When NOT to use AI

There’s a version of AI consultancy that tells every client the same thing: yes, you need AI, here’s how we’d build it, here’s the price.

That’s not how Inzen works.

Some of the most valuable conversations we have start with us telling a business to slow down. Not because AI isn’t capable — it is — but because the wrong solution to the right problem is still the wrong solution.

So before we talk about what AI can do, here’s when we’d tell you not to bother.

When the problem isn’t defined clearly enough

AI doesn’t fix ambiguity. It amplifies it.

If you can’t describe the problem you’re solving in plain language — the specific workflow, the specific friction, the specific outcome you want — then you’re not ready to build anything yet. An LLM won’t clarify your thinking for you. It will produce confident-sounding output against a vague brief and call it done.

The first step is always problem definition. Everything else follows from that.

When the data isn’t there

Custom AI applications are only as good as the data they’re built on. If your processes aren’t documented, your data isn’t structured, or your systems don’t talk to each other — AI won’t solve that. It will inherit it.

We see this regularly. Businesses with years of institutional knowledge locked in email threads, spreadsheets and people’s heads. Before AI can help, that foundation needs work.

When a simpler tool would do the job

Not every automation problem is an AI problem. Sometimes a well-built workflow in your existing tooling, a properly configured integration, or a straightforward rules-based system is faster, cheaper and more reliable.

AI adds the most value where there’s genuine ambiguity, nuance or unstructured input involved. If the logic is simple and consistent, reach for a simpler tool first.

When the organisation isn’t ready

This one is underestimated. AI projects fail not because the technology doesn’t work, but because the people and processes around it aren’t ready to absorb the change.

This is consistently reflected in industry evidence, including McKinsey's State of AI findings on organizational readiness barriers and the NIST AI Risk Management Framework, which both stress governance, accountability and operational controls as prerequisites.

If there’s no internal champion, no clear ownership, or no appetite to adapt existing workflows, the project will stall after delivery. We’ve seen it happen. The tool gets built. Nobody uses it.

When the risk profile is too high for the maturity level

There are use cases where the consequences of AI getting it wrong are significant — regulated industries, patient-facing applications, legal or financial decisions. That doesn’t mean AI can’t be used there, but it does mean the bar for reliability, observability and governance is considerably higher.

For UK businesses handling personal data, the ICO guidance on AI and data protection is a practical starting point before committing to high-stakes deployments.

Rushing into a high-stakes deployment without the right architecture and oversight isn’t a shortcut. It’s a liability.

So when should you use it?

That’s a harder question to answer in a blog post — because the honest answer depends entirely on your specific situation, your data, your team and your goals.

What we can tell you is that the businesses getting genuine value from AI right now share a few things in common: a well-defined problem, a realistic scope and someone who understands both the technology and the business context well enough to bridge the two.

If you’re not sure which side of the line you’re on, that’s exactly the conversation we’re here to have.

Get in touch -> [email protected]

Let’s build something.

Inzen