The Operator's Brief · No. 01 · Field service · Routing
The Override Tax
Your software plans tomorrow perfectly. Then your dispatcher moves one job, for a reason she is right about, and the eleven stops behind it land wherever they land. That is the tax. Not the job she moved. The eleven she didn't.
Free to readBy Ben Guthmiller, Founder & Editor8 minBrief · Workup · Play
Get each new issue free
What this issue finds
Axiom shows why the cost of a dispatcher's override is not in the job she moved but in the stops sequenced behind it, models the annual override tax on a field-service route, and uses the UPS ORION routing program as the teardown for why the fix was never a software problem.
01
What is going on here that I did not know?
Every route has a Mrs. Alvarez
Somewhere on your route tomorrow there is a customer whose schedule isn't in your software. It's in your dispatcher's head. Ours is Mrs. Alvarez, who leaves for dialysis at nine on Tuesdays and Thursdays, which means she has to be the first stop of the day. Everybody in the office knows this. Nobody has ever typed it anywhere.
So before the trucks roll, Dana moves her to the front. Dana has run your dispatch for six years and she is right to do it. The problem is the eleven stops that were sequenced behind Mrs. Alvarez. Dana can't redo eleven stops in her head with the phone ringing, so she doesn't, and they land wherever they land.
Exhibit 01
Moving one job costs you 44 minutes on the eleven behind it
Illustrative model. A twelve-stop pool route, same twelve jobs both ways, only the order changes.
As planned
Time on jobsTwelve stops, sequenced
As driven, once Mrs. Alvarez moves to first
Same time on jobs
+44 minExtra driving
Same twelve jobs, same durations. The extra time is all driving, and the day just got longer by that much.
NOTES · Jobs modeled at 30 minutes each; driving between stops is modeled, not measured. SOURCES · Axiom model, August 2026. Axiom estimate.
Forty-four minutes, out of one defensible decision. Now do that eight times a week, across twelve trucks, for a year. That is the override tax, and the whole play fits in a sentence: write down every override for four weeks, then use an AI to turn what you wrote into rules your software can read. No new software, no consultant, about five hours of your own time across the month.
The tax is behind the job, not in it.
02
Why do most attempts at this fail?
Two people are handing you this bill
They are telling you completely different things, and the fix is different for each. Miss the difference and the whole thing falls over.
Exhibit 02
Your dispatcher and your techs are reporting two different problems
What the two sides of an override log are actually telling you, and what each one needs.
Dana · dispatch · before the trucks roll
"A rule is missing."
She knows about the dialysis, the Kellermans' Tuesday-only gate, and that Miguel is the only tech trained on the variable-speed pumps. Your software knows none of it.
→ Put it in the software as a rule
Miguel · field · at a locked gate
"Your numbers are wrong."
That job is booked at 40 minutes and has never once taken less than 90. The address points at the front office instead of the equipment pad. It looks like a rule. It's a wrong number.
→ Fix the number, don't make it a rule
SOURCES · Axiom, August 2026. The two categories are Axiom's.
03
So what is my number?
How often, times how deep
The depth is the whole game, and almost nobody thinks about it. Twenty-two stops a day means eleven jobs sit behind every override. Three stops a day means almost nothing does. Same software, same override, wildly different bill.
Exhibit 03
What your overrides cost you every year
Three sliders. If the override count isn't written down anywhere, guess high, then go count for a week.
Roughly what the override tax costs you a year, at $32 a loaded hour
$50,000
NOTES · Depth is (stops − 1) ÷ 2. Assumes $32 a loaded hour and fifty working weeks. The 2.6 minutes per scrambled job is an Axiom estimate and the load-bearing figure. SOURCES · Axiom model, August 2026.
04
Has an owner like me actually run this?
The teardown: UPS, 2003
Every month we put a business on the table with real numbers and call it in public. This month it's the company that invented your problem, at a scale that is hard to look at. UPS has 55,000 Mrs. Alvarezes, and not one of them is written down anywhere the company can look up.
Teardown No. 01 · Logistics · 2003United Parcel Service
55,000
delivery routes
160
stops per driver / day
$50M
value of one mile saved / yr
10 yrs
before it works, if it works
The ask. $250 million and ten years, to build routing software that tells 55,000 veteran drivers what order to drive in.
The upside. Shave one mile off each driver's day and that is $50 million a year, forever. They believe there are eight miles in there. Payback in roughly nine months.
The risk. 55,000 drivers judged on missed deliveries, who did not ask for any of it. What the software needs to know is written down nowhere. If it were, they wouldn't need the ten years.
You're in the room in 2003. Do you fund it?
$250 million, ten years, 55,000 drivers who never asked. Decide before you read on.
It took thirteen years, and six of them went on discovering it was never a math problem.
The software worked early. It produced routes drivers would not drive: a customer with a standing daily pickup started getting a different time every morning, because the system worked the route out from scratch each day. Better on paper, worse on the road. So UPS rebuilt the rules out of what the drivers actually knew. The mechanism inside the system today: when a driver goes off the plan, it asks him why, and his answer goes back in.
A quarter of a billion dollars and thirteen years, and the machine at the end asks the driver why he overrode it and writes down what he says.
That isn't an anticlimax. That's the finding. The software was never the hard part, and it isn't the hard part for you either. Of every number on this page, the cost of getting knowledge out of somebody's head is the only one that has moved since 2003, and it moved to roughly nothing.
05
Where does this pay, and where doesn't it?
Find your trade. Stops per day decides it.
Exhibit 04
Stops per day decides whether you run this
Typical daily stop counts by trade, and the call each band gets.
Run it
12+ stops a day. Pool, pest control, medical courier, linen, vending, propane, lawn. One override scrambles eight to twelve jobs behind it.
Worth a month
4 to 10 stops. HVAC, plumbing, electrical, appliance, locksmith. The wrong job times are usually worth more than the routing.
Skip it
Under 4 stops. Cleaning, roofing, install, moving, concrete. Nothing sits behind an override. Run the same four weeks against quoting instead.
SOURCES · Ranges are Axiom estimates from category norms, August 2026, not survey data.
Take your densest route and count how many jobs sit behind a typical override. That single figure decides whether any of this is worth your month, and it takes ten minutes.
Know a dispatcher who'd recognize Dana?
Part 02 · The Workup
You’ve felt the wound. Now measure it in your own numbers — the full model, and a calculator set to your plan.
One case, one number you believe that turns out wrong, the model and the play. Free, the first Sunday, in your inbox before it goes anywhere else.
No spam, no teasers, no dark patterns. One click to leave.
Already running these on your own jobs? The reading stays free. The Bench is the tools.See The Bench · $49/mo →
Published with this issue · the four weeks step by step, and the arithmetic behind every number on this page. Axiom Research · The Operator's Brief · No. 01