispute linkage rate, representment win rate, days-to-resolution.

Why now. Real-time rails compress timelines. RTP processed $481B in Q2-2025, and FedNow counts 1,400+ participants—funds move faster, so evidence must as well.

Part 2 — Fee overcharges: unblending scheme rules from processor markups

Problem. Blended MDR obscures the split between scheme-mandated components (interchange; network dues/assessments) and processor economics (markups, gateway add-ons). Dues/assessments look tiny—often ~0.13–0.15% domestically—but compound at volume; cross-border fees can add 2–3%. Without line-level recomputation, drift goes undetected.

Solution with the graph.

1. Contracts → code. We convert your pricing schedules and public network tables (e.g., Visa U.S. Interchange effective Oct-2025) into a versioned rate engine keyed by brand/product/MCC/CP-CNP/geo/FX/tiers.

2. Deterministic recompute for every settlement leg. For each leg, the engine calculates expected interchange + dues/assessments + contracted processor fees, then compares to the invoiced amount, producing a variance vector (bps + currency) with root-cause labels: tier misclassification, cross-border misflag, assessment uplift, FX spread, gateway markup, rounding.

3. Anomaly models for drift. Seasonality-aware baselines alert only when variance is statistically meaningful and persistent (e.g., ≥0.8 bps for 5+ days in an EEA cross-border corridor). That keeps teams focused on recoverable dollars, not noise.

Outcome. Across diversified PSP mixes, merchants commonly uncover 0.5–2.5 bps of recoverable leakage once fees are recomputed line-by-line and monitored for drift. Pair this with macro fee totals and the P&L impact becomes board-level. (For context: $172.05B in U.S. fees in 2023; the figure rose further in 2024).

Deep dive: We unpack fee math and recovery steps in our companion post, “AI-Powered Fee Validation for Merchants”—read it on the Optimus blog.

Why one graph enables both use cases

• Same identifiers, different outcomes. The keys that win disputes (auth ID, ARN/UTR, settlement IDs, shipment proofs) are the same keys you need to recompute fees and audit processor invoices.
• Controls meet close. Evidence packs for disputes and rate-engine diffs for fees both flow into the reconciliation workbench and onward into subledger/ERP postings with maker-checker.
• Instant rails reset expectations. With RTP/FedNow and UPI scaling (UPI processed 20.0B transactions in Aug-2025), finance controls must operate on streams, not only month-end files.

Explore how we operationalize this inside Payment Reconciliation.

Operating model: lean team, measurable outcomes

• Exception taxonomies.
  • Disputes: true fraud, first-party misuse, merchant error, logistics mismatch.
  • Fees: tier miss, assessment uplift, cross-border misflag, FX spread, gateway markup, rounding, contract-not-mirrored.
• SLAs & queues. Prioritize high-value corridors and aging disputes; target ≥90% resolution <24h for evidence-complete disputes and fee variances.
• CFO-grade KPIs.
  • Representment win rate, days-to-resolution
  • Recovered basis points (RBP) and fee variance detected (FVD) per 10k settlements
  • Approval-adjusted unit cost (true cost per successful txn)
  • Close predictability: operational T+2, GL T+5

60-day rollout (practical and fast)

Days 0–15 — Wire the graph. Connect two PSPs + one bank; ingest 60 days of payments/settlements/disputes; confirm keys (PSP Txn ID, ARN/UTR); map logistics data (carrier, delivery, proof).

Days 16–30 — Turn on packs & recompute. Auto-build dispute evidence packs; enable line-level fee recomputation; start anomaly monitors; route both dispute and fee variances to the workbench with maker-checker.

Days 31–60 — Recover & report. File representments with packaged evidence; open recovery cases on sustained fee variances; publish a RBP leaderboard by connector; feed realized outcomes back into thresholds. Post clean entries to ERP via subledger.

All of the above runs natively in Optimus Payment Reconciliation, with dashboards and audit-ready exports.

The payoff

Unifying finance around a single event-lineage graph converts two chronic expense lines—chargebacks and fee overcharges—into controllable, measurable cost levers. You’ll resolve more disputes, faster. You’ll stop paying markups masquerading as scheme dues. And you’ll do it with a lean team, in real time, as instant rails scale.

If you want a quick, evidence-backed read on your opportunity, send us 60 days from two PSPs. We’ll return a dispute-evidence baseline, a fee-variance map, and a rollout plan to make both wins routine on Optimus Payment Reconciliation.