Automation
What a qualifying agent should and should not do
The safe scope for a qualifying agent is narrow and worth defining before it is built: gather facts, classify the enquiry, route it, and set an expectation. Anything that commits the business — a price, a date, a technical assurance — belongs to a person.
Draw the boundary first
The useful question is not what an agent can do. It is what it is allowed to do. Write that down before anything is configured, because the boundary is the design.
A scope we would defend:
- Ask for the facts a person would have asked for
- Classify the enquiry by service and urgency
- Route it, with the full original message attached
- Confirm receipt and state when a person will reply
What sits outside it
Prices, dates, availability and technical assurances. Not because a model cannot produce them fluently — it can, which is exactly the problem. A confidently wrong lead time is indistinguishable from a promise, and the customer will hold you to the version they were told.
Never overwrite the original
Whatever the agent extracts, keep the raw enquiry beside it. Summaries lose the detail that turns out to matter, and the person picking the enquiry up needs to be able to check the agent’s reading rather than inherit it.
Test it with the awkward enquiries
Configure it against the messages that are vague, angry, half-finished or obviously for another business. The clean enquiries always work. Behaviour at the edges is what decides whether the thing is safe to leave running, and it is the part that never gets tested before launch.
Article FAQs
Still have questions?The short answers.
Can the agent give a price if we publish our prices anyway?
It can quote a published figure verbatim and link to where it is published. What it must not do is calculate, estimate or apply a published range to a specific job, because that reads as a commitment and it is the one thing customers remember exactly.
What happens when the agent does not understand the enquiry?
It should hand over and say so. An agent that guesses produces a wrong classification, wrong routing and a wrong first reply. Escaping to a person with the raw message intact costs one step; a misroute costs the enquiry.
