Bound the task
Begin with one repeatable enquiry type and an approved knowledge source. Decide which data the assistant may access and which questions require a human.
Back to project explorationsDESIGN WALKTHROUGH / AI & automation
Useful AI, with an accountable next step.
A design exploration for a team that wants to organise inbound enquiries and prepare suggested replies while retaining human approval.
Original design exploration. Fictional scenario. No client deployment or measured outcome is claimed.
01 / THE CHALLENGE
Understand the whole journey.
An assistant can help organise information and propose a starting point. The team still needs a reliable source, an owner and an explicit decision about what may be sent to a customer.
A focused first phase
Agree the users, responsibilities and acceptance criteria for this journey, then review a prototype before committing to a larger build.
- 01Receive an enquiry
- 02Retrieve approved context
- 03Prepare a suggested reply
- 04A person reviews and sends✓
02 / THE DESIGN DECISIONS
Make the important choices visible.
Keep the evidence close
Show the source behind a suggested answer. Make uncertain or missing information visible so the reviewer can correct the response rather than assuming it is complete.
Keep sending under review
Drafting and sending are separate steps. A named team member checks the recipient, source and wording before any customer message leaves the system.
03 / VALIDATION
Test what the journey depends on.
- Source access and customer boundaries
- Unsupported questions and incorrect suggestions
- Reviewer changes and approval history
- Provider failure and manual fallback
04 / MEASUREMENT
Agree evidence before claiming a result.
These are proposed measures to discuss during discovery. A baseline and targets would be agreed for your project.
- Reviewer time per enquiry
- Corrections needed before approval
- Questions escalated to a person
