Professional services · OperationsExperimental build
Enquiry Operations Assistant
Built to answer a practical question: how much of the first ten minutes of handling a new enquiry can be removed without losing judgement? The assistant reads the enquiry, classifies it, summarises it and proposes the next step, and a human decides.
- Type
- AI operations assistant
- Services
- Workflow design, AI integration, Internal tools
Challenge
Inbound enquiries arrive as free text across email and web forms. Someone has to read each one, decide what it is, how urgent it is, and who should act, before any value is created.
Objective
Produce a consistent structured record for every enquiry within seconds, and recommend an action, while keeping a person in control of the reply.
Strategy
Treat the model as a careful first reader, not a decision-maker. Constrain its output to a fixed schema, validate it, and show the reasoning so the operator can disagree quickly.
Solution
A small service with a secure API route, a structured-output prompt, validation, and a review surface that lists enquiries by priority with the suggested action ready to confirm.
Result
An experimental build used internally. No client metrics are claimed. The public AI Lab on this site shows the same workflow as a labelled simulated preview.
Design and technical decisions
- Fixed output schema (summary, category, priority, suggested action) validated before anything is stored.
- Server-side processing with per-user limits and no retention of enquiry text beyond the record itself.
- Graceful degradation: when the model is unavailable the enquiry is still stored and queued for manual triage.
- Every automated field is editable in one click, and edits are logged to improve the prompts.