What to automate first.

Written by the Incognito Labs engineering team 16 September 2026 8 min read

The first thing you automate is not really about the hours it saves. It is about whether anyone believes the next one is worth doing.

Start with the twenty-step process that crosses four systems and involves three people’s judgement, and you will spend six weeks on something that half works and nobody trusts. Everyone quietly goes back to the old way, and automation becomes a thing that was tried once here and did not suit us.

Start with something small, frequent and unglamorous, and it runs quietly from the first week. That buys the appetite — and the evidence — for the harder one.

Score each candidate process against these four. A good first job scores well on all of them; anything scoring badly on the fourth is a trap regardless of the others.

1. It happens often. Frequency is what turns a small saving into a real one. Ninety seconds saved on something that happens four times a day is worth more over a year than an afternoon saved on something that happens twice. Frequency also means you find out quickly whether it works.

2. It follows rules, not judgement. If you can write the decision down as “if this, then that” without using the word “usually”, it is automatable. If every case needs someone to weigh things up, you are not automating a process, you are trying to automate a person.

3. Mistakes in it are expensive or annoying. Re-keying an address wrong costs a redelivery. Missing a follow-up costs the job. Where human error is both likely and costly, removing the human step pays twice: the time, and the mistakes that were never going to show up in anyone’s timesheet.

4. It has clean edges. This is the one people skip. You need to be able to say exactly what starts it, and exactly what “finished” looks like. A process that starts “when someone notices” and ends “when it feels done” has no edges to attach anything to.

Image placeholder Suggested: a scoring grid of four or five candidate processes against the four traits, with one clear winner highlighted in the lime accent.

Here is the uncomfortable part, and it is the reason a lot of automation money is wasted: most businesses do not have an automation problem first. They have a process problem.

The process lives half in an inbox, half in a spreadsheet, partly in one person’s head and partly in a system nobody fully trusts. Automating that does not fix it. It makes the confusion faster and harder to see.

Before anything gets built, you should be able to answer four questions in one sentence each:

• What starts it? A form submission, an email arriving, a date, a status change — something observable, not “when Dave gets to it”.
• What normally happens, in order? The happy path, written down, with no branches yet.
• Who owns the exceptions? Every automation meets a case it cannot handle. If there is no named human to hand it to, it will fail silently, which is much worse than failing loudly.
• What does finished look like? A record in a particular state, a message sent, a slot booked. Something you could check.

If you cannot answer those, that is the work. It is also free, and you can do it this week without buying anything. Half the value of our own first phase is simply writing this down with someone who asks awkward questions about it.

A good automation supplier will talk you out of things. These are the ones we push back on:

• Anything where the judgement is the value. Quoting a complicated job, handling an upset customer, deciding whether to take work on. You can automate the admin around these. Do not automate the decision.
• Processes that change every few months. Automation encodes assumptions. If the assumptions move constantly, you will spend more maintaining the automation than doing the task.
• Anything that happens twice a year. Frequency again. A checklist is the right tool here, and it costs nothing.
• The first human contact, unless you replace it with something better. Answering a new enquiry automatically is fine only if the automatic answer is genuinely useful — booking a slot, confirming a time. An auto-reply that says “we have received your message” is worse than nothing, because it spends the goodwill without doing the job. More on that in what a missed call actually costs.

For a typical service business, running mostly on email, a spreadsheet and an accounting package, this is roughly the order we would go in — easiest and most certain first.

First: the thing that chases people. Payment reminders, unsigned documents, unconfirmed appointments. High frequency, entirely rule-based, and nobody enjoys doing it, which means it is the task most likely to be skipped when the week gets busy. That is precisely why automating it produces money rather than just time.

Second: the thing that copies information. Enquiry into CRM, order into delivery note, booking into calendar. It is pure re-keying, it has clean edges, and the error cost is real. This is also the one that quietly reveals whether your data is in order — covered in seven signs your business has outgrown its spreadsheets.

Third: the thing that assembles reports. The Monday morning number-gathering. Once the data is landing in one place reliably, this stops being a task and becomes a screen. We wrote about what belongs on it in what a business dashboard should actually show.

Fourth, and only now: the thing that answers people. Out-of-hours enquiries, booking requests, triage. This is the highest-value automation in most service businesses and the one everyone wants to do first. It goes fourth because it needs the previous three working to be any good — it has to know your availability, write to your records, and tell someone when it cannot cope.

Before you switch anything on, write down two numbers: how long the task currently takes in a week, and how often it currently goes wrong. Ten minutes with a notepad is enough. Without them you will have an automation that everyone agrees is nice, and no case for the next one.

After a month, check three things: how many times it ran, how many exceptions it produced, and whether anyone has quietly gone back to doing it by hand. That last one is the honest measure. People do not file complaints about software they do not trust — they just stop using it.

If the exception rate is high, that is not a failure. It usually means the rules were more complicated than anyone admitted, and now you know exactly where. Fix the rule, not the tool.

Pick the task in your business that is most repetitive and least interesting. Write down what starts it, what happens, who handles the odd cases, and what done looks like. Time it for a week.

If the four answers come easily and the weekly total is more than an hour, you have found your first automation. If the answers do not come easily, you have found something more valuable: the process that needs fixing before any software touches it.