Look for a repeatable handoff
Start by listing work that repeatedly moves between people or tools. A new enquiry might be copied into a CRM, a document might need a summary, or an approved record might need to update another system. We look for a task with a clear beginning and end. “Automate our operations” is difficult to scope; “prepare an enquiry brief for review” gives the team something concrete to assess.
Separate rules from judgement
Map the decisions in that workflow. Some steps follow fixed rules, such as checking that an email field is present. Others require interpretation, such as preparing a summary of an open-ended request. Our approach is to keep those responsibilities explicit. Where an AI-generated draft could affect a customer or an important business decision, define who reviews it and what they need to see before approving the next action.
Plan the exceptions first
Ask what should happen when a field is missing, a connected service is unavailable or the same request arrives twice. Decide where incomplete work waits and who is notified. We also ask which systems may be read or updated and what information is necessary for the task. These boundaries turn an appealing demo into a more practical project brief.
Measure a small pilot
Record how the task works today: time spent, common corrections and delays between steps. Run a limited pilot with representative inputs, then compare the result against that baseline. Include the effort spent reviewing and maintaining the automation. Expand only when the team can explain what improved, what still needs attention and who owns the workflow.
