The AI Opportunity Assessment: A CEO's 20-Minute Test Before You Spend a Dollar

Most AI budgets are spent backwards. A team gets excited about a tool, finds a use for it, and only discovers the data problem or the integration problem after the money is committed. An AI opportunity assessment flips the order: decide whether a project is worth doing — and whether it's even possible — before anyone writes code.

You don't need a six-week consulting engagement to do the first pass. Here's the same framework we use, compressed into something a CEO and one operator can run in about twenty minutes.

1. Size the prize from your own baseline

Forget borrowed industry statistics. Start from a number you actually own. Pick one painful, repetitive, high-volume process and write down its baseline: how many units a month, how long each takes, what an error costs. If AI removed even part of that time or error rate, what's the value at stake — as a range, not a fantasy? If you can't estimate a baseline, that's your first finding: you're flying blind, and measurement comes before automation.

2. Score feasibility — the weakest link caps the project

A use-case is only as strong as its weakest link. Score five quickly:

• Data: does the data needed exist, and is it reachable (API, warehouse, database, clean export) — or trapped?
• Signal: is there enough pattern in the data for a model to do something useful?
• Error tolerance: what happens when it's wrong, and can a human catch it in time?
• Integration & write-back: can the result flow back into the system where the work happens?
• Adoption: will the people who'd use it actually change how they work?

A brilliant use-case sitting on unreachable data is not a strong project. It's a data-access project wearing an AI costume.

3. Decide build vs. buy vs. foundation model

Not everything should be custom-built. For commodity needs, an off-the-shelf product is faster and cheaper. For anything that's a genuine differentiator, or that has to live deep inside your own systems and data, custom usually wins. And many problems are solved best by a foundation model with the right prompting and your data retrieved alongside it — no training required. Pick the cheapest path that actually wins.

4. Prioritise to a wedge

Resist the urge to "do AI everywhere." Choose one smallest valuable slice — the wedge — that you can ship, measure and point to. A single proven win buys you the credibility and the data to scale. A sprawling everything-at-once programme buys you a steering committee.

5. Name the risks, the metric, and the owner

Before you greenlight, write down the top two or three ways this could fail and how you'd manage each; the metric you'll judge it by (baseline → target → guardrail); and the single human who owns the outcome. No sponsor, no adoption — it's that blunt.

The honest part most assessments skip

The most valuable output of a good assessment is sometimes "not yet" or "this isn't an AI problem." If the fix is really a rules change, a process redesign or a reporting clean-up, AI will only make it slower and more expensive. Saying so before the budget is spent is the cheapest win available to any leadership team.

If you'd like this done in minutes rather than a workshop, that's exactly what we built AI Blueprint for. It walks the same framework — sizing the prize, scoring feasibility, checking whether your data is reachable, choosing build-vs-buy — and hands you a build-ready plan you can act on, including an honest verdict when the answer is "not yet."