> ## 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.

# A Core Migration Is Quoted as a Build. It Is Paid for as a Proof.
- URL: https://thepaymentsoperator.com/a-core-migration-is-quoted-as-a-build-it-is-paid-for-as-a-proof/
- Published: 2026-02-09T03:20:00.000Z
- Updated: 2026-11-18T11:18:20.000Z
- Description: The budget estimates construction. The money goes on discovery, duplication and evidence. In a well-run migration, proving correctness costs more than achieving it.
- Author: Armando

*The budget for replacing a payments core is an estimate of construction. The money is spent on something else entirely: demonstrating, transaction by transaction, that the new system behaves exactly like the old one, including in the ways nobody can explain.*

Every few years an institution concludes that its payments core must be replaced. The reasons are usually good and usually one of four: the licence economics have stopped making sense, the vendor has put the version on a path to end of support, the platform cannot carry the volume or the products the business now wants, or a corporate event — a separation, an acquisition — has made the current arrangement untenable.

The proposal that follows is a construction estimate. Licences, implementation, integration, infrastructure, a contingency, a date.

I have never seen one that was wrong about the construction. I have rarely seen one that was right about the total, and the gap is always in the same place.

A core migration is not a build. It is a proof. The deliverable is not a working system; a working system is table stakes. The deliverable is *evidence, acceptable to several audiences, that the new system does what the old one did* — and the old one does a great many things that nobody has written down.

## What is actually being replaced

Here is the thing that makes payments cores different from other systems, and it is worth stating plainly because it is the root of everything else.

A mature payments platform's behaviour does not live in its code. It lives in its parameters.

Twenty years of product decisions, exception handling, client-specific arrangements, regulatory accommodations and one-off fixes are expressed as configuration: fee tables, limit structures, routing rules, posting rules, retry behaviours, dunning schedules, reason-code mappings, hierarchy definitions, calendar settings, a hundred flags whose names made sense to somebody in 2011.

That parameter set is the institution's actual business logic, and three things are true about it at once. It is enormous. It is largely undocumented. And it is correct — in the specific sense that it produces the behaviour the business currently depends on, including the behaviour nobody intended.

So when a migration begins, the real scope is not "implement the new platform." It is "discover what the current platform does, in sufficient detail to reproduce it, from a system that cannot be interviewed."

The only complete documentation of the old system is its production behaviour. Which means the migration's central activity is archaeology, and archaeology does not fit neatly into a Gantt chart because you cannot estimate what you have not yet found.

## The five costs nobody quotes

**The dual run.** You cannot move a card portfolio in one night. Any responsible plan moves it in waves, which means that for a long period — months, often more than a year — two platforms are in production at once.

Two platforms in production is not twice the work. It is more, because now a third thing exists: the daily obligation to prove the two agree. Balances on both. Reports on both. A reconciliation between them whose only purpose is to demonstrate that the migration has not broken anything, and which disappears the day it is finished.

That work is real, continuous, skilled and temporary. It needs people. It is almost never in the budget, because it produces no deliverable anyone wants to own.

**The fields that lie.** Every legacy platform has fields whose contents do not match their names, because somebody in a hurry repurposed them. A spare indicator that now means something about a product. A free-text field carrying a code that a downstream report depends on. A date field used to store something that is not a date.

Migration surfaces every one of these, and each one is a meeting: what is this, who uses it, what happens if we drop it, who will notice. Multiply by several hundred. Most of those meetings end with the conclusion that nobody knows and nobody dares remove it, so it is migrated faithfully, including the ambiguity.

**Recertification.** Every external connection that was certified against the old platform gets certified again. Networks, schemes, acquirers, clearing houses, regulators, the parties on the other end of every interface.

Those calendars are not yours. A certification slot that moves by six weeks moves your go-live by six weeks, and there is no amount of internal effort that compresses it. This is the most predictable source of delay in any migration and it is routinely modelled as if it were under the project's control.

**The people who know.** In any mature institution, the complete picture of how the core actually behaves exists in a small number of heads. Not in a document, not in a wiki — in people, usually long-serving, often not senior, frequently the ones who are quietly closest to retirement.

Those people are also the only ones who can run the current platform while the migration happens. Which creates the migration's cruellest constraint: the scarcest resource is simultaneously the one keeping production alive and the one the project cannot proceed without, and no org chart solves it.

**The business did not stop.** This is the one that makes everything else harder. Through all of it, the institution is still launching products, still meeting scheme mandates, still responding to regulators, still fixing what breaks.

So every change made to the old platform during the migration has to be made twice, and the migration's target is moving. A mandate that lands in month nine is implemented in both systems, tested in both, certified in both.

The honest version of the plan is not "migrate the core." It is "migrate the core while doing everything else, with the same people, for eighteen months." Said that way, the real question stops being technical. It becomes: **what are we going to stop doing?**

If a migration plan does not contain an explicit list of things the institution will not do during the period, that list exists anyway. It is simply being written by exhaustion rather than by management.

## Proof, and who is asking for it

The word I keep coming back to is proof, because the hardest part of the last six months of any migration is not making the system work. It is convincing several different audiences, each with different standards of evidence, that it does.

Operations wants to see a parallel period with no unexplained differences. Finance wants to see that balances tie. Risk wants a documented control. Internal audit wants a trail. The regulator wants to understand what changed and when. The client, if you are a processor, wants a guarantee that its customers will notice nothing.

Each of those is a separate evidentiary standard, and none accepts the others' work as a substitute. The volume of material required to satisfy them all is enormous, it is produced by the same people who are running the migration, and it is almost never estimated as its own workstream.

I would go as far as this: in a well-run migration, the cost of proving correctness exceeds the cost of achieving it. If your plan does not reflect that, your plan is a construction estimate for a project that is mostly evidence.

## What makes one succeed

I do not have a method. I have four things that distinguish the migrations I have watched go well from the ones that did not.

**A named list of what stops.** Written before the start, signed by somebody who can enforce it, revisited quarterly. The list is a commercial decision, not a project artefact, and it should be made by the people who will be disappointed by it.

**Comparison automated from day one.** Not at the end. From the beginning, running both systems against the same inputs and diffing the outputs, every day, with the differences triaged by someone whose job that is. Migrations that build this early find their problems in months with slack. Migrations that build it late find the same problems in the month before go-live.

**Scope defined by behaviour, not by feature.** "Implement instalments" is a feature. "Produce the same instalment schedule as the current system for these ten thousand historical cases" is a behaviour, and only the second one is testable against the thing you are replacing.

**An explicit decision about the parameters you will not carry.** Some of that twenty-year configuration is dead. Migrating all of it faithfully is the safe choice and it imports the whole archaeology into the new platform, where it will be somebody else's problem in fifteen years. Deciding what to drop is dangerous and valuable, it requires someone senior to accept the risk, and it is the only opportunity the institution will get for a generation.

Most migrations choose faithfulness, for understandable reasons. It is worth knowing that you are choosing it.

## From a Panamá desk

Two things look different from a smaller market, and both cut the same way.

**There is no slack.** A large institution absorbs a migration by adding people. In a regional processor, the people who would run the migration are the people running production, and there is no bench. Which means the sequencing question — what stops — is not a planning nicety here. It is the entire plan. An institution our size that tries to migrate without stopping anything is not running a risky project; it is running an impossible one, slowly, until something gives.

**And the obligations do not shrink.** This is the structural fact about operating in a small market that I keep returning to. A scheme mandate has one deadline for everybody. A regulatory requirement has one date. The work those generate is roughly fixed, regardless of portfolio size, so it consumes a far larger share of a small institution's capacity than a large one's.

During a migration, that fixed load lands on a team already committed twice over. It is the single most common reason regional migrations slip, and it is almost never named as a risk, because naming it requires admitting that capacity, not competence, is the constraint.

The practical counsel I would give any board in this region considering one: do not ask how long the migration will take. Ask what the institution will not do for that period, and whether you are prepared to tell the clients who were expecting those things. If the answer is that nothing will be deprioritised, the estimate you have been given is not a plan. It is a hope with a date on it.

## My judgment

The recurring error is category, not estimation. Institutions budget a core migration as a construction project because that is what the proposal looks like, and then spend the money on discovery, duplication and evidence — three activities that are poorly estimated precisely because they produce nothing a steering committee can point at.

If I could change one thing in how these are proposed, it would be to split the budget into two numbers, openly: what it costs to build the thing, and what it costs to prove the thing. Put the second number next to the first and the conversation changes, because everybody can see that the second is larger and nobody has to discover it in month fourteen.

The second belief, and I hold it firmly: the real asset at risk in a migration is not the platform. It is the institutional memory encoded in a parameter set that nobody can read. A migration is the only moment an institution is forced to confront how much of its own operation it does not understand, and the organisations that treat that discovery as valuable — that write it down this time — come out of it stronger regardless of the platform they land on.

The ones that treat it as an obstacle migrate the ignorance along with the data, and buy themselves the identical project in fifteen years.

## Three questions for the people who own this

**1.** Does your migration plan contain a named, signed list of what the institution will *not* do during the period? If not, that list is being written anyway, by whoever burns out first.

**2.** Are the two numbers separated — what it costs to build, and what it costs to prove? If there is one number, the second one is hiding inside the contingency and the contingency is too small.

**3.** How many people could reconstruct your current platform's behaviour from scratch? Name them. Now ask what the plan is if two of them leave in the same quarter, because that is a normal thing that happens.

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 vendor, no platform, no document and no figures. It is about a pattern, and everything in it comes from my own experience of running technology through platform change and from conversations with people who have run others.

As with everything here, it is written as a pattern and never as a case. No institution, client, vendor, platform, contract term, timeline or figure from anywhere I have worked appears in it, and none will. The five costs and the four success factors are my own organising scheme, not an industry framework — somebody who has run more of these than I have would order them differently, and I would genuinely like to be told how. Nothing here reflects the position, portfolio, clients or vendors of my employer.