How AI dispatch triage works for a trades business
What happens between a customer describing a burst pipe and a technician being on the way, and which parts of it software can be trusted with.
Dispatch is a triage problem wearing an admin problem's clothes. The hard part is not writing down an address. It is deciding, in ninety seconds, whether this call means someone drives out tonight or Tuesday — and being right often enough that neither the customer nor the on-call technician stops trusting the answer.
What a dispatcher is actually doing
- Working out what has actually happened, from a description by someone who is not a plumber.
- Deciding how urgent it truly is, as distinct from how urgent it sounds.
- Checking the address is inside the service area.
- Finding a slot that matches the skill the job needs.
- Getting the details to whoever is driving.
- Setting the customer's expectation about when.
Steps three to six are procedure. Steps one and two are judgment, and they are where the cost of a mistake sits.
Where software is genuinely good
Capture is the obvious win. An AI dispatcher takes the address, reads it back, confirms the callback number, records access instructions, and does it identically on the fortieth call of the day. Human dispatchers make transcription errors when three lines are ringing. Software does not get tired at 4pm.
Service-area checking is a second clean win, because it is a lookup. So is availability, if your calendar is connected. So is texting job details to the assigned technician the moment the call ends, which is a task humans routinely defer and then forget.
Where it needs guardrails
Urgency classification is the risky one, and it fails in both directions. Over-classify and you wake the on-call technician for a dripping tap. Under-classify and a customer sits with water coming through a ceiling because the system heard "leak" and scheduled Thursday.
The workable approach is to write the emergency criteria explicitly rather than leaving them to inference. No heat below a stated temperature. Any active water ingress. Gas odour. No power. Anything involving a vulnerable occupant. Given rules like those, classification becomes closer to a checklist than a judgment — and where a call does not fit the rules, the correct behavior is to escalate rather than guess.
The safety script matters more than the booking
For genuine emergencies, what the agent says before the technician arrives can matter more than the appointment. Where the stopcock usually is. Turn off the power at the consumer unit. If you can smell gas, leave the building and ring the emergency number, then ring us back.
That should be written by someone in your business who knows the trade, not improvised by a model. Put it in the employee's instructions verbatim.
What good looks like after a month
- Emergency calls dispatched in minutes, at any hour, with the same information every time.
- Routine calls booked without anyone being interrupted.
- Out-of-area calls identified before a van is committed.
- Every call recorded, so a disputed job has a transcript rather than two recollections.
- Your experienced dispatcher spending their day on the twenty percent that needed them.
See what an AI dispatcher is configured to handle, and read what AI agents get wrong before you put one in front of emergency work.