AI and the Blended Rate: What Payment Card Optimization Can and Can't Fix

The blended rate moved 0.054% in a single month.

At the volume being processed, that translated to over a million dollars in unexplained cost increase. The job was to find out why.

Card mix shift? Checked. Gas price impact? Checked. Settlement timing? Checked. Network rule change? Checked. What the analysis found instead was gift card transaction volumes at individual sites that were mathematically impossible given their fuel throughput. The ratio simply didn't exist in legitimate retail.

That is what forensic analysis of a blended rate finds when you follow it to the answer. And that is what AI can now do automatically — at every site, every day, in real time. But only if someone is governing the AI. And right now, almost nobody is.

How Payment Card Economics Actually Work

Most executives understand that accepting a credit card costs money. Few understand how much, why it varies, and what can be done about it. That gap is where millions of dollars disappear every year — and where AI is beginning to change the equation.

Every credit card transaction involves four parties: the cardholder, the merchant, the acquiring bank, and the issuing bank. The card network — Visa, Mastercard, Amex, Discover — sits in the middle setting the rules. On a standard $100 purchase, the merchant receives somewhere between $97.50 and $98.50. The difference splits three ways: interchange goes to the issuing bank and represents the largest piece, assessment fees go to the card network, and the merchant discount goes to the acquiring bank.

Interchange is the number that matters. Everything else is relatively fixed. Interchange moves — by card type, transaction type, data quality, volume, and timing. Understanding why it moves, and by how much, is the entire game.

Why Fuel Card Economics Are Completely Different

Standard retail interchange is straightforward — a percentage of the sale. Fuel breaks that model in ways that make it one of the most complex payment card environments in the world.

The core problem is transaction size volatility. A regular retail transaction averages $45 to $65 and is relatively stable. A fuel transaction averages $40 when gas is $2.50 per gallon and $110 when gas is $5.00 per gallon. Same physical transaction — same gallon of gas — completely different dollar economics. When interchange is tied to transaction value it moves with gas prices. When it is tied to a per-transaction flat fee, it does not.

This creates the fixed versus percentage problem. Under a percentage model, 1.8% of a $40 transaction is $0.72. The same rate on a $110 transaction is $1.98. Under a fixed fee model of $0.75 per transaction regardless, the effective rate on a $40 sale is 1.875% and on a $110 sale is 0.68%. Who wins depends entirely on where gas prices go. Contracts negotiated at $2.50 per gallon look completely different at $5.00 per gallon. At billions in annual volume, fractions of a percent are millions of dollars.

The Blended Rate — The Single Most Important Number

The blended rate is your effective average interchange rate across all transactions, all card types, all networks for a given period. The formula is simple: total interchange paid divided by total transaction volume. The interpretation is anything but.

At a fuel retail site, on any given day, a merchant accepts standard consumer credit cards across all major networks, rewards cards carrying higher interchange, premium rewards products at the highest rates, PIN debit capped by the Durbin Amendment, signature debit, commercial and corporate cards with Level 2 and Level 3 data requirements, and fleet cards from WEX, Voyager, and other networks — each with its own interchange schedule, settlement timing, and data requirements.

Factor	Moves Rate Up	Moves Rate Down
Card mix	More premium rewards cards	More PIN debit
Data quality	Missing Level 2/3 fields	Complete data capture
Transaction size	Smaller (fixed fee contracts)	Larger transactions
Settlement timing	Wrong batch window	Optimized window
MCC coding	Wrong category code	Correct category
Fraud/chargebacks	Higher chargeback rate	Clean transaction file

When the blended rate moves and nobody can explain why — that is the forensic signal. Every legitimate rate movement has a cause. Unexplained movement means something is happening in the transaction mix that is not supposed to be there.

The Level 2 and Level 3 Opportunity Most Operations Miss

Commercial cards — corporate cards, purchasing cards, fleet cards — have a tiered interchange structure based on data richness. Level 1 data is basic: transaction amount, date, merchant name. Level 2 adds sales tax amount, customer reference, and merchant zip code, unlocking interchange discounts of 0.30 to 0.50%. Level 3 adds line item detail, odometer readings, vehicle and driver identification, and freight amounts — unlocking discounts of 0.50 to 1.00% compared to Level 1.

The problem is that point-of-sale terminals at fuel sites are frequently not configured to capture and transmit Level 3 data fields. Fleet card transactions default to Level 1 when they qualify for Level 3. At billions in commercial card volume, that represents a significant and entirely avoidable expense. This is a process review finding that consistently pays for itself many times over — and exactly the kind of systematic gap that AI identifies and corrects at scale.

The Durbin Amendment — What Changed Everything

Before 2011, debit interchange was effectively unregulated. The Durbin Amendment changed that permanently. For large issuers, debit interchange was capped at $0.21 plus 0.05% of the transaction. More significantly, merchants were given access to at least two unaffiliated debit networks for routing — meaning for the first time, merchants could choose which network processed their debit transactions.

At scale, this routing choice is worth real money. PIN debit through one network versus another carries different economics post-Durbin. Optimizing that routing decision across billions in debit transactions annually is one of the clearest AI optimization opportunities in payment card operations — and one of the most commonly neglected.

What AI Can Actually Optimize — and What It Cannot

The distinction matters. AI in payment card operations is not a replacement for human judgment — it is a force multiplier for human judgment. The blended rate decomposition that once took days of forensic analysis can run automatically every night. The anomaly that signals fraud at a specific site surfaces immediately rather than hiding in quarterly reconciliations. The debit routing decision that required a treasury analyst to review network schedules happens in milliseconds at the point of sale.

But the AI does not know what the regional manager promised a major fleet account in a side conversation. It does not know that a processor relationship is strategically important beyond its current economics. It does not know when the contract window opens for renegotiation in a way that changes the calculus. Those remain human decisions — and the governance framework has to make clear which decisions belong to the AI and which belong to the humans overseeing it.

The Governance Question Nobody Is Asking

Payment card AI is being implemented across the industry right now — in routing optimization, anomaly detection, fraud scoring, and settlement management. The optimization case is well understood. The governance case is not.

When AI manages debit routing decisions at billions in transaction volume — who authorized the routing logic? When the blended rate moves unexpectedly because the AI made a routing decision at 2 AM — who reviews it and how quickly? When a new fraud pattern emerges that the anomaly detection model was not trained on — who owns the gap?

These are not hypothetical questions. They are the governance questions that determine whether AI in payment card operations creates value or amplifies risk. The same forensic discipline that finds fraud in a blended rate anomaly needs to be applied to the AI systems now managing those rates automatically.

AI does not just optimize payment card economics. It makes them transparent in ways that structured opacity previously prevented. That transparency is an asset for organizations with strong governance. It is an exposure for organizations without it.

What Good Governance Looks Like

A governed AI implementation in payment card operations has clear answers to several questions. It knows which decisions the AI makes autonomously, which require human review before execution, and which require human approval regardless of what the AI recommends. It has defined thresholds — when the blended rate moves by more than X basis points the AI flags it for review rather than continuing to optimize around it. It has audit trails — every routing decision, every anomaly flag, every settlement timing adjustment is logged and reviewable. And it has a human in the loop who understands the underlying economics well enough to know when the AI is wrong.

That last requirement is the hardest one. The institutional knowledge required to govern payment card AI effectively is exactly the knowledge that decades of complexity and offshoring have made scarce. The people who understood why the blended rate moved, what the settlement file was supposed to look like, and what a gift card volume anomaly meant — those people are retiring or have moved on. Replacing their judgment with AI without replacing their governance function is not optimization. It is risk transfer.