Why Your Vans Drive More Than They Work

An engineer starts in the east end, drives across town for an eleven o’clock, comes back for a two o’clock four streets from where they started, and finishes in the west.

Nobody planned that. It is the order the calls came in, and the scheduler had eleven other things to do.

The cost is bigger than the fuel

Fuel is the visible number and the smallest one.

An hour of unnecessary driving is an hour of a skilled person’s day that cannot be billed, and it comes out of the only part of the week that generates revenue. Across six engineers, an hour each, that is most of another engineer’s output, permanently unavailable.

Then the knock on effects. Arrival windows get wide because nobody can predict the travel, so clients are told between eight and one. Jobs overrun into the next slot, the last appointment of the day gets moved, and that client is now the one who tells people you are unreliable.

Why the scheduler cannot fix it

Not a competence problem. The task is genuinely hard and it changes continuously.

A good schedule has to weigh travel time, skills, parts availability, contractual response times, client preferences, engineer working hours and the jobs already booked. That is a large problem for six engineers and forty jobs, and a person solving it well would need most of the morning.

Then at nine fifteen an emergency arrives, somebody calls in sick, a job overruns by two hours, and the whole thing has to be solved again. So it never gets solved properly. It gets solved once, badly, and then patched all day.

What the scheduling software did not do

Most field service systems will optimise a route. Very few handle the part that actually breaks the day.

They assume the job is what the ticket says. A job logged as a routine service that turns out to need two hours and a part nobody carries is what destroys an afternoon, and no route optimiser prevents it, because the information that would have predicted it is sitting in a free text description or in the client’s call.

They also assume the plan holds. Real days require constant re-planning, and a system that produces a beautiful morning schedule and then goes quiet is not much help by eleven.

Where AI closes the gap

Two things, and the first is the one nobody expects.

It reads what the job actually is. A description written by whoever took the call, in their own words, tells you far more than the category they selected from a dropdown. An AI system reading that against thousands of past jobs can estimate the real duration, the skill needed and the parts likely required. Getting the duration approximately right is worth more than any routing improvement, because the schedule fails when the estimate is wrong, not when the route is imperfect.

It re-plans continuously. When the nine fifteen emergency lands, the whole day gets re-solved in seconds, including who is closest, who has the right ticket, and which existing appointments can move without breaching a contract. Then it tells the affected clients before they ring you.

It also handles the part nobody has time for: a message to each client the evening before, and a narrower arrival window on the morning, updated if the engineer is running late. That single change removes most of the calls asking where somebody is, which is itself a meaningful share of your office workload.

What to fix before you touch scheduling

An honest caution, because this is where these projects fail.

If your job durations are recorded as whatever was booked rather than what actually happened, no AI system can plan the day, because it has nothing true to learn from. Start by capturing real start and finish times for a month. It is unglamorous and everything else depends on it.

Same with the van stock. If nobody knows what is on each vehicle, the second visit for a missing part will keep happening regardless of how good the route is.

The engineers will know before the data does

Before any system, spend a morning in the van with two of them.

They can tell you exactly where the day goes: the jobs that are always mis-described, the client sites where access takes forty minutes, the part that is never in stock, the office practice that sends them across town twice. None of that is in your job records and all of it is costing you hours every week.

It is also the fastest way to find out whether the problem is scheduling at all. Sometimes the answer is that jobs are booked with no travel allowance between them, which is a booking rule rather than an optimisation problem and takes an afternoon to change.

What not to automate

Two things, and both matter more in field service than in most businesses.

The decision to send somebody to a genuine emergency stays with a person. The cost of being wrong is a vulnerable client without heat overnight, and no efficiency gain justifies that risk.

And do not automate the message that tells a client you are running two hours late on a job that has already been rescheduled once. That call is about the relationship, and an automated notification on the second failure reads as a business that has stopped paying attention.

The number to measure

• Wrench time. The proportion of a paid day actually spent on site working. Most field businesses that measure it for the first time find something between 40% and 55%, and are surprised.

• First time fix rate. Every return visit is a job you did twice and billed once.

• Jobs per engineer per day, tracked over months rather than weeks, because this is the number that turns an improvement into visible capacity.

• Arrival window width. The client facing measure, and the one that decides whether they use you again.

Moving wrench time from 45% to 55% is roughly an extra engineer’s worth of capacity across a team of six, without hiring anybody. That is normally the largest single improvement available to a field service business, and it is entirely a scheduling and estimating problem rather than a labour one.

AI Optimize estimates the real length of a job from how it was described, re-plans the day when something changes, and tells the client before they ring you. That work sits under Workflow Automation.