Ask anyone who has worked a dispatch board during the first heat wave of the summer and they will tell you it is one of the hardest jobs in the trades. Every booked call, late tech, part that just arrived, and customer who needs to reschedule lands on one person, who is also trying to keep the board profitable.
So it is no surprise that AI dispatching is one of the most talked-about features in field service software right now. ServiceTitan, for example, has written openly about how its dispatch tools recommend or automatically assign technicians based on skills, location, availability, and predicted job value.
The idea is sound. But in the shops we work with, the first thing that breaks is not the algorithm. It is the data underneath it.
What an AI dispatcher actually reads
Whatever tool you use, automated dispatching makes decisions from a handful of inputs:
- Technician skills and certifications. Who can work on a rooftop unit, a boiler, a tankless heater, a commercial panel, a refrigeration rack.
- Zones and drive time. Where each tech starts the day and which areas they cover.
- Job types and durations. How long a no-cool diagnostic, a drain clearing, or a quarterly PM actually takes.
- Job value and priority. Which calls are emergencies, which are likely replacements, which belong to a commercial account with a contracted response time.
- Technician performance. Close rates, average tickets, callbacks.
If any of those is wrong, the system makes confident decisions on bad information. And it does it every ten minutes.
Where the data usually breaks
The skills spreadsheet nobody owns
Most companies keep tech skills in a spreadsheet, a whiteboard, or the dispatcher's head. It was accurate the day it was built. Since then, two techs got certified on a new line of equipment, one stopped doing commercial work, and a new hire is still riding along. Nobody updated it, because it was everybody's job and nobody's job.
Job types that do not mean anything
"Service call" and "Misc" are not job types. When half the jobs in your system are booked under a generic code, the software cannot learn how long things take or which tech is best at them.
Durations that were set once and never revisited
If every job is booked as a two-hour window, the board looks full when it is not, or empty when the day is about to fall apart.
Commercial work mixed in without its rules
Commercial accounts often come with contracted response times, required certifications, site access rules, and a specific tech the facility manager trusts. If those live in an email thread instead of the account record, the dispatcher remembers them and the software does not.
What to fix first
Before turning on any kind of automated dispatching, we usually work through four things with the operations team:
- Build the skills matrix from real history. Pull every completed job from the last 12 months, group it by job type and technician, and use that as the starting point. Then have each department lead review it. One owner keeps it current.
- Clean up job types. Collapse the generic codes into a short list that actually describes the work, separate residential from commercial, and map old jobs to the new list so the history is usable.
- Measure real durations. Compare booked windows to actual time on site by job type. Update the defaults.
- Put commercial rules on the account. Response time, required skills, preferred tech, access notes. If it is in a contract, it belongs in the record.
None of this requires AI. All of it makes AI dispatching work.
Start in assist mode
Most platforms let you choose between recommendations and full automation. Start with recommendations. The dispatcher sees the suggested assignment and the reason behind it, and accepts or overrides it.
Track the overrides. Every time a dispatcher says no, there is usually a reason: a skill that is not recorded, a customer preference, a site rule. Each one is a data fix waiting to happen. When overrides drop to a small, predictable set, you are ready to let the system take more of the board.
Decide what you are optimizing for
Revenue and drive time pull in different directions. Sending your best closer across town for a likely replacement can be the right call. Doing it every day burns out your tech and leaves maintenance work unassigned. Many teams end up with different rules by department: service and sales techs weighted toward revenue, maintenance techs toward route efficiency, commercial accounts toward contracted response times.
That is a business decision, not a software setting. Make it explicitly, write it down, and then configure the tool to match.
The real payoff
When the data is right, the benefits go beyond a smoother board. Dispatchers get time back for parts, estimates, and warranty claims. Customers get accurate arrival windows. And the owner finally has a trustworthy picture of who is doing what work, how long it takes, and where the margin is. That picture is what every other automation in the business will be built on.