Built in weeks, maintained for years

An in-house reconciliation pipeline is cheap in its first year and expensive in every year after. What five years of building can cost, and when buying makes more sense.

The Actuals teamActuals · 5 min read

A payment provider updates its API, and the in-house reconciliation pipeline breaks. Two engineers spend three days debugging it while the close waits. Nobody budgeted those three days. Nobody will budget the next time either.

Most CTOs we talk to who built their own data integration layer made a sensible call at the time. Volumes were lower, the scope was small and the team had capacity. When we ask whether they would build it again today, the answer is honest: "No. But we’ve already invested heavily and switching feels risky."

That is the trap. Not the decision to build, but the cost of keeping it.

Why it starts cheap

The first version connects one PSP, such as Stripe or Adyen, to one order system and one ERP. The data model is simple, and a proof of concept runs in production within weeks. It costs €20-30k in engineering time, which looks cheap next to a platform licence of €35k a year.

But the layer is never just a pipe. Settlement timing does not match order timing. One provider settles in T+2, another in T+3. One system wants raw transactions, the next wants journal entries with the right GL codes. Someone has to connect each API, translate the formats, handle the timing gaps and find out why data went missing.

How it grows

Then the business grows, and each year adds something.

  • A second PSP with its own settlement format. Two weeks of work, and now there are two pipelines, so every matching bug gets fixed twice.
  • A new country. German VAT works differently, so the ERP needs transactions categorised in another way.
  • A marketplace channel. Those orders flow in differently and the existing logic does not apply, so a third pipeline gets built.
  • Edge cases. A refund processes before the original payment settles, and the matching flags it as an error. At 0.1% of orders, that is 1,000 orders a month for finance to check by hand.

What began as a simple data flow is now 10k lines of business logic, spread over several files and engineers. Some edge cases only the original engineer knows. Workarounds for provider quirks were never documented. Some data is transformed in three places, for three different reasons. At some point a second engineer is hired, only to keep an internal tool running.

What five years can cost

The figures below are an example, not a quote. They follow a company that grows from €25M to €80M revenue. Most companies grow slower than that, but we have seen this pattern many times, in different forms.

  • Year 1, €25M revenue. Build and maintenance take 1.5 engineers, about €130k.
  • Year 2, €40M. A second PSP integration plus one FTE on maintenance, about €145k.
  • Year 3, €65M. The integrations get fragile and a second engineer joins, about €230k.
  • Years 4 and 5, €80M and up. The pipeline is now business-critical and needs on-call cover and monitoring. Each year costs €280k or more.

Over five years, direct engineering alone comes to more than €900k. Then add what those engineers did not build, and the features that slipped. Add the trouble of hiring for "maintains internal data pipeline" too, and the real cost passes €1.3M.

A platform for the same company costs €30-84k a year in licence fees, depending on the vendor, volume and complexity. Add a one-time implementation of €28-42k and about €7k a year for maintenance. Over five years that is €215-500k. New payment providers, API changes, new countries, audit trails and monitoring are the platform’s work, not your engineers’.

Product or infrastructure?

The honest question is whether this data integration is your competitive advantage, or the infrastructure under it. If data processing is your product, as in a fintech platform, build it and own it. If you run an e-commerce, marketplace or subscription business where the data layer simply has to work, let a platform carry it. Your engineers can then spend their time on what makes your product unique.

The usual objection is "our business is too complex". Some of our clients are genuinely complex. Interrail processes two million orders a year across multiple countries with complex tax rules. Recharge operates in 160 markets with different settlement models. Last Mile Solutions handles EV charging sessions with roaming logic across networks. Their complexity sits in what flows, not in how it has to flow. What they need is a platform that can express their tax rules, settlement definitions, exception handling and routing, without engineers maintaining code.

Building still makes sense in a few cases. A regulator may require you to own your control systems, or your data may not be allowed to leave your own infrastructure. Even then, private deployments and on-premise options exist. An engineering team that enjoys maintaining the pipeline is not one of those cases.

The debt that looks like an asset

Internal systems work, often for years. The code is stable and everyone knows how to maintain it, so it feels like an asset. Then you need a new data source, settlement type or edge case, and it turns out to be debt.

Switching costs time as well. Data has to be migrated, the team retrained, the new system validated. But staying costs time too. It is only spread over five years instead of a few months, and it includes the work your engineers never got to.

If you already built it

  1. Keep maintaining it. Accept that scaling now depends on engineering capacity, that the debt keeps growing, and that knowledge leaves when a core engineer does.
  2. Migrate to a platform. That is a hard project of 3-4 months and €35-70k in engineering time. In the example above, it pays for itself in year 1-2.
  3. Go hybrid. Keep your own system for logic that is truly unique, and use a platform for the common work: PSP matching, settlement handling and GL posting. That works when about 20% of the logic is unique and 80% is common.

Most companies wait until the system is fragile and moving feels impossible. The better moment is while it is still stable, which in the example falls between €25M and €80M revenue.

So sit down with your CTO and ask one question. In five years, do you want two engineers to be specialists in an internal pipeline, or building what sets your business apart? If the answer is the second, the decision is no longer build or buy. It is when to move.

In practiceAdyen, Stripe, PayPal. Three report formats, one truth?

Written by

The Actuals teamActuals