What can AI do for scheduling and dispatch?
AI for scheduling and dispatch is at its best when it proposes the day and a person approves it. It can hold every constraint at once, the skills, the licences, the travel, the parts, the promised windows, the traffic, in a way no human can while also answering the phone. What it cannot do is know that this customer is difficult, that this apprentice should not be sent alone, or that the job at the end of the street is worth doing today because you are already there.
So the machine drafts the run. The dispatcher reads it and changes what needs changing. That takes minutes instead of an hour, and the day still belongs to somebody who understands it.
What is actually going wrong today?
What is going wrong is usually not the plan itself, it is the replanning that starts the moment the day meets reality.
Most field businesses can build a decent schedule the night before. What breaks it is nine in the morning: a no access, a part that did not arrive, a job that turns out to be twice the size, someone off sick. Every one of those forces a rebuild of the rest of the day, and the rebuild happens in someone's head while three people are waiting on the phone.
The value is concentrated there. A machine that can reshuffle the remaining day in seconds, with the constraints intact, and tell you what it would do, is worth more than one that produces a beautiful plan at 6am.
What constraints does it need to know?
All the ones the dispatcher currently holds silently, which is why writing them down is most of the project.
- Capability. Who is qualified, licensed and insured for what. This is a hard constraint, not a preference.
- Geography and travel. Real drive times at real times of day, not straight line distance.
- Windows and promises. What the customer was actually told, including the ones agreed verbally.
- Parts and equipment. Whether the van has what the job needs, or whether a supplier stop comes first.
- Duration reality. How long this type of job actually takes for this technician, from your own history, not from the standard estimate nobody believes.
- Human factors. Pairings that work, customers who need a particular person, apprentices who should not go alone.
That last group is the one that gets left out and the one that causes rejection. A schedule that is optimal and socially impossible will be overridden every morning until people stop looking at it.
Should it dispatch automatically?
Not at first, and possibly not ever for the whole day. Start with the machine proposing and a person releasing.
The reason is the same one that applies everywhere: dispatch is irreversible in practice. Once a technician is driving, undoing the decision costs real money and a customer's morning. Propose and approve keeps the speed and keeps the accountability, which is the principle in human in the loop AI.
Where automatic release does become reasonable is a narrow lane, proven over months. Standard maintenance visits, single skill, generous windows, regular customers. Let it run that lane unattended once the evidence says the dispatcher was not changing those anyway, and keep the gate on everything else.
Will the dispatcher lose their job?
No, and building it as though they might is the fastest way to make the project fail.
The person who schedules your work is the single most valuable source of information in this build, because the constraints live in their head and nowhere else. If they believe the machine is there to replace them, the constraints do not come out, the machine is built wrong, and it gets quietly ignored. That is not sabotage, it is entirely rational.
Build it as their tool. They stop doing arithmetic under pressure and start doing the judgement they were always better at. That is a genuinely better job, and it is the honest framing, not a management line. We take the same position in will AI replace my staff.
How do you know it is working?
Measure jobs completed per technician per day, travel time as a share of the day, on time arrival against the promised window, and how many of the machine's proposals the dispatcher changed.
That last number is the important one and the one nobody thinks to collect. A high override rate is not a failure, it is a map of the constraints you did not write down. Read the overrides, add the missing rules, and watch the rate fall. When it stops falling, you have found the genuinely human residue, and that is where the gate stays permanently.
Take those numbers before you start. Once the machine is running, the old ones are gone, which is the point made in how to measure ROI on AI.
Start with the reshuffle
If you want one place to begin, take the mid morning rebuild. It is the most painful, the most frequent, and the easiest to prove value on inside a month.
Get in touch and describe how your day usually falls apart. For the safety obligations that sit behind fatigue, driving and who you send where, Safe Work Australia is the reference, and business.gov.au covers the rest of the operating ground.