AI Use-Case Evaluation Framework

Not every process needs AI. Use this framework to decide whether an AI-based solution suits a specific business problem, before committing to one.

Work through the steps in order. The first two are about the problem; only the third is about the technology.

Step 1 — Define the problem

Record, in plain language:

• The problem itself
• The current process
• Who is affected

If the problem cannot be stated without naming a technology, it has not been defined yet.

Step 2 — Identify the work

What does the process actually involve?

• Repetitive tasks
• Classification
• Search
• Summarisation
• Content generation
• Prediction
• Decision support
• Data extraction
• Communication

Step 3 — Evaluate suitability

Rate each dimension low, medium or high:

• Repetition — how often the work recurs
• Data availability — whether the information needed exists and is reachable
• Potential time saving — what would actually be freed up
• Human-review feasibility — whether someone can realistically check the output
• Integration feasibility — whether it can reach the systems involved
• Risk — what an undetected mistake would cost

A use case rating high on repetition and data availability, and low on risk, is a good first candidate. One rating high on risk and low on review feasibility is not, however attractive it looks.

Step 4 — Define the human role

• What should remain under human control?
• What should the system assist with?
• What should the system never decide on its own?

Step 5 — Define the first experiment

• Use case
• Smallest useful version
• Data required
• Human reviewer
• Success criteria

---

Start with a valuable workflow, not an AI feature.

Related reading: AI agents explained.