The AI projects I turn down, and why

I turn down more AI work than I take on, and I think that is a feature rather than a problem. Knowing which projects will not work is most of the skill, and a clear no early saves a client far more than my fee. Here are the ones I say no to, and what I suggest instead.

The idea with no problem attached

This is the most common one. "We want to do something with AI" is not a project, it is a wish. If there is no specific, painful, measurable problem underneath it, there is nothing to build towards and the work drifts. So I turn these down, or rather I turn them into a different conversation. Not "what could AI do here", which has a thousand answers and no priority, but "what is the most expensive repetitive thing your business does". Start there and a real project usually appears on its own.

The one where the numbers do not work

AI earns its keep on repetition. A process that happens rarely, or touches very few cases, will not save enough to justify building for it, however impressive the demo. It is the single most common way a tempting idea turns out to be a bad investment, and I wrote a whole piece on the filter I use for it in the three questions I ask before spending a penny. When a project fails the frequency test, I say so, because spending twenty thousand pounds to save two hours a month is a poor trade no matter how good the technology is.

A clear no is more useful than an expensive maybe.

The one with no consistent input

AI is very good at handling variation within a pattern and hopeless when there is no pattern at all. If every case in a process is a genuine one-off, with nothing shared for the system to learn from, it is a weak candidate however appealing it looks. I would far rather tell you that at the start than take the work and hand back something that shines in the demo and falls apart on real inputs.

The one where a mistake is expensive and nobody wants a human in the loop

Some processes can absorb the occasional AI mistake cheaply; a slightly awkward draft that someone reviews before it goes out costs nothing. Others cannot: a wrong figure in a filing, an offensive message sent straight to a customer. Those can still use AI, but only with genuine human checks designed in. If the appetite is "fully automated, no oversight" on a high-stakes process, I turn it down, because that is not a risk I am willing to take with your business on your behalf.

The one that is not ready

Sometimes the idea is sound but the ground is not prepared. The data is a mess, no one really owns the process, or there is no decision-maker who can actually say yes and make it stick. AI does not fix an unclear process, it automates it, mess and all, and then you have an automated mess. When something is not ready, the honest move is to fix the process first and add AI second, not to paper over the cracks with a clever tool.

The one that genuinely needs a team

My model is one senior operator with AI leverage, which I call AI-Leveraged Delivery, and it fits a great deal of AI implementation but not everything. A programme that truly needs many specialists working in parallel, or heavy on-site presence, is not a fit for how I work. When that is the case I say so and point you towards the right kind of team, rather than stretching to take work I am not the best answer for.

Why saying no is worth more than saying yes

Every one of these, taken on regardless, tends to end the same way: money spent, little changed, and often a pilot stuck in Pilot Purgatory. Turning them down early costs the client nothing and saves them a great deal. It is also, frankly, part of why the projects I do take on tend to work. They have been through the filter. If I say no to yours, I will tell you exactly why and what would turn it into a yes, because that is genuinely more useful than a polite maybe and an invoice.

What a good one looks like

The projects worth doing share a shape. A real, frequent, well-defined problem. Consistent inputs. A sensible attitude to the cost of a mistake. And a business that is ready to actually use the result. When yours has that shape, the work is a pleasure and it pays for itself quickly. When it does not, the most valuable thing I can do is tell you, early and plainly. That is worth more than my fee, and on most days it is most of the job.