What to automate first, and what to leave alone

Every shop that has ever bought software has automated the wrong thing first. Not through carelessness — through a reasonable instinct that turns out to be backwards. You automate whatever annoyed you most last week, which is usually whatever was most irritating rather than whatever was most expensive.

Irritation and cost are barely related. The thing that wound you up on Friday took ten minutes. The thing that cost you four hundred pounds happened quietly and nobody noticed.

The test that actually works

Before automating anything, ask three questions in this order:

Does it happen the same way every time? Automation is repetition applied at speed. A task that goes differently depending on who the customer is, or what you found in the loft, is not repetitive — it is judgement wearing a routine's clothes.

Does it need a decision about a person? Whether to give someone longer to pay, whether to price a difficult job high or take it as a favour, how to tell a customer their unit is finished — these are relationship decisions. A small shop's advantage over a national chain is that a human makes them.

If it goes wrong, how expensive is that? A reminder sent twice is embarrassing. A quote sent automatically at the wrong price is a job you do at a loss or an argument you have in a driveway.

Automate where the answers are: yes, no, cheap. Leave the rest.

What passes the test in almost every shop

Appointment reminders. Identical every time, no judgement, and the failure mode — a duplicate text — costs nothing. It also directly reduces no-shows, which is one of the few places automation returns money the same week.

Paperwork you retype. Anything you enter twice because two systems do not talk. The retyping is not just slow; it is where transcription errors get introduced, and those surface months later as a dispute about what was agreed.

Follow-up prompts — the prompt, not the message. A list of quotes nobody has chased in ten days is enormously valuable. The message that goes out should still be yours.

Recurring service scheduling. Maintenance visits due in six months are the definition of repetitive, and forgetting them is silent lost revenue.

Anything that only reminds you. Internal nudges have no customer-facing failure mode at all, which makes them the safest category to be aggressive with.

What to leave alone, and why

Pricing a non-standard job. Every trade has jobs where the right price depends on access, condition, and how much the last person botched it. Automated pricing here is either too high, losing you work, or too low, which you find out on site.

Chasing money. This is the one people are most tempted by and it deserves the most caution. A dunning sequence does not know that this customer is slow because their mother died, and that one is slow because they always are. It also puts you in a different legal category in some countries — automated collections messaging is regulated far more tightly than a service reminder, and the penalties are per message.

Draft the reminder automatically. Send it yourself.

Anything that speaks to a customer about a problem. Delays, breakages, things found on site. Automated tone is fine for "your technician is on the way" and actively damaging for "we cannot come today."

Diagnosis. Photo-reading tools are getting genuinely useful as an assistant — a second opinion on a nameplate, a prompt about a symptom. As a decision-maker they are confidently wrong in ways that cost far more than the time they save.

The order I would go in

1. Reminders to yourself first. Zero risk, immediate benefit, and it teaches you what you actually forget.
2. Appointment reminders to customers. Highest return for the lowest danger.
3. Stop double-entry. Whatever you type twice, fix that next. It is usually the largest silent time cost in a small shop.
4. Recurring work. Get maintenance visits onto a schedule that reminds you rather than living in your head.
5. Drafted messages for anything sensitive. Prepared by the software, sent by you.

Notice that four of the five involve no customer-facing automation at all. That is not caution for its own sake — it is where the returns are.

We went looking for someone else's case study

This is the point where an article like this produces a shop that automated something and saved a fortune. We do not have one of our own — we are new enough that our customer list is a rounding error — so the plan was to borrow somebody else's and say so.

Here is what we found instead. The examples we found were published by the companies selling the software in them. Their headline improvements were large, but we did not find an independent audit behind the examples we reviewed.

We are not going to repeat those figures as though they were evidence, because we spend most of our time telling people to check vendor claims and it would be a poor look to launder someone else's on the way past.

So: no case study today. What we can tell you is the mechanism, which is what the rest of this article is, and you can test it against your own week for free. Ask us again in a year and we intend to have our own numbers here — measured on real shops, with the method written down, and with the uncomfortable ones left in.

If that turns out to be optimistic, at least it will be on the record.

What the current AI wave changes, and what it does not

The current wave makes drafting cheap: quotes, follow-ups, summaries of what a technician wrote. That is real and worth using.

What it does not do is take responsibility. If an automated quote is wrong, the customer is not arguing with a model — they are arguing with you, in a driveway, holding a piece of paper with your name on it.

So the rule that has held through every previous wave of this still holds: automate the typing, not the promise.