What is an API integration?

An API integration (Dutch: API-koppeling) is a programmatic connection that lets two systems exchange data automatically, so information entered once appears wherever it is needed without anyone retyping it.

An API itself is just an agreed way for one program to ask another for something. The integration is what you build on top: the mapping, the timing, the error handling and the decisions about what happens when the other side behaves unexpectedly — which is most of the work and almost none of the sales conversation.

What an API integration actually consists of

The connection is the small part. What takes the time: field mapping (their customer number is your debtor code, their date format is not yours), authentication and how tokens are refreshed before they expire, triggering — does data move on a schedule, on an event, or when someone presses a button — and error handling, which is where integrations are actually judged. A connection that works when everything is fine is a demo.

Idempotency: the failure nobody quotes for

Systems retry. A timeout, a queue redelivery, a double click — and the same instruction arrives twice. If processing it twice does not produce the same result as processing it once, that retry becomes a duplicate order, a duplicate payment, a twice-applied credit.

The fix is a stable identifier per logical operation, stored by the receiver and checked before execution. It is not difficult; it has to be designed in. Ask any supplier: how do you detect a duplicate, and on which key? An answer involving timestamps or comparing content is not an answer.

Rate limits, backoff and the Monday morning problem

Most APIs cap how often you may call them. Integrations built and tested on quiet data hit that ceiling the first time a real backlog runs — classically a Monday morning, when the weekend queue flushes at once.

What to specify up front: the documented limit, what the API returns when you exceed it, and whether your integration backs off and retries or simply drops the work. Exponential backoff with a cap is the standard answer, and a queue in front of the call is the difference between a slow Monday and a lost one.

The upstream change that breaks it silently

The most common cause of a mature integration failing is not your code. It is the other system adding, renaming or repurposing a field — often in a minor release nobody told you about.

Silent failures are worse than loud ones. An integration that errors is noticed within hours; one that quietly maps an empty value into a required field can run for weeks. Two defences pay for themselves: validate what you receive against expected shape rather than trusting it, and alert on volume anomalies — zero orders on a Tuesday is a signal even when nothing threw an error.

What an API integration costs

A single straightforward integration between two systems with documented APIs is typically a few days to a couple of weeks of work. What multiplies it: undocumented or legacy endpoints, bidirectional sync (which needs conflict rules), and anything touching money or stock, where the error handling matters more than the happy path.

At Crux Digits integration work sits inside the standard route: a €2,500 audit to map the endpoints, the volumes and the failure cases, a €20,000 proof of concept where the flow is genuinely uncertain, and production from €50,000. Many integrations are smaller than that and the audit will say so.