Choose a task, not an entire department
“Improve customer service” is too broad to test. “Answer questions about opening hours and preparation for a first visit” gives you a defined starting point. Choose a task that happens often enough to observe, has reasonably stable information and has a person responsible for exceptions. Record what is explicitly outside the pilot.
Prepare knowledge and realistic examples
Collect the answers your team currently trusts, rather than every document you own. Remove obsolete guidance and decide who approves changes. Build a small set of anonymised examples: ordinary requests, missing details, ambiguous wording and a request that must reach a person. Avoid real patient records or unnecessary personal information in preparation material.
Agree the handoff before the first live request
An escalation is only useful if someone receives and handles it. Name the receiving team, its availability and the information it needs. Decide what the customer should be told when no colleague is available. If a calendar or another system is involved, check account permissions and the exact supported action before adding it to the pilot.
Measure quality as well as speed
Record a baseline for the chosen task. Useful measures include correct answers, appropriate handoffs, missing details and the editing effort left for the team. Count only outcomes you can verify. A message sent, a link shown and an appointment confirmed are different results. Set review dates and decide in advance what would justify continuing, adjusting or stopping.
Keep this checklist
- One task and a named owner
- Approved knowledge and anonymised test examples
- Access, handoff and exception rules
- Baseline, review dates and success criteria
Prepared by the Feleo team. Examples are planning guidance, not customer results. Confirm the scope of your own pilot before launch.