> ## Content Index
> Fetch the complete content index at: https://thepaymentsoperator.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Everything in Payments Is Reconciliation. Nobody Has Ever Been Promoted for It.
- URL: https://thepaymentsoperator.com/everything-in-payments-is-reconciliation-nobody-has-ever-been-promoted-for-it/
- Published: 2025-04-14T03:21:00.000Z
- Updated: 2026-11-18T11:24:42.000Z
- Description: A transaction is not an object — it is the same claim written six times by six parties. The best measure of a payments operation is the age of its oldest unexplained break.
- Author: Armando

*A payment is not an object. It is a claim, made six times, by six parties, on six clocks — and the entire industry is the agreement about when those six claims are the same claim.*

I have spent most of my working life around payment systems, and the longer I do it the more convinced I become that there is only one problem in this industry wearing a hundred costumes.

Every incident is two records disagreeing. Every dispute is two records disagreeing. Every regulatory finding, every late close, every angry merchant, every "the money left my account and the shop says it did not arrive" — two records disagreeing. The technology is different each time. The problem is the same problem.

It is reconciliation, and it is the least glamorous word in the business.

## A transaction is not a thing

The mental model most people carry — including, in my experience, most executives who sponsor payments projects — is that a transaction is an object. It happens, it has properties, it moves through systems, and at the end it is somewhere.

That is not what a transaction is. A transaction is a *claim about an event*, written down independently by several parties who were not in the room together.

Follow one card purchase and count the records it generates.

There is the authorization: a message that says a reservation was made. There is the clearing record, which says what was actually charged, and which may differ from the authorization in amount, in timing, and occasionally in whether it exists at all. There is the settlement file, which says what money moved between institutions. There is the issuer's ledger entry, which says what the cardholder owes. There is the line on the cardholder's statement, which is a human-readable approximation of that entry and is frequently the only version the customer ever sees. And there is the merchant's own record, plus the credit to the merchant's bank account, which arrives net of things the merchant may or may not understand.

Six or seven representations of one event. Produced by different systems, owned by different institutions, generated on different clocks, in different formats, with different notions of what counts as the same.

None of them is the transaction. The transaction is the *agreement* among them, and reconciliation is the work of establishing that agreement.

Which is why I think the word undersells itself. Reconciliation is not an accounting chore performed after the real work. It is the product. Everything else is the means by which claims are generated; reconciliation is where they become a fact.

## The four families of break

After enough years you stop seeing individual breaks and start seeing families. There are four, and almost everything belongs to one of them.

**Timing.** Two systems agree about what happened and disagree about when. A cut-off at midnight in one timezone and at midnight in another; a batch that closed at 23:58 and a transaction stamped 00:01; a weekend; a leap second; a daylight-saving change that one system observes and another does not. Nothing is wrong except the boundary, and the boundary is where the report is drawn.

Timing breaks are the most common and the least dangerous, because they resolve themselves by the next cycle. They are also the reason so many reconciliation processes are quietly tolerant of differences that persist for less than a day — which is sensible, and is also exactly the tolerance a real problem hides inside.

**Representation.** The same value, written differently. A currency with a different number of minor units. A rounding rule applied in one place and not another. A field truncated to a length that was generous in 1994\. An amount expressed as cents in one file and as a decimal in another, and a conversion that is correct for every value except the ones at the edges.

These are the breaks that are invisible until they are enormous, because they scale with volume and they are deterministic. A rounding error does not happen occasionally. It happens every time, quietly, in the same direction.

**Lifecycle.** The event is not one event. An authorization is followed by a partial capture, then a second capture, then a refund, then a chargeback, then a representment, then a reversal of the reversal. Each one is a separate record referring to the first. The break occurs when a later record cannot find its original — because the reference was not carried, or was carried in a field that two systems populate differently, or because the original was archived by a retention policy written by people who had never been asked this question.

Lifecycle breaks are where real money is lost, because an orphaned refund is a payment made to somebody for nothing, and nothing in the system will tell you about it.

**Silence.** The hardest family, because it has no evidence. A file that did not arrive. A batch that was empty when it should not have been. A connection that returned success and delivered nothing. An interface that has been failing for six weeks and whose failure is logged somewhere nobody reads.

A break that appears is a problem. A break that does not appear is a loss, and the difference between an organisation that finds these and one that does not is whether anybody is checking for the absence of things. Almost nobody is, because monitoring is built to watch what happens, not what fails to.

## The part that is about people

Here is the uncomfortable half.

Reconciliation is detected by systems and *resolved* by people. The matching engine finds the exception; a human being decides what it means. And that human being, in most institutions, is in the least visible part of the organisation, on the lowest salary band of anyone who touches money, with the least career mobility and the oldest tooling.

That is not an accident. It is a consequence of how the work is perceived. Reconciliation produces no feature, closes no deal and appears in no release note. Its best possible outcome is that nothing happened. There is no way to be visibly excellent at it, and visibility is how careers are made.

So the function is chronically under-invested, chronically under-trained, and chronically staffed by people who are extremely good at a job nobody has asked them to explain. And it is precisely where the losses are, which is the whole irony: the institution puts its thinnest resources on the control that catches everything else.

I would go further. If you want to know how well a payments operation is actually run, do not look at the architecture diagrams or the uptime figures. Ask to meet the people who clear exceptions, and ask how long the oldest open item has been open.

That number is the single best indicator of operational health I know of, and it is almost never on a dashboard.

## Every new product adds a reconciliation

The operational rule I would write on the wall, if I could write one thing on a wall:

**Every new product, rail, partner or currency adds a reconciliation. If the business case does not contain it, the business case is incomplete.**

This sounds obvious and is routinely ignored, because the reconciliation is invisible at the moment of the decision. A new partner is presented as a revenue line and an integration. Nobody says that it also creates a new pair of records that must be made to agree, every day, forever, by somebody who has not been hired.

The cost is small per product and it is permanent and it compounds. An institution with fourteen partners has fourteen daily agreements to maintain, and the fourteenth was approved on the assumption that the team absorbing it had slack, which it did not, because the same assumption was made about the thirteenth.

This is how operations teams end up in a state that everyone describes as a staffing problem and that is actually an accumulation of undeclared obligations.

## And now there are new rails

Which brings me to the thing I keep thinking about as the settlement layer starts to change.

Every proposal to move value over a new rail — a tokenised dollar, an instant transfer scheme, a direct account-to-account arrangement — is presented as removing an intermediary. In reconciliation terms it does the opposite. It adds a representation.

A settlement that happens on a chain produces a record with its own notion of finality, its own timestamp, its own identifier format, and its own relationship to a bank ledger entry that was written by a completely different kind of system. Now you have both, and somebody has to make them agree, daily, including on the days when one of them was running and the other was not.

I am not arguing against the new rails. I think several of them are genuinely better. I am saying that the honest version of the pitch includes a sentence nobody puts in the deck: *this creates a new break class, and you do not currently have anyone who knows what a normal break looks like on it.*

That sentence is worth more than most of the technical evaluation, and I have never once seen it in a vendor proposal.

## From a Mexico City desk

Two observations specific to operating here, and the first one costs nothing to fix.

**The calendar is a reconciliation problem.** Operating across the Mexican and United States calendars means two sets of banking holidays that do not align, in a relationship where a very large share of the volume crosses that border. Every mismatched holiday is a day when one side's cycle ran and the other's did not, and every one of those days produces a population of breaks that are not errors.

The institutions that handle this well do one simple thing: they publish, in advance, a combined calendar that marks which cycles run on which days in both countries, and they tell the exceptions team what a normal break looks like on each of those days. The institutions that handle it badly rediscover it every year, usually in November, usually at the worst moment of the quarter.

It is a spreadsheet. It takes an afternoon. I have rarely seen it exist.

**And the multi-currency edge.** A peso-denominated book settling against dollar-denominated obligations produces representation breaks continuously — rounding, minor units, the rate applied at the moment of conversion versus the rate applied at the moment of posting. These are the breaks that are small, deterministic, and always in the same direction, which is the definition of a leak rather than an error.

If nobody at your institution has measured the cumulative annual value of rounding and conversion differences, that is not an oversight. It is an unexamined line item, and in my experience unexamined line items in payments are never zero.

## My judgment

Three things I believe about this, in descending order of confidence.

First, with complete confidence: reconciliation is the discipline of the industry, and the fact that it is treated as back-office administration rather than as the central control is the largest single misallocation of attention in payments. Every architectural decision should be evaluated partly on what it does to the agreement between records, and almost none are.

Second, firmly: the best available measure of an operation's health is the age of its oldest unexplained break. Not uptime, not throughput, not approval rate. Uptime measures whether the machine ran. The oldest open item measures whether anybody understands what it did. An institution with 99.99% availability and an unexplained difference that has been open for seven months does not have a well-run operation. It has a well-run machine and a blind spot.

Third, and more tentatively: the coming decade of new rails will be won operationally rather than technically, and the institutions that win will be the ones that treated reconciliation as a design input rather than an afterthought. Every new settlement asset is a new representation of an old event. The organisations that already know how to make representations agree will absorb them cheaply. The ones that do not will discover their exposure the way we always have — slowly, in a month-end close, in a number nobody can explain.

And the title is not a joke. Nobody has ever been promoted for reconciliation, and that is a defect in how we value work rather than a judgment about its importance. If I ran a larger organisation I would put the best operator I had on exceptions for a year and then promote them loudly, specifically so that everyone could see what the institution actually considers difficult.

## Three questions for the people who own this

**1.** How old is your oldest unexplained break? If nobody can answer in a minute, the answer is older than anyone would like.

**2.** For the last three products or partners your institution added, was a reconciliation obligation named in the business case — with a person assigned to it? If not, those obligations are being absorbed by somebody who did not agree to absorb them, and you will meet that fact as attrition.

**3.** What in your monitoring would detect a file that never arrived, as opposed to a file that arrived wrong? Those are different checks and most operations only have the second one.

Reply to this email with a line or two. The sharpest answers will shape the next issue.

## Sources

**None.** This essay cites no news event, no document and no figures. It is about a structural pattern, and everything in it is drawn from my own experience in payments operations and from conversations with people who run others.

As with everything I publish here, it is written as a pattern and never as a case. There are no institutions, clients, vendors, platforms, contract terms or figures from anywhere I have worked, and there will not be. The taxonomy of break families is my own organising scheme rather than an industry standard — other operators would draw the lines differently, and I would be interested to see how. Where I say "in most institutions" or "in my experience," those are generalisations offered so you can measure them against your own, not findings. Nothing here reflects the position, portfolio, clients or vendors of my employer.