The Green Sheet Online Edition
September 28, 2026 • 26:09:02
Four analysts, one statement, four honest answers
Hand the same merchant statement to four experienced analysts and ask each for the effective rate. You will get four numbers, each reached through a defensible calculation. None of them will be dishonest.
Consider a restaurant month: $250,000 in gross settled purchases across 5,000 sales, $6,500 in net processing fees after a $99.85 rebate, $6,000 of refunds and $1,500 of chargeback principal, and $484.85 of monthly, PCI and equipment charges that do not move with volume. Fees are withheld daily, so deposits land at $236,000.
The first analyst divides net fees by net deposits and reports 2.7542 percent. The second divides gross fees by gross volume, never touching the rebate, and reports 2.6399 percent. The third divides net fees by gross volume and reports 2.6000 percent. The fourth counts only the volume-driven fees and reports 2.4061 percent.
Nearly thirty-five basis points separate the highest from the lowest. Apply the highest and the lowest to the same $3 million of annual volume and the gap is worth roughly $10,400. The merchant paid $6,500 either way—what moved was the reported rate, and with it the size of any savings a competing proposal can claim without stating a single false number.
Which is why the analysts I trust stopped arguing about rates years ago and compare dollars and cost per transaction instead. That instinct is right, but dollars only compare when both sides counted the same charges over the same period, and that is the part nobody wrote down. Put two of these analysts on opposite sides of a deal and the comparison becomes an artifact of method rather than a fact about the merchant. Nobody lied. The number simply is not reproducible, and the merchant has no way to tell.
Why the informality held, and why it is ending
For three decades this was survivable, because statement analysis was a craft. The assumptions lived in the analyst's head, and when a merchant or a competitor pushed back, a person could say what they had counted and why. Informality held because a human always stood behind the number.
That is what is changing. Statement analysis is increasingly performed by software: parsers, spreadsheets that outlive whoever built them, tools that return a rate in seconds.
This is not a forecast. NMI acquired Fee Navigator in June 2026, citing a knowledge base built on more than 100,000 merchant statements. U.S. Bank said in May 2026 that its in-house analyzer had processed more than 30,000 statements in the preceding year. This publication covered AI statement analysis as early as October 2021. The technology is not new to anyone reading this; what is new is its share of the work.
Software does not carry an assumption in its head. It encodes one. An assumption nobody wrote down is no longer a craft judgment; it is a defect that scales silently across every merchant it touches, and cannot be cross-examined.
There is precedent for the fix, though not here. The U.K. Payment Systems Regulator directs the largest acquirers to give merchants a summary box of specified information and to run an online quotation tool. The EU Interchange Fee Regulation makes acquirers itemize the merchant service charge, interchange and scheme fees by card category and brand unless the merchant asks in writing for blended pricing.
What has never existed on the U.S. side is a processor-neutral, openly published methodology stating how a statement audit is computed. I can find no industry body that publishes one, and no commercial tool I reviewed does either.
Declare the basis before the arithmetic
The remedy is not better analysis. The arithmetic was never hard. The remedy is declaring the basis first.
Start with the denominator. Use gross settled purchase volume and its matching settled count. Do not substitute net deposits, funding totals or authorization volume: refunds, chargeback principal, reserves and loan repayments all move deposits without changing the cost being measured.
Declare the corollary too — fees arising from credits and disputes stay in the numerator even though credit volume never enters the denominator. That is a stated convention, not an oversight, and it biases the rate upward for a high-return merchant.
Then reconcile the numerator. "Total fees" hides several different totals. Classify every processing-related charge exactly once: interchange, assessments, processor markup, authorization, monthly, PCI, equipment, chargebacks and an explicit other for genuinely ambiguous costs.
Tiered and flat-rate statements need one more category, because they print a single blended discount that fuses interchange, assessments and markup and cannot be taken apart from the document; a bucket like that should say so rather than be filed under whichever component it most resembles. Gross processing fees are the exact sum of those groups. Net processing fees are gross fees minus processing-fee credits. Only then divide.
This matters more than it sounds, and it is worth checking on your own desk today. On several common statement formats the processor's discount is printed below the itemized fee total rather than inside it: the statement shows a total fees due figure, a separate discount due figure, and an amount deducted that equals the two added together. An analyst who sums the fee sections and stops at the printed total omits the markup entirely.
I ran that check across 68 real monthly statements from two unrelated processors. On one, itemized-only reporting gave 3.0701 percent against an all-in 3.8701 percent — 80 basis points, roughly a fifth of what the merchant actually paid, missing. On another, 2.9050 against 3.5050 percent. On a third format, 20 basis points. Same document, same analyst, two defensible readings, and nothing on the page announces which one you just performed.
The same corpus made three other habits measurable. Eleven statements produced an effective rate above 100 percent—dormant months where fixed fees dwarfed a few dollars of volume, the worst reading 8,037 percent—arithmetically true, analytically worthless, and belonging flagged rather than averaged into anything.
Seven carried an annual charge inside a single month, once 44 percent of that month's fees, inflating any month-over-month comparison that included it. And on a dual-priced account the headline rate read 3.6461 percent while the merchant's own cost of acceptance was 0.0328 percent, because the surcharge collected from cardholders settles inside card volume and pays most of the bill. Both numbers are defensible. They differ by a factor of 111.
Publish the rounding rule as part of the equation. Money work should not depend on floating-point behavior or whichever mode is active in someone's spreadsheet. Our model rounds the effective rate half up to six decimal places and average ticket to two; 0.022918 displays as 2.2918 percent. That feels fussy until two systems disagree at a boundary, at which point a disclosed rule turns an argument into a testable question.
Say plainly what the number cannot prove
A validator can prove that required fields exist, that fee groups reconcile, that dates are ordered and that computed fields match declared inputs. It cannot prove a human transcribed the source correctly, decide whether a fee is contractually permitted, or say whether a merchant should switch providers.
It cannot even tell you that the line labeled interchange is interchange—a padded pass-through reconciles to the cent and validates clean. Catching that takes the brand's published tables and transaction detail, which no statement-level record carries. The contract makes the arithmetic checkable, not the pricing honest.
That distinction belongs in the audit itself. "The arithmetic reconciles" is defensible. "This merchant is definitely overpaying" usually is not, unless the reviewer can name the comparison basis and the assumptions behind it.
American Express deserves its own line, because it moves the denominator without touching the numerator: under OptBlue the discount sits on the acquirer statement, while a direct agreement bills off it, leaving the volume inside the comparison and the cost outside.
Source statements also stay private. A reusable audit record needs statement-level totals only —never card numbers, security codes, PIN data, bank or routing numbers, tax identifiers or credentials. Automated screening reduces accidental disclosure; it does not replace a human reading the record before it leaves the merchant's control.
Six lines to write down before you quote a rate
None of this requires new software. It requires writing six things at the top of the analysis, before the arithmetic starts, and letting the merchant see them. State the period and the volume definition—gross settled purchase volume and its matching settled count, named as such. State the numerator: every processing charge for that period, classified once, with gross fees, credits and net fees shown separately rather than collapsed into one total.
State the rounding rule and the precision.
State what is excluded and why, naming any charge billed off this statement, a direct American Express agreement being the common one. State whether the month is comparable at all, because a dormant month, an annual charge landing in one cycle, or a surcharge settling inside volume each produce a rate that should not be ranked against anything. And state what the audit does not establish.
Six lines. A competitor who declares them can be checked. One who will not has told you something.
Doing this costs no proprietary pricing and no merchant data—only the definitions, the equations, the rounding rules and the limits, versioned so later changes are visible.
Lifted Payments published its own model on those terms under CC BY 4.0, with a spreadsheet template, JSON Schema, decimal-safe validator, methodology and a synthetic test corpus, archived at doi.org/10.5281/zenodo.21761714 and maintained at liftedpayments.com/payment-processing-statement-audit. It is not an accounting opinion, an interchange-qualification determination, or a guarantee of savings.
Take the word neutral skeptically, because I sell merchant processing. It is not the publisher that is neutral; it is the artifact. Every equation and rejection case is published and re-runnable, so anyone can prove me wrong without my cooperation.
I would rather the industry argue about the definitions than keep quietly using four of them. When a merchant can see the basis, reproduce the math and understand the limits, a statement audit survives the merchant's accountant, the incumbent's rebuttal and a second look six months later.
Daniel Wilson Kemp is founder and CEO of Lifted Holdings. Lifted Payments provides merchant processing and builds payment software, including smart-terminal applications, gateway workflows, POS integrations and custom payment services. He writes about pricing transparency and reproducible measurement in the acquiring channel. Learn more at Notice to readers: These are archived articles. Contact information, links and other details may be out of date. We regret any inconvenience.
