The AI opportunity map — finding where it actually pays off

The most expensive AI mistakes I see aren’t technical failures. They’re well-executed projects aimed at the wrong target — real effort spent automating something that didn’t cost much, or chasing a flashy capability no customer was asking for.

Deciding where to apply AI is the highest-leverage decision you’ll make, and it deserves more rigor than a brainstorm. Here’s a way to think about it.

Two questions, honestly answered

For any candidate, ask:

If this worked perfectly, what would it be worth? In money, time, risk reduced, or revenue enabled — quantified, not hand-waved.
What would it take to make it dependable? The full cost: data, evaluation, the last mile of delivery, change management, and ongoing upkeep.

Most ideas collapse under the first question. A capability can be genuinely impressive and still not move a number anyone cares about. The ones that survive both questions are your real portfolio.

Where AI tends to earn its keep

Patterns that repay the effort:

High-volume judgment that’s currently a bottleneck. Work where humans are the constraint and “good enough, instantly” beats “perfect, eventually.”
Unstructured data you already own but can’t use. Documents, transcripts, tickets, images — value that’s trapped because it was too costly to read.
Personalization or synthesis at a scale people can’t match. Not replacing judgment, but extending it across far more cases than a person could handle.

Patterns that usually disappoint:

Automating something that was already cheap and fast.
Capabilities with no owner accountable for the business outcome.
Anything where being mostly right is worse than not doing it at all — unless you’ve designed carefully for the uncertain case.

Sequence for momentum, not just value

The highest-value opportunity is often not the right first one. Early wins should also be legible — easy to measure, easy to trust, easy to point to when you ask for the next round of investment. A smaller win that builds organizational confidence can unlock a larger one that would have been politically impossible to fund cold.

So the map has two axes: value if it works, and readiness to deliver. Start where those overlap. Prove the pattern. Then reach for the harder, more valuable work with a track record behind you.

The output

Done well, this produces a short, defensible list: a handful of opportunities, each with a rough value, an honest cost, and a reason it’s sequenced where it is. That artifact is worth more than any individual model. It’s the difference between an AI strategy and a pile of AI experiments — and it’s usually a few focused weeks of work to build.