Pipeline duration is usually tracked as an infrastructure metric. It is really a behavioural one.

What slow CI actually buys you

• Bigger commits, because nobody wants to pay the wait twice.
• Longer-lived branches, and therefore harder merges.
• Approvals given without a second look, because a re-run is a coffee break.
• A quiet norm of pushing to main on Friday afternoon, since the pipeline will not finish before anyone leaves anyway.

Where the time usually goes

In the pipelines I have profiled it is rarely the tests. It is dependency resolution with a cold cache, container image builds that reinstall the world because a layer was ordered badly, and stages running in sequence that have no dependency on one another. Fixing the layer order alone has repeatedly cut builds in half.

Set a budget, put the duration on a dashboard, and treat a regression the way you would treat a latency regression in production. It is the same kind of problem.