AI Opportunity Check
Pick one job somebody in your business does by hand. Eight questions, and a straight answer on whether it is worth automating. Four of the six possible answers are reasons not to build anything.
Have a specific job in mind first. “Our admin” is too broad to answer honestly; “sending the same quote template after a WhatsApp enquiry” is the right size.
How this works
No score. Six conditions, checked in a fixed order, stopping at the first one that matches. A points total would average away the thing that decides these questions: a job can happen fifty times a week and still be the wrong candidate, because every one of those fifty is slightly different.
The six rules, in the order they run
The order is not arbitrary. It runs from the cheapest reason to stop to the most expensive. There is no point weighing up how automatable something is if it barely happens, and no point discussing exceptions if getting it wrong could hurt somebody.
- The job happens a few times a year, or happens monthly and takes under fifteen minutes.Not worth automating. The time saved will not repay the cost of building and maintaining it.
- Getting it wrong unnoticed for a week would cause something serious, and judgement is involved.Automate the preparation, not the decision. A person stays in the loop.
- The rules cannot be written down for a new starter, or most cases are exceptions.Write it down and simplify it first. There is nothing settled enough to automate yet.
- Nobody would own it once it exists.Find an owner first. Unowned systems fail quietly.
- It happens at least weekly, inputs always arrive in the same shape, the rules are fully writable, exceptions are not the norm, and no judgement is needed.A strong candidate. Build it properly.
- Anything else.Worth doing in a narrower version: automate the predictable core, send the rest to a person.
Your result names which of these six applied and quotes back the answers that triggered it, so you can check the tool did what this list says. Change one answer and you should be able to predict what it says before you see it. If you cannot, the fault is ours and we would like to know.
The eight questions (8 questions)
Published here so the whole thing is readable without using it, and so it still works as a thinking checklist if you would rather answer these in a meeting than on a screen.
How often does this job happen?
- Several times a day
- Weekly, or a few times a week
- Monthly
- A few times a year
How long does it take each time?
Count the interruption as well as the task. A five-minute job that pulls somebody out of something else costs more than five minutes.
- More than an hour
- Between fifteen minutes and an hour
- Under fifteen minutes
Does the information you start with arrive in the same shape every time?
A form with the same fields is the same shape. A WhatsApp message where the details are wherever the customer happened to put them is not.
- Always, it comes from the same place in the same format
- Usually, with some tidying up
- Rarely, it arrives differently each time
Could you write the rules down well enough for a new starter to follow them without asking?
- Yes, completely
- Mostly, but they would still need to ask about some cases
- No, you have to know the business to do it
How often does something come up that does not fit the normal way of doing it?
- Rarely, nearly every one is the same
- Sometimes, maybe one in five
- Most of them are a bit different
Does somebody have to use judgement about a person or a situation?
Deciding whether a customer is upset, whether to make an exception, or whether something looks wrong, all count as judgement.
- No, it is the same decision every time
- Sometimes, for the awkward ones
- Yes, every time
If this were done wrong and nobody noticed for a week, what would happen?
- Some annoyance, and we would sort it out
- We would lose money, or lose a customer
- Something serious: money out the door, a legal or safety problem, real harm to somebody
Once it exists, who looks after it when it breaks?
Everything built eventually breaks, usually because something it connects to changed. This question is about who notices and who fixes it.
- A specific person, and they know it
- Somebody would, roughly
- Nobody, honestly
What this cannot tell you
It cannot tell you what the work would cost, because that depends on what the job has to reach and how tidy those systems are. It cannot see the constraint you did not mention, and it takes your answers at face value: a process that feels predictable from a distance often is not once somebody writes it down, which is why rule 3 exists. And it says nothing about whether AI specifically is involved. Plenty of the jobs that come out of this well need a rule and a notification, not a model.
Where to go from here
Where do you want to start?
A website that brings in no enquiries. A business nobody finds on Google. The same job done by hand every week. Tell us which one it is.
Have a project in mind? Send an enquiry.
- Free 30-minute call
- No obligation
- Talk to a person, not a bot
- We publish our prices