Unblending the Bill: Separating Scheme Rules from Processor Markups to Find the Real Cost of Payments

Why this matters: your “blended” MDR is a mirage. Interchange tiers, network dues/assessments, cross-border/FX add-ons, gateway markups, and rail-specific fees all move underneath it—and a few basis points of drift at scale is real money. In 2023 alone, U.S. merchants paid $172.05B in processing fees; recovering just 1 bp on a nine-figure volume business is a material, recurring gain.

This piece lays out a CFO-grade, AI-powered payment reconciliation approach to unblend every settlement line, recompute what you should have paid under scheme rules, and isolate what’s truly processor markup versus scheme-mandated—so you can recover leakage, reprice corridors, and negotiate from evidence.

The blended-rate problem (and why spreadsheets lose)

Blended pricing promised simplicity; it delivered opacity. Interchange itself varies by card brand/product/MCC/auth method and often drives 70–90% of total cost—so small classification errors snowball when you’re at 50–100M+ txns/month. Network dues and assessments look tiny in isolation (~0.13–0.15% domestically) but compound at scale, and cross-border fees can add 2–3% on international flows. Without line-level recalculation, “mystery bps” hide in the bundle.

The macro trend makes inaction expensive. Real-time rails are accelerating settlement and compressing your window to catch errors: The RTP network processed $481B in Q2-2025, while the FedNow Service surpassed 1,400 participating institutions in mid-2025. In India, UPI crossed 20.0B monthly transactions in August 2025—a preview of the data variety and velocity every global merchant will face.

The operating model: contracts → code → recompute → recover

On Optimus Payment Reconciliation we treat “cost of payments” as a deterministic math problem with statistical guardrails.

1) Parse contracts into a versioned rate engine

Your processor/PSP agreements and the public network schedules (e.g., Visa U.S. Interchange Reimbursement Fees) become machine-readable rules: brand, product, MCC, card-present/-not-present, authentication method, geographic pair, cross-border flags, FX handling, and tier thresholds—time-boxed so you can prove which rules applied on a given day.

2) Recompute every settlement line

For each settlement leg, Optimus deterministically computes expected interchange + dues/assessments + contracted processor fee, then compares it to the invoiced amount. The output is a variance vector (bps & currency) with a cause code: tier mis-application, assessment uplift, cross-border misflag, FX spread, gateway markup, rounding, or “contract update not mirrored.”

Example: assessment charged at 0.15% where domestic benchmarks suggest ~0.13–0.15%; paired with cross-border flags that should not apply for the issuer geography—flag as likely markup + misclassification, with links to the rate engine and supporting schedules

3) Detect drift with anomaly models

Line-by-line math finds obvious misses; AI catches the subtle ones. We baseline each corridor/rail for seasonality, ticket size, and mix, then alert only when variance is statistically meaningful and persistent (e.g., ≥0.8 bps for 5+ days on EEA cross-border premium). That keeps the fee desk focused on value, not noise.

4) Close the loop with evidence packs

For each variance cluster, Optimus generates an auditor-ready dossier: affected IDs, expected vs. invoiced math, the rate-engine clause/version, and a reconciliation timeline. Finance routes it via maker-checker to PSPs/acquirers; outcomes retrain thresholds and lock in recovered savings—all within the reconciliation workbench.

What you uncover when you unblend

• Markups vs. scheme dues. Network dues/assessments are standardized at the scheme level, yet we routinely see “brand fees” uplifted inside bundles. Unblending exposes whether you’re paying the scheme rate or a processor markup on top.
  
  
• Tier boundary slips. Volume shifts across programs can place transactions in a higher tier than contracted; the engine computes the counterfactual at the correct tier and quantifies the delta against the invoice.
  
  
• Cross-border misflags & FX spread creep. Misclassified issuer geography or program codes, plus opaque FX spreads, create recurring bps leakage that only shows up in line-level recompute.
  
  
• Program/rule updates not mirrored. Networks revise schedules; if your processor lags in implementation, you pay yesterday’s math on today’s traffic. The published Visa schedules make these changes explicit.
  

The CFO scorecard (and why 0.5–2.5 bps matters)

We recommend measuring four things weekly:

1. Recovered Basis Points (RBP) by connector/region/brand—your headline P&L win.

2. Fee Variance Detected (FVD) per 10k settlements—your pattern-finding velocity.

3. Time-to-Evidence Pack & win rate on fee disputes—your operational muscle.

4. Approval-adjusted unit cost (true cost per successful txn), not headline MDR.

Context matters: with $172.05B in U.S. processing fees (2023), reclaiming even 0.5–2.5 bps across selective corridors is six- to seven-figure annual savings for a large merchant, before you count second-order effects like better routing and corridor repricing. Household-level estimates peg the macro burden of swipe fees at ~$1,100 per U.S. family—an external signal of the system’s scale and why line-level governance pays for itself.

Architecture that makes finance and engineering both happy

• Canonical data model. Normalize payments, settlements, fees, refunds/chargebacks, payouts, and bank deposits with durable keys (PSP Txn ID/ARN/UTR + internal GUID). This is how we join cleanly before math. See Payment Reconciliation.
  
  
• Stream first, store second. Webhooks/SFTP/bank APIs feed event streams; we maintain raw + curated zones for replay and audit. Real-time rails (RTP, FedNow) demand low-latency controls—not month-end autopsies.
  
  
• Deterministic + probabilistic matching. Tier-1 exact keys; Tier-2 fuzzy (amount±fx, time windows, last-4, merchant ref) with explainable confidence; Tier-3 enrichment for late/partial files—so fee recompute runs on high-fidelity linkages.
  
  
• Controls & audit. Maker-checker for postings, immutable logs, and downloadable close packs—because speed without governance isn’t helpful to a CFO.
  
  

Implementation: 60 days to a defendable “true cost” number

Days 0–15 — Baseline. Connect two PSPs + one bank; ingest 60 days; parse contracts into the rate engine; publish a variance heatmap by brand/region/MCC.
Days 16–30 — Automate. Turn on line-level recompute; enable anomaly alerts; route fee variances to the workbench with maker-checker; start producing evidence packs.
Days 31–60 — Recover & reprice. Open recovery cases on persistent clusters; publish an RBP leaderboard by connector; align routing policy and corridor pricing to your approval-adjusted unit cost. (All inside Payment Reconciliation.)

Conclusion

The goal isn’t to wage war on fees—it’s to separate the non-negotiable (scheme rules) from the negotiable (processor economics) with math, telemetry, and proof. Once you can explain each basis point, you can decide—today—not six months from now—what to fix, what to route differently, and where to renegotiate.

If you want a quick, evidence-backed read on your true cost of payments, send us the last 60 days of settlements from two PSPs. We’ll return a variance map, an RBP forecast, and a remediation plan inside Optimus Payment Reconciliation—so “blended” stops being a black box, and starts being your lever.