1. Name the business problem

Write the problem without using the term AI. For example: “New inquiries wait too long for an initial review.” This keeps technology choices tied to work that matters, rather than starting with a tool and searching for a use.

Describe the current process, the people involved and the consequence of doing nothing. Record what is known and what still needs to be measured.

2. Map one workflow

Identify the trigger, inputs, decisions, actions and final handoff. Include exceptions. An incomplete process map often hides the work that makes implementation harder than a demonstration suggests.

Ask which parts require interpretation and which parts are straightforward rules. Not every useful automation needs AI.

3. Check the foundations

  • Do you have permission to use the required information?
  • Is the information accurate enough for the task?
  • Can the existing tools support the necessary connections?
  • Who can approve access and resolve questions?
  • What happens if an integration becomes unavailable?

4. Set boundaries before features

Define read and write access, approval steps, escalation and retention. Keep payments, sensitive advice, employment decisions and binding commitments under appropriate human control. Choose providers and handling arrangements before processing private information.

5. Choose useful measurements

Compare the current process with the pilot using task completion, time spent reviewing, error rate, operating cost and user feedback. Collect a baseline before making a savings claim. Estimated capacity released is not automatically an increase in profit.

Use the automation ROI calculator for a transparent scenario, not a financial forecast.

6. Agree what a pilot delivers

A scoped pilot should identify the workflow, supported cases, excluded cases, test approach, responsibilities and go/no-go criteria. Include an operating plan, not just a launch date. A system needs an owner when a model changes or a business rule no longer applies.

7. Decide how to move forward

Launch gradually when the agreed criteria are met. Keep a manual fallback and review early activity. If the pilot does not justify expansion, document what was learned rather than adding more features to hide the problem.

For help turning these questions into an implementation plan, explore AI strategy or complete the Opportunity Finder.