The seat you are actually automating
Start with the fact that the seat is real. In May 2023, the Bureau of Labor Statistics counted 14,900 dispatchers working inside building equipment contractors, the HVAC, plumbing, and electrical shops, at a mean annual wage of $46,660, out of 206,090 dispatchers nationally. That is a funded position with a real payroll line, not a workaround somebody invented because the whiteboard filled up.
So the question is not whether software can delete the seat. The question is which half of that seat's day is arithmetic and which half is judgment. Most of the shops we work with run somewhere between 3 and 20 field techs, small enough that every tech is a named person with known strengths, large enough that a whiteboard stops working around 8 trucks. At that size the dispatcher is doing two jobs at once, and only one of them scales.
The arithmetic half
The arithmetic half is the part that runs on data the software already holds. Who is on the clock. Who holds the license or the manufacturer cert this job requires. Who is closest to the address. What is already committed to the afternoon. How long the last three jobs of this type actually took, not how long the book says they take.
None of that requires a human. It requires a system that has the job history, the tech profiles, and the calendar in one place, which is exactly what a CRM is for when it is wired to the schedule instead of sitting beside it. The shops we talk to describe a normal day as 4 to 8 residential service calls per tech, with 1 to 3 emergency inserts in season. Every insert forces a full re-solve of the board. A person doing that by hand at 10am is guessing; by 2pm they are guessing worse.
Then there is the mechanical follow-through, which is where most of the real leakage lives. Texting the customer the window. Updating the record when the tech runs long. Nudging the estimate that went quiet on day four. Asking for the review while the job is still fresh. That work is boring, it is high-volume, and it is the first thing to fall off a busy desk. Automation does not do it better than a great dispatcher. It does it at 4:45 on a Friday, which a great dispatcher does not.
Why the nearest truck is usually the wrong answer
Naive routing optimizes drive time because drive time is easy to measure. It is the wrong objective function, and the labor numbers explain why.
BLS puts HVAC mechanics and installers at 440,900 jobs in 2025 with about 40,600 openings projected each year through 2035 and median pay of $61,010 as of May 2025. Electricians run 821,000 jobs with 72,700 annual openings, plumbers and pipefitters 510,600 jobs with 42,000 openings. Both of those trades require licensure in most states. A license is a boolean. A manufacturer cert is a boolean. No amount of clever averaging turns the tech 20 minutes away into the tech who can legally close the job 90 minutes away.
Now price the tradeoff instead of arguing about it. The IRS 2026 business standard mileage rate is 72.5 cents per mile through June 30 and 76 cents from July 1. A 70-mile round trip to send the right person costs roughly $53 in vehicle cost. Against a callback, which burns a second truck roll, a second window, and a customer's patience, $53 is not a close call. Shops report callbacks in the low single digits of completed jobs, and the ones that stay there are the ones that stop treating proximity as the tiebreaker of first resort.
Drive time is still a real cost, just not the deciding one. We price a single roll line by line, fuel to tech minutes, in what a truck roll really costs at fall 2026 fuel prices.
Optimize for first-visit completion. Let drive time break the tie, not make the call.
What the software should hand you, and what it should hand back
Here is the shape that works. The router reads the board, applies the hard constraints, prices the soft ones, and proposes an assignment with its reasoning visible: this tech, this window, because of the cert and the 14 minutes. Then it waits. You confirm, or you drag it somewhere else and it re-solves the rest of the day around your change.
That last clause is the whole design. Those studies found the effect twice, at 32% versus 73-76% in the first and 47% versus 70% in the replication. People will use a model they can push on. They abandon one they cannot. A dispatcher who gets overruled by the board on a Tuesday will quietly go back to the whiteboard by Thursday, and you will be paying for software nobody opens.
Which means the override is not a courtesy feature bolted on for comfort. It is the adoption mechanism. Build the boundary in on purpose and publish it.
The calls that stay with you
Some decisions never move, and it is worth saying which ones out loud.
Sequencing a customer relationship stays with you. The account that has been on the books eleven years and just had a bad experience does not get routed by cost. Who eats the overtime stays with you. Which job slips when three cannot fit stays with you, because that is a business decision wearing a scheduling costume. Whether to send the apprentice with the senior tech on a job that is technically simple but politically delicate stays with you.
And the pricing call stays with you. Software can assemble the estimate, hold the line items, send it, and chase it. It should not decide what this customer, on this day, with this history, is charged.
Where OptimizeIt draws it
We built to that boundary. The estimate and quote workflow, the CRM, and the follow-up automation own the mechanical layer: proposals go out, the record updates itself, the quiet quote gets chased, the review ask fires on real job history rather than on somebody remembering. Dispatch routing proposes assignments against qualifications and drive time and re-solves when you move something.
On the phone side, Plus includes the business phone line so calls stop scattering across personal cells. Pro adds the AI Voice Agent, which answers, captures the details, and books into the same calendar the router is reading, so an after-hours call arrives as a job rather than a voicemail. Pricing is $79 for Core with 2 seats, $159 for Plus with 4, and $239 for Pro with 6, on annual billing, with extra seats at a flat $25. The full grid is on pricing.
If your board still lives on a wall and you want the trade-specific version of how it breaks, we wrote that up for plumbing shops in when a plumbing shop outgrows the whiteboard. If the leak is happening before dispatch even starts, read what a missed call really costs. And if you are deciding where to start at all, what to automate first in the trades is the sequencing argument.
The Honest Take
This is not for a two-truck shop where the owner is one of the two trucks. At that size you are the router, you already know every job on the board, and software that asks you to confirm assignments is adding a step to a decision you made in the driveway. Come back when the whiteboard stops fitting.
It is also not for anyone shopping for a dispatcher they never have to think about. That product does not exist, and the shops that buy the promise are the ones that end up running two systems, the real one on paper and the expensive one nobody opens. If you want a board that does the math and then hands the judgment back, that is a different thing, and it is the thing worth buying.
Frequently asked questions
What does an AI dispatcher actually decide on its own?
The arithmetic. It reads the board, checks who is licensed and certified for the work, prices the drive against the job, and proposes an assignment with the reasoning attached. It also handles the mechanical follow-through: texting the customer a window, updating the job record, nudging the estimate that has gone quiet. It proposes the assignment. It does not silently commit the ones that carry a relationship or a warranty behind them.
Why does keeping an override matter so much?
Because adoption tracks control, not accuracy. In the Dietvorst, Simmons and Massey algorithm-aversion research, only 32% of people chose to use a forecasting algorithm they could not modify, while 73-76% chose the same algorithm once they could adjust its output, even inside a tight band. A second study replicated the pattern at 47% versus 70%. A dispatcher who cannot overrule the router will stop opening it by week three.
Should the router optimize for the shortest drive?
Not by itself. Drive time is cheap and legible, so it dominates naive routing. At the IRS 2026 business standard mileage rate, a 70-mile round trip runs roughly $53 in vehicle cost. One callback on a job the wrong tech could not finish costs more than that in truck time alone, and it spends a senior tech's day twice. Optimize for first-visit completion, then let distance break the tie.
Does an AI dispatcher replace the human dispatcher seat?
That is not the honest pitch. BLS counted 14,900 dispatchers inside building equipment contractors in May 2023, at a $46,660 mean wage, so the seat is funded work, not a legacy artifact. What changes is what fills the day. The mechanical part, the board math and the status texts, moves to software. The judgment part, the calls where a relationship or a license is on the line, is what the seat is actually paying for.
What can OptimizeIt handle when nobody picks up the phone?
Depends on the tier. Plus at $159 a month includes the business phone line, so calls land in one place with the job record attached instead of scattering across personal cells. Pro at $239 adds the AI Voice Agent, which answers, captures the job details, and books into the same calendar the router is reading. Both feed the same CRM, so the follow-up automation and the review ask fire off real job history.
How many people can be on the account?
Core includes 2 seats at $79 a month on annual billing, Plus includes 4 at $159, and Pro includes 6 at $239. Extra seats are a flat $25 a month each. For a shop running 3 to 20 field techs, the usual shape is the owner and the dispatcher on named seats first, then techs added as the board moves off paper. Full breakdown lives on the pricing page.