Why we'll tell you not to use AI

There is a version of an AI consultancy that is very easy to run. A client asks “should we use AI for this?” and the answer is always yes. Yes to the chatbot. Yes to the copilot. Yes to the agent. Every yes is another engagement, and the incentive structure quietly rots the advice.

We don’t run that version. Here is why, and here is what “no” actually looks like.

AI is not free, and it is not always cheap

The headline cost of a model call is small. The real cost is everything around it: the integration work, the data plumbing, the evaluation harness, the monitoring, the human review queue, the retraining when the world shifts. For a high-value, high-volume process, all of that is worth it many times over. For a process that runs forty times a month, it is often a six-figure solution to a four-figure problem.

When we assess a project, we cost the whole system, not the API line item. Sometimes the honest answer is that a spreadsheet and a clear owner will outperform anything we could build, for a tenth of the price. We will say so.

Some work demands a human, and always will

There is a category of work where the entire point is that a person is accountable for the judgment. High-stakes, low-frequency decisions. Situations where the cost of being confidently wrong dwarfs the cost of being slow. Relationships where the other party needs to know a human weighed it.

You can put AI around that work — preparing the brief, surfacing the precedent, drafting the first pass. But automating the judgment itself doesn’t remove the human-in-the-loop requirement; it just hides it until the day it fails an audit. We will tell you which parts of a process are genuinely automatable and which are load-bearing human judgment dressed up as a workflow step.

The pilot that should never have been funded

The most expensive AI project is the one that gets six months and a team before anyone asks whether it should exist. By then there is sunk cost, there are reputations attached, and “kill it” has become politically expensive.

Our assessment work exists to move that decision to the front. Before budget is committed, we give you a written verdict: fund, delay, or kill — with the reasoning. The projects we recommend killing are not failures of imagination. They are the reason the projects we do back can get everything they need.

What “no” is actually worth

A consultancy’s “yes” is cheap — everyone selling AI says yes. The “no” is the expensive part to give, because it costs us the engagement. That is exactly why it is the most valuable thing on the menu. It means that when we say a project is worth everything you can give it, you can trust that the sentence cost us something to say.

Radical honesty isn’t a personality trait we’re advertising. It’s the only way advisory work is worth paying for. If we’ll tell you not to use AI, you can believe us when we tell you where it changes everything.