How to spot which job in your business AI should do first
The hard part of AI is not the tech, it is knowing where to point it first. Here is how to find the job that pays off fastest.
You know AI could help. You just cannot work out where to start. So you keep meaning to, and somehow never do.
That stuck feeling is not a sign you are behind or slow. It is the normal place everyone stands at the beginning, looking at a tool that could do almost anything and freezing on exactly that.
And it matters which way you jump. Pick something flashy that does not really move the needle, watch it underwhelm, and you quietly write the whole thing off.
The skill that separates people who get value from AI from people who do not is unglamorous: choosing the right first job. Get it right and your first attempt works, you trust it, you do more. I work alongside these tools every day, and this choice is where most of it is won or lost. So here is how to choose well.
Look for the task that is repetitive and dreaded
The best first job is usually the one you sigh about. The weekly report nobody enjoys. The same enquiry answered fifty times. The data copied from one place into another.
Repetitive, rule-ish, and a bit soul destroying. AI is good at exactly the work humans find tedious, which is a gift, because that work is everywhere.
Ask yourself: what do I, or someone on my team, do every week that is repetitive, takes real time, and does not need much judgement? That is your shortlist.
Score it on time, frequency and pain
For each candidate, ask three questions. How long does it take each time? How often does it happen? How much does everyone hate it?
A job that is slow, frequent and dreaded is gold. A job that is slow but happens once a year is not worth automating yet. This simple scoring stops you chasing the exciting idea and points you at the boring one that quietly gives you hours back.
There is a real engineering reason the repetitive, clearly described task is the right first one, and it is not just that it is easy. A task you can spell out in a few steps can be pinned down with a tight system prompt, the standing instruction that tells the model exactly what good looks like, and wrapped in guardrails and a validation layer that check each output before it reaches anyone. Frequent work is also where you get enough examples to do few-shot prompting, showing the model a handful of past cases so it copies the pattern instead of guessing. The high-stakes, fuzzy, once-a-year job is the opposite: thin on examples, hard to specify, expensive when it drifts, exactly the territory where a probabilistic tool is least dependable. Pick the task whose rules you can actually write down, and most of the reliability problem is solved before you start.
Avoid the high-stakes job for your first go
It is tempting to aim AI at the most important, most sensitive thing you do. Resist that for the first project.
You want an early win where a mistake is cheap and easy to spot, not one where an error costs you a client. Build trust on something low risk, then move up.
Pick the one you can describe clearly
If you can explain a task to a new starter in a few clear steps, AI can probably help with it.
If even you are not sure how the task really works, that is a sign to fix the process first, not to throw AI at the confusion. Clarity in, value out.
Choosing the right first job is most of the work, and it is the part people skip. Spend ten minutes on the choice and your first AI project is far more likely to be the one that sticks, and the one that makes the next ten obvious.
If you can feel there is time to save but cannot pin down where to start, helping you find that first high-value job is exactly where we begin.
Book a quick chat →Related: How to know what to hand to AI, and what to keep.
Common questions
What should I use AI for first in my business?
Start with a task that is repetitive, frequent and a bit dreaded, where a mistake is cheap to catch. That combination gives you a quick, low risk win you can trust, which makes everything after it easier.
How do I know if a task is right for AI?
If you can describe it to a new starter in a few clear steps, and it does not need much human judgement, it is a good candidate. If even you cannot explain how it works, fix the process before bringing in AI.
Why do AI projects fail?
Often because the wrong first job was chosen, something too vague, too high stakes, or too rare to be worth it. The technology is rarely the problem. Picking the right place to start is the real skill.