Business service vs application service in CSDM — when to use which

Most explanations of this give you two definitions and leave you exactly where you started. Here is the test that actually settles it in a workshop.

Ask who would complain if it stopped, and in what words.

If the answer is “payroll didn’t run” — someone describing an outcome the business cares about, in business language — you are looking at a business service. It is the thing being consumed. It has an owner who would be asked to explain the outage to people who do not know what a server is.

If the answer is “the API tier is down” — someone describing a deployment, in operational language — you are looking at an application service. It is a running thing with a topology underneath it, and it can be watched.

One business service is usually supported by several application services. The payroll business service might sit on top of the payroll application service in production, which in turn depends on a database, a web tier and an integration. Users consume the first and operations monitor the second.

The mistake that costs the most

Modelling every application service as a business service, because the sponsor wanted a long list of services on a slide.

You end up with hundreds of “business services” that no business person recognises, nobody owns, and nothing is ever mapped to. Impact analysis then reports that an outage affects forty business services, which is technically true and tells the bridge call nothing. The list looks like progress and behaves like noise.

The opposite mistake is rarer but worse: one business service for the whole estate, supported by everything, which can never be used to decide anything.

The part the definitions skip

Neither placement is discovered. Discovery can tell you a process listens on a port. It cannot tell you that the thing it found is what the business means when it says “payroll”. Somebody decided that, in a workshop, probably once, and the decision is now sitting in the same table as the discovered facts with nothing marking it as a judgment.

That is why this argument recurs every eighteen months: the model records the answer but not the reasoning, so when someone new asks why an application was placed where it was, nobody can say, and the only way to settle it is to have the workshop again.

The practical defence is cheap. Write down the test you used — one sentence, in your own words, kept with the model — and record who decided. Then the next person inherits a rule rather than a result.

The longer argument, the CSDM documentation, including what evidence can and cannot settle in a service model.