The most expensive twenty minutes in a field service business happen at about half four in the afternoon.
A job has overrun. The engineer rings in. Somebody in the office now has to work out who picks up the two calls that were meant to follow it, whether the customer at the second one can be moved to Thursday, which van is nearest, who is closest to being out of hours, plus whether the part needed tomorrow is actually on the right vehicle. That person is usually the owner. They are usually doing it while standing up.
Nobody has ever put that twenty minutes on an invoice. It happens most days, it happens again at seven in the evening when the paperwork comes back, plus it is the single clearest example I know of a business running on somebody's memory rather than on a system.
That is the job worth talking about, because this year it became a realistic thing to hand over.
What actually changed
For about two years, most AI in a small business meant a chat box. You asked it something, it answered, you copied the answer somewhere useful. Genuinely handy for writing a quote or untangling a supplier email. Not much use at half four, because the problem at half four is not that you do not know what to do. You know exactly what to do. You just do not have time to do it twelve times.
What moved this year is that these systems started taking actions rather than only producing text. The industry calls them agents, which is a poor name, because it sounds like something that replaces a person. A more honest description is that it is software that can be given a goal plus a set of tools, then work through the steps itself instead of waiting for you at each one.
The practical difference: a chat box can tell you which customer to ring. An agent can look at today's jobs, work out the knock-on effects of the overrun, propose a new running order, draft the three texts, then wait for someone to say yes.
That last part matters more than anything else in this article. The good implementations stop and ask. They do not ring your customers on their own.
Why scheduling is the obvious first place to look
Because the cost is not where people think it is.
Ask most owners where the waste is in their week plus they will say drive time. Drive time is visible, it is on the tracker, it is easy to be annoyed about. It is also mostly fixed. Your patch is your patch.
The expensive part is re-planning. Every re-plan is a small research project: what changed, who is free, who is qualified for it, what does the customer already expect, what did we promise, what will it cost us to break that promise. Twelve vans generate a surprising number of those in a week, plus each one lands on the person least able to spare the time.
This is the same point I make about any business, whether it has vans or covers. Look for the job that gets redone, not the job that takes longest. The redone job is where the hours quietly go. I have written the longer version of that exercise up as the bottleneck audit, which works the same way whether your bottleneck is on the road or behind a counter.
The three places it actually lands
Scheduling plus re-scheduling. The realistic version is not a system that plans your week from scratch. It is one that holds the current plan, notices when reality has diverged from it, then puts a revised plan in front of you with the reasoning attached. You approve or you override. The value is not that it is cleverer than you. It is that it does the boring recalculation in four seconds rather than twenty minutes, plus it does it at half four when you are tired.
The callout. Out of hours, the question is nearly always the same. Is this an emergency, who is on, what did we agree in the contract, what is the charge. That is a decision tree, plus decision trees are exactly what this technology handles well. Getting the triage drafted plus the right engineer identified before anyone picks up the phone takes a genuine chunk out of the worst part of the week.
The paperwork after the job. Job sheets, photographs, parts used, the certificate, the invoice. Most of that information already exists by the time the engineer gets back in the van. It just exists in the wrong shape, in a notes app or on a phone camera roll. Turning it into the document somebody needs is transcription plus formatting, which is work no human should still be doing at seven in the evening.
Where it does not work, which is worth saying plainly
It will not fix a scheduling problem that is actually a capacity problem. If you are consistently promising more work than twelve vans can do, better sequencing buys you a week of relief, then you are back where you started with a subscription to pay.
It is not good at the things your engineers are good at. Deciding a job is not safe to complete, reading a customer who is upset for reasons that have nothing to do with the boiler, knowing that the address on the ticket is the flat above the shop rather than the shop. That is judgement built on years, plus none of this touches it.
It also needs your data to be in some sort of order. Not perfect, but consistent. If half your job history lives in a shared mailbox plus the other half is in somebody's head, the first honest job is not automation at all. It is getting the record straight.
And I would be careful about anything sold as fully autonomous. In this trade you are sending people into other people's homes plus commercial premises, against certificates that carry your name. A system that acts without a human confirming it is a liability wearing a productivity badge.
Four questions worth asking a supplier
Most of what is being sold into this trade right now is the same underlying capability wearing different branding. The differences that matter are rarely the ones on the front of the brochure, so these are the questions I would put to anyone selling you a scheduling assistant.
What happens when it is wrong? Not if. When. You want to hear about an approval step, an audit trail, plus a way to see why it proposed what it proposed. If the answer is that it is very accurate, that is not an answer.
Where does our data go plus who else sees it? You hold customer addresses, access details, sometimes alarm codes. That is worth a straight answer in writing rather than a reassuring noise on a call.
What happens when we leave? Ask how you get your job history out, in what format, at what notice. A supplier who is comfortable with this question is usually comfortable for a reason.
Can we start with one crew? If the only option is everybody on day one, that is a commercial decision dressed as a technical constraint. Walk unless you have a very good reason not to.
None of that is specific to AI. It is the same list you would use for any system that ends up holding your operation. The reason it matters more here is that these tools are moving quickly, plus a fast moving market produces confident salespeople well before it produces mature products.
How to work out whether this is for you
Do this for one week, on paper, no software required.
Every time the day's plan changes after it has been set, put a mark on a sheet. Next to the mark, write roughly how long it took to sort out plus who sorted it. That is it.
At the end of the week you will have a number. Most owners I do this with are surprised by it, plus the surprise is rarely pleasant. If the number is small, you have a well run schedule plus your bottleneck is somewhere else, probably in quoting or in chasing money. If the number is large, you have found the thing worth automating first, plus you now have a baseline to measure any change against.
That is the whole method. Find the bottleneck, prove it with a number, then work out what can realistically be handed over. Buying software before you have that number is how firms end up paying monthly for something nobody opens.
Where to start, if the number is large
Start with the smallest reversible thing. One workflow, one van team, four weeks, with the plan that a human still approves every action before it reaches a customer. Keep the paper sheet running so you can compare like for like.
Do not start with the whole schedule. Do not start by replacing a system your team already trusts, however clunky it is. The clunky system has years of hard won exceptions built into it, plus you will only discover what they were by breaking them.
If you want to see what this looks like applied to a trades or engineering business specifically, there is more detail on our AI for field service and trades page. If your problem sits further upstream, in the admin around the work rather than the work itself, the automation side covers that ground instead.
Either way, run the week first. The sheet of paper costs nothing plus it will tell you more about your business than any demo will.