How to Choose Your First AI Pilot: A UK SME Checklist

By WaypointAI · Reviewed 10 November 2026 · For UK SME owners and operations managers

Choose a recurring task with a clear owner and a result you can check. The best first pilot makes a business decision easier: expand it, improve it or stop.

Start with the process you can describe

Before comparing products, write down what happens today. Who receives the work? What information do they need? What do they produce, and who checks it? If the team cannot explain the process, automating it will make the confusion harder to diagnose.

Look for work that happens often enough to observe. Preparing routine enquiry drafts or assembling a weekly report may be easier to evaluate than a rare, high-stakes decision. These are examples to assess, not promises that AI will suit every business.

Use five questions to shortlist a pilot

  1. Is the pain visible? Record frequency, time spent and common corrections. Separate measured figures from estimates.
  2. Can you judge a good result? Agree what an acceptable output contains and what would make it unsafe or unusable.
  3. Is suitable information available? Check access, accuracy, ownership and whether redacted examples can support early testing.
  4. Can a person keep control? Name the reviewer and decide which actions require approval.
  5. Can you return to the manual process? A pilot should have a clear stopping point and a workable fallback.

Compare candidates without false precision

For each candidate, mark likely value, delivery effort and risk as low, medium or high, then write one sentence explaining the evidence. Do not add the scores into a supposedly scientific answer. A high-value workflow can still be unsuitable if the business cannot check its output or approve access to the information it needs.

A useful pilot brief fits on one page: task; owner; input; output; approved data; review step; baseline; acceptance test; costs; fallback. Missing answers are discovery work, not gaps to hide with optimistic assumptions.

Define the test before configuration

Choose examples that reflect normal work and awkward cases. An enquiry workflow might include an incomplete message, a request outside your services and conflicting information. Decide how the workflow should respond: prepare a draft, ask a question or pass the case to a person.

Keep a small set of examples separate from those used during configuration. Repeatedly tuning against the same examples can make a demonstration look convincing without showing whether the workflow handles unfamiliar work. Record failures and changes so the final review can explain what improved and what remains uncertain.

Include the cost of checking

A fast first draft is useful only if the checking and correction do not consume the saving. Measure the entire task, including setup, review and exception handling. Read our guide to measuring pilot value before agreeing a target.

Know when to pause

Pause when access is unclear, source information is unreliable, nobody owns the process or an important error cannot be detected before it causes harm. Sometimes a clearer template, a rule-based workflow or an existing software feature is the better first investment. A pilot that establishes this has still answered a useful business question.

If you need help choosing the starting point, WaypointAI's AI Discovery & Readiness Assessment reviews opportunities, risks and dependencies before implementation is agreed.