Where do I start with AI? Three marks of a good first task

There is a point where a lot of people get stuck. You have heard enough to suspect there is something in it. You may have tried it twice. And then it sits there again.

The reason is rarely scepticism. Most often it is the choice: people begin with the wrong task.

Start with something important, unfamiliar and delicate, and you get a mediocre result – exactly where it ought to be right. After that, AI gets left alone. So it pays to choose the first task deliberately. Three marks are enough to spot it.

The question before all others

"Is AI worth it for me?" is the most common opening question – and it is too big a question. Put that way it can honestly only be answered with "it depends", which helps nobody.

The more useful question is: is AI worth it for this one job I am doing this week anyway? That has an answer, and it arrives in twenty minutes rather than in a strategy paper.

The big question calls for a forecast about your whole business. The small one calls for an attempt. And out of five or six such attempts, the answer to the big question emerges by itself – based on what you have seen, not on a guess.

Frequent, annoying, harmless

Some jobs suit a first attempt considerably better than others. Three marks make the difference.

Frequent. The job comes round regularly – weekly, better still daily. That matters more than it sounds: the first time, some effort goes into finding your way. On a job that comes round once a quarter, that effort evaporates. On a weekly one it pays back from the third round.

Annoying. The job costs you effort, time or both. That might be texts you find hard going; summaries nobody enjoys writing; sorting through notes. Where things already run smoothly there is little to gain – and the difference barely registers.

Harmless. The job survives a mediocre first attempt. An internal draft, a piece of preparation, a collection of ideas. Something you would look over anyway before it goes anywhere. That takes the risk out of the attempt – and the pressure off you.

Together the three make a filter that runs in a minute: frequent, annoying, harmless. Run through your last few working days in your head and look for the job all three apply to. Usually it puts its hand up straight away.

If several qualify, the third mark decides. Take the most harmless one – the learning is the same, the exposure the smallest. The important jobs come round by themselves once you know what the tool delivers and where you have to sharpen up.

The first attempt

For the chosen job, a request that describes the situation and fixes the result is enough. You type it into the input box of your AI assistant. This frame carries almost any first task:

For the follow-up message from the example, that means: role – an experienced customer contact. Result – a short message that follows up an open proposal in a friendly way. Situation – proposal for a company party, sent a fortnight ago, no reply since, the date is getting closer. Form – four sentences, friendly, no pressure.

Compare the result with what you would otherwise have written. That single comparison says more about the benefit than any testimonial – because it concerns your job, your tone and your standard.

For the second attempt the fuller frame is worth it: Writing good prompts means filling in four fields sets out which four details decide the outcome, and how to spot the one that is missing.

What happens in the first week

Usually the first attempt already turns out usable, and the next ones turn out better. In between sits the thing that makes the difference – you notice which details were missing and supply them next time round.

The time saving usually arrives later than the relief. What you notice first is that something goes more easily, that starting costs no effort. The minutes saved follow once the routine settles.

And something else shows up quickly: jobs you would never have considered in this context. Once you have watched an awkward message come together in two minutes, you look at next week differently.

The most revealing spot is where you change the draft. That is where your expertise sits – what the AI did not yet know about your business, your customers and your tone. Make a habit of writing those changes into the request next time and the path gets shorter with every round.

After five or six such jobs you hold something no brochure supplies: your own experience of where this tool carries your work and where it lets you down. That is the answer to the big question from the start – only evidenced.

When waiting is the wiser choice

There are cases where holding off pays, and it is only fair to mention them.

When your work is almost entirely hands-on. Anyone who spends the day at the machine, with patients or on site writes little. There AI helps more at the edges, for instance with proposals or documentation. That can be worthwhile without turning anything upside down.

When it is not yet settled which AI tool you use for client and business data. That gets decided once, in principle. Until then, practise on tasks without names, addresses or other confidential details.

When you are in a peak workload right now. Learning something new needs a little calm. Two quiet weeks later the start comes easier than in the middle of the storm.

In every other case the effort for a first attempt is modest: one job, twenty minutes, one comparison. Little enough to do on any given Tuesday morning – and telling enough to know more afterwards than before.

The next step

One job, twenty minutes, one comparison – that is all the beginning needs. After that the big question answers itself, and it does so for your business rather than in general.

If you would like a head start: the AI Skill Level Check asks five questions and is free. You leave your email address and receive a practical guide matched to your answers. Marketing emails only arrive if you tick that box separately.