The trainee test: how to pick the first job to hand to AI
When a small business tries AI and gives up, the tool usually isn't the problem. The first job was.
People tend to start with one of two things. Either something trivial ("rewrite this email"), which works but saves nothing worth talking about. Or something huge ("sort out our month-end"), which fails, because nobody could hand a job that size to a new starter either.
The jobs worth handing over sit in between. There's a simple way to find them.
The trainee test
Imagine a capable trainee starts on Monday. Bright, keen, knows nothing about your business yet. Would you give them this job in their first month, with a written brief and you checking the result?
If yes, it's a good candidate for AI. If no, ask why not. The reason is usually one of the five below, and it tells you whether to fix the job or pick a different one.
Five questions
1. Does it come round regularly? Weekly or monthly is ideal. A one-off job means writing a brief you'll use once. A weekly one means writing it once and reusing it fifty times.
2. Could you write it down as steps? If you can explain it to a trainee in a page, you can explain it to AI. If the honest answer is "it depends, I just know", the knowledge is in your head. Write it down first, because that alone is worth doing.
3. Is the material in one place? The job should work from things you can paste in or attach: an export, a set of notes, last month's version. If it means logging in to four systems and asking two people, start with a smaller piece of it.
4. Can someone check the result quickly? This matters most. A good first job is one where checking takes a few minutes and mistakes are easy to spot. A summary you'll read anyway passes. A figure that goes straight into a VAT return without anyone looking does not.
5. Is a mistake cheap? Early on, something will be wrong. Pick a job where a wrong first draft costs you a correction, not a customer or a fine. Internal reports and first drafts are good. Anything sent straight to a client or a regulator is not, yet.
What usually passes
Five yeses in a row are common in finance, admin and operations work:
- the weekly sales or cash summary for the managers' meeting
- turning meeting notes into actions and owners
- a first draft of the monthly management commentary from the numbers
- sorting and summarising supplier or customer queries before someone replies
- checking a list against a policy and flagging the exceptions for a person
Each one comes round regularly, can be written as steps, works from material you already have, and gets read by a person before it goes anywhere.
What usually fails, for now
- Anything with no checker. If nobody will read it before it's used, it's not a first job.
- Judgement calls about people. Hiring, pay, disputes.
- Jobs that are really ten jobs. "Do the month-end" fails. "Draft the commentary for the month-end pack" might pass.
"For now" matters. As the brief improves and you learn where the tool goes wrong, jobs move from the second list to the first. That's the point of starting small.
Pick one this week
Write down three jobs that come round every week or month. Run each through the five questions. Pick the one with the most yeses and the quickest check, and hand that over first, using a written brief like the one in the previous article.
Where are you now?
Choosing the job is only one of the four Ds. The readiness check below takes about three minutes and shows whether the job, the brief or the checking is where you're most likely to get stuck.