Plan for the people who will use AI

Define the destination before the demonstration

A pilot can show that an AI tool completes a task under controlled conditions. Introducing it into daily work raises additional questions: who can use it, which information it can access, how a result is checked and who responds when it fails. Make those questions part of the pilot scope.

Name the people expected to use the capability and the workflow it will enter. Agree what useful performance looks like using examples from that workflow. A convincing demonstration is evidence about a capability. The release decision also needs evidence about the conditions in which it will operate.

Include operational ownership

Someone needs to own access, evaluate changes and respond to problems after the pilot team moves on. Identify that person or team early. Include them in decisions about monitoring, support and how users report a problem.

Thoughtworks’ guidance on scaling AI beyond proofs of concept includes cross-functional teams, supporting platform work and adoption for a specific user group. The practical implication for a buyer is to make that surrounding work visible in the plan, with responsibilities and dependencies alongside the model or application work.

Test the awkward cases

Choose examples that resemble the work people will actually face, including poor inputs, missing context and requests outside the intended scope. Agree when the tool should proceed, when it should ask a question and when it should stop.

Keep a record of the test cases and the decisions behind them. When a prompt, model, connection or business rule changes, those examples provide a basis for checking whether the workflow still behaves as intended.

• Can users see the source information needed to check a result?
• Can they correct or reject an output without losing their work?
• Does a failed connection leave the task in a recoverable state?
• Is there a named owner for issues that the user cannot resolve?

Make training specific to the work

Give people practice with the tasks, permissions and exceptions they will encounter. Teach them how to recognize uncertainty and what to do with it. Include the managers and process owners who will support the new way of working.

A pilot should end with a clear decision: release the capability within an agreed scope, address named gaps or stop the work. We can help connect that decision to implementation and enterprise AI training, so the transition into everyday use is planned from the start.