Where do I start with AI?
Most companies start where it is easiest — not where it matters most.
The most common answer is: with a small internal use case. That is rarely wrong and almost never sufficient.
The asymmetry
Organisation — finance, HR, reporting, administration — is usually easiest to implement and has the least strategic effect. Production — how your service is actually made — often carries the larger productivity effect. Product — what you sell — often carries the greatest strategic effect and is examined least often.
This is not a law. But the order in which companies search is almost always the reverse of the order in which it counts.
Two beginnings at once
A pragmatic start has two movements. First, an initiative in Organisation that delivers visible value within a quarter, so the organisation experiences AI doing something real. Second, a serious and uncomfortable examination in Product: what happens to our offering if a competitor integrates AI first?
The first builds credibility. The second builds strategy. Doing only the first is the most common mistake.
The test for a first use case
Recurring, text-heavy, low-risk, reviewed by a human anyway, and measurably costly in time or money. If that holds, it is a good first case. If it needs judgement, relationships or accountability, it is not an automation case — it may be a case for support rather than replacement.