The work AI should take off your plate first

The first question most owners ask is what the impressive thing would be — the part of the business AI could take over that would make a good story at a dinner. It is the wrong place to start. The work worth handing over first is almost always the dullest thing on your list.

On the consultation we walk through a week of your actual work and put each recurring task against three questions. The ones that pass all three are what the summary recommends building first.

• Does it repeat? Something that happens once a quarter costs more to automate than to keep doing by hand.
• Could you write it down? If you could explain the steps to a new hire in a paragraph, a system can follow the same paragraph.
• Would you notice if it were wrong? Work with a visible output — a reply you can read, a draft you approve — is safe to hand over. Work that fails quietly is not.

Three yeses and it is a candidate. One no and it waits, however appealing it sounds.

It is rarely the strategy, the pricing, or the part of the job you trained for. In most businesses the first candidates look like this:

• The first reply to an enquiry — the one that decides whether someone waits for you or calls the next name on the list.
• The follow-up nobody gets to: the second message, a few days later, to a person who did not answer the first.
• The drafting — the same email in twenty variations, the listing description, the caption, the recap of a call.
• Moving information from one place to another: form to spreadsheet, spreadsheet to invoice, inbox to calendar.

None of it is interesting. All of it is on your evening.

The first thing to hand over is the thing you do the same way every time.

Some work should stay yours, and saying which is part of the audit. Judgment calls — what to charge, who to take on, when to say no — run on things that are not written down anywhere, including your read of the person in front of you. The relationship is the other one: the call where somebody needs to be heard is not a workflow, and treating it like one is how a business loses the customers it already has.

We also leave anything where a mistake would be both expensive and invisible — money moving, compliance, anything a regulator would want a record of. Those can come later, with a person in the middle, once the boring pieces have earned some trust.

The impressive build is the one you can least easily check. It touches more of the business, it is harder to explain, and when it goes wrong you tend to hear about it from a customer rather than from the system. A dull build does one thing whose output you can read the same day, which means you know quickly whether to keep it.

There is a second reason. Once a small thing runs without you, you get a clear look at the next bottleneck — and it is often not the one you would have picked beforehand. Building in that order costs less and tells you more.

You do not need to prepare anything for a consultation. But if you want a head start, look at last week rather than at a typical week: write down what you actually did after five o'clock, then mark the lines that were the same as the week before. That list is where we start, and it is usually shorter and more boring than people expect.