Where to start automating: the four tasks that pay back first

Ask ten business owners what they would automate first and most will name the thing that annoys them most. That is rarely the same as the thing that costs most. Annoyance and cost are different measurements, and only one of them shows up on the P&L.

Here are the four tests we apply before recommending anything. A task that passes all four is worth automating now. A task that passes two is worth a note and nothing more.

Test 1: does it repeat on a schedule you can name?

"Every time an order comes in" is a schedule. "Whenever a customer needs something unusual" is not. The first can be written down as rules; the second needs judgement, and judgement is what you want your people spending their day on.

This test alone eliminates most of what people ask us about first. Handling an angry customer feels expensive because it is stressful. It is not repetitive, so it is not the place to start.

Test 2: does it cost more than you think?

Do this on paper before anyone quotes you anything. Take one task and answer three questions:

• How many people touch it?
• How many minutes does each spend, per day?
• What does an hour of their time cost you, fully loaded?

Multiply, then multiply by working days. The number is usually larger than expected, for a reason worth understanding: the cost is spread across several people in small slices, so nobody experiences it as expensive. Fifteen minutes a day feels like nothing. Across six people and a working year it is not nothing.

If the arithmetic gives you a number you would not pay a contractor to make disappear, do not automate it. Move on to the next task.

Test 3: what does it cost when it goes wrong?

This is the test people skip, and it is often the one that decides the ranking. Manual data entry is not expensive because it takes time. It is expensive because a transposed digit in an invoice or a mistyped email address costs a customer, and you usually find out weeks later without connecting the two.

Ask of each task: what happens when someone gets this wrong, and how long before we notice? The longer the gap between the error and the discovery, the more automation is worth — not because software does not make mistakes, but because it makes the same mistake every time, which means you find it once and fix it once.

Test 4: is the input already structured?

A task whose input arrives as a form, a spreadsheet row or an API call is cheap to automate. A task whose input arrives as a phone call, a photo of a receipt or a WhatsApp message that starts "hey, quick question" costs several times more, because something has to turn the mess into structure first.

That does not mean the messy ones are off the table — turning unstructured input into structure is exactly what the current generation of AI is good at. It means they are not where you start. Start where the payback comes in weeks, prove the approach works, then spend that credibility on the harder one.

The four tasks that usually win

Across the businesses we look at, the same four keep coming out on top. Not because they are exciting, but because they pass all four tests:

Task	Why it wins
Moving data between two systems that do not talk	Perfectly repetitive, structured input, and errors surface late
First-response follow-up on new enquiries	Repetitive, and the cost of being slow is a lost sale you never hear about
Recurring reports someone assembles by hand	Scheduled, structured, and it eats senior time rather than junior time
Keeping the CRM honest	Cheap per instance, ruinous in aggregate: every downstream decision uses it

What not to do first

Do not start with the customer-facing thing. It is tempting — it is visible, it demos well, and it feels like progress. It is also where a mistake is seen by the people whose opinion you can least afford to lose.

Start somewhere internal, where a failure costs you an afternoon instead of a customer. Once it has run for a month without anyone thinking about it, you have earned the right to point it at the people who pay you.