The No List
When not to use AI.
As far as we can tell, no UK consultancy has published this page. Plenty say they will tell you when AI is wrong for you; none of them say in advance what wrong looks like. Here are our rules, written down, so you can hold us to them.
The nine rules
If your idea trips one of these, we will say so on the free call rather than three weeks into a paid engagement.
The answer has to be exact and auditable
Invoicing, payroll, VAT, statutory reporting, anything a regulator or an auditor will pull apart. These need arithmetic that is right every time and can be traced. A model that is right 97% of the time is not a saving, it is a liability with a delay on it.
Ordinary software already does it well
A great deal of what gets pitched as AI is a database query, a workflow rule or a properly configured integration. We will name the boring solution, because it is cheaper to build, cheaper to run and it does not drift.
A licence you already pay for covers it
Microsoft 365, Power Automate, features dormant in your CRM, capability already bundled in your line-of-business system. We host and support your estate, so we usually know what you are paying for. Frequently the honest recommendation is “turn this on”.
Your data is not good enough yet
Too scattered, too inconsistent, too undocumented. Feeding poor data to a model does not produce insight, it produces confident nonsense — and it will be believed, because it is fluent. Fix the data first; that is often the whole project.
Nobody stays in the loop
If a consequential decision would go out with no person accountable for it, the answer is no — regardless of how good the model is. Not because the technology cannot, but because you cannot defend it afterwards.
The volume is too low to repay the build
Our rough threshold: if the same task takes less than about two hours a week across the business, automating it will not repay building it — and it certainly will not repay maintaining it. Maintenance is the cost everyone forgets.
The process is undocumented because it is judgement
Some processes are not written down because nobody has got round to it. Others are not written down because every case genuinely turns on experience. The first is a documentation job. The second is not an automation candidate, and telling them apart is most of the value of a Verdict.
Automating it removes the reason people buy from you
Nearly six in ten UK small firms that have not adopted AI say their business is built on close personal relationships. They are usually right. If the human contact is the product, automating the contact does not cut cost — it cuts revenue, on a lag, and you will blame something else.
You are doing it because someone asked what your AI strategy is
A board question, a competitor’s press release, a post on LinkedIn. All perfectly understandable, none of them a business case. If this is the real driver, the cheapest honest deliverable we can give you is a well-argued page explaining why you are not doing it yet.
So when is the answer yes?
To be clear, we are not against any of this. We build AI features ourselves — our own fitness platform uses them, and they are the only thing we charge for in it. The yes cases have a recognisable shape.
Good candidates
- High volume, repetitive, and someone currently does it by hand
- Judgement is genuinely light — the task is transcription, extraction, classification or drafting
- Being occasionally wrong is survivable, and a person checks before anything irreversible happens
- The input already exists in a consistent form: documents, photographs, transcripts, structured records
- You can measure the before and the after, so payback is a fact rather than a feeling
- The freed-up time has somewhere to go that you have actually decided on
Why we publish this
Because a promise you cannot check is worth nothing. “We will tell you if AI is not right for you” is now said by roughly a third of the UK market, and almost none of them will tell you in advance what their criteria are.
These are ours. If you bring us something that trips one of these nine rules and we take your money anyway, you have this page to point at.
Not sure which side of the line you are on?
That is exactly what the free 30-minute Reality Check is for. Bring the idea, we will tell you which rule it trips, or that it does not trip any of them.