Where AI creates value: choosing the problem

Step 1 Warm-up recall

From M1.L1: name the family of a system that routes invoices by a fixed rule.

Your answer

Private: nothing you type here is sent or stored.

Step 2 Learning goal

By the end of this lesson you can:

• state tests for a good AI use case (Foundation);
• apply the tests to a candidate (Core);
• explain why a poor candidate should wait (Core).

Step 3 A familiar example

A task done once a year by one person is rarely worth automating, however clever the tool. A task done thousands of times a month with clear inputs often is.

Step 4 Clear explanation

Four tests help to sort promising from poor use cases.

1. Frequency and volume. The task is repeated often enough for savings to add up.
2. Usable data. The inputs exist in a form a system can use, and you may legally use them.
3. Tolerable errors. An occasional mistake is affordable or caught before harm is done.
4. Measurable result. You can state the current cost, time or quality and see a change.

A candidate that passes all four is worth a value case. One that fails the second or third test should wait until data or controls improve. A candidate that fails the fourth cannot be managed: without a measure you cannot tell whether it worked.

Notice also what the tests omit: whether the technology is fashionable. Leaders often start with a tool and search for a problem; reverse that. The entry on defining project scope and managing risk describes how to bound a small first project.

Step 5 Worked example (4 marks)

Candidate: using generative AI to draft board minutes that the company secretary will review.

1. Frequency: monthly meetings, modest volume. (1 mark)
2. Data: meeting notes exist and are confidential, which limits tool choice. (1 mark)
3. Errors: a reviewer catches them, so tolerable. (1 mark)
4. Measure: hours saved per meeting. Verdict: small gain, so a low priority unless volume is high. (1 mark)

Step 6 Learner practice (instant-marked)

Step 7 Recap

• Four tests: volume, data, tolerable errors, measurable result.
• Fail the data or error test and wait.
• Start with the problem, not the tool.

Step 8 Exit ticket (ungraded)

Apply the four tests to one AI idea in your organisation.

Your answer

Private: nothing you type here is sent or stored.

Step 9 Go further (Extension)

List three processes with high volume and a measurable cost in your organisation.