How to find the workflows worth automating first
Start with work that repeats, crosses tools, creates delays, and has a clear owner. The best first automation is rarely the flashiest one. It is the workflow your team already understands and can measure.
Look for manual glue, not isolated tasks
Manual glue is the invisible work people do to keep disconnected systems moving. It includes copying enquiry details into a spreadsheet, chasing an approval in chat, renaming files, updating a customer record, and reminding someone about the next step.
A single task may take only a few minutes. The opportunity appears when that task repeats, depends on memory, or delays the next person in the process. Map the complete flow from trigger to outcome rather than automating one click in the middle.
Score each candidate on four signals
A simple scorecard keeps the decision grounded in operational value rather than novelty.
- Frequency: how often does the workflow run?
- Friction: how much copying, waiting, checking, or rework does it create?
- Clarity: are the inputs, decisions, exceptions, and desired outcome understood?
- Consequence: what happens when the workflow is late, incomplete, or wrong?
Choose a contained first pilot
A strong pilot has a defined starting event, a limited number of systems, and a human owner. Lead response is a common example: an enquiry arrives, an approved response is sent, useful questions are asked, and the result is recorded for the team.
Keep human approval where judgement, sensitivity, or commercial risk demands it. Automation should make responsibility clearer, not hide it.
Measure the workflow before and after
Capture a baseline before building. Useful measures include response time, handoff time, completion rate, error rate, and the number of manual touches. Pick the measures that reflect the actual purpose of the workflow.
The result should be a visible operating improvement that the team can inspect and refine. That is more valuable than deploying an AI tool without a shared process around it.