Insights
Start with the strategy you already have
Most AI programmes begin in the wrong place. The fix is not a better tool. It is a better starting question.
Most of the AI work we are asked to look at begins the same way. Someone has seen what the technology can do, felt the pressure not to be left behind, and started asking how to use it. The question sounds reasonable. It is the wrong one, and it quietly shapes everything that follows.
Starting from the technology means working backwards to a justification. A capability goes looking for a problem. Pilots multiply, each defensible on its own terms, and none of them clearly tied to what the organisation was already trying to achieve. A year later there is activity to point to and very little that has changed. The programme was busy. It was not aimed.
The better starting question is the one the organisation should already be able to answer. What are we trying to do over the next few years, and what stands in the way. Strategy is not something to be reverse-engineered from a technology. It is the thing that should decide whether the technology is worth touching at all.
The order matters more than the tooling
When you begin from strategy, the sequence changes. You name the direction first. You are honest about the constraints, meaning the budget, the people, the data you actually have rather than the data you wish you had, and the tolerance for disruption in a business that still has to run while you change it. Only then do you ask where technology genuinely helps, and you evaluate each option against that direction rather than against a demonstration.
This sounds obvious written down. It is rare in practice, because starting from the technology is easier and more exciting, and because a market full of noise rewards the appearance of motion. The discipline is in resisting that, and in being willing to conclude that a great deal of what is on offer does not serve where you are trying to go.
Constraints and appetite are inputs, not obstacles
There is a habit of treating constraints as problems to be argued away. We think they are the most useful information you have. An organisation with limited engineering capacity and a low appetite for operational risk should make very different choices from one with the opposite profile, even if both are in the same industry facing the same technology. Advice that ignores this produces roadmaps that look impressive and never survive contact with the real business.
Appetite deserves the same honesty. Some leaders want to move early and can absorb the cost of being wrong. Others cannot, and should not pretend otherwise. Neither is the correct answer. The correct answer is the one that fits, and fit is only visible once the strategy, the constraints, and the appetite are on the table together.
What good looks like
A programme that starts from strategy is easier to explain and easier to stop. Each piece of work has a reason that predates the technology, so it is clear when something is working and clear when it is not. The roadmap becomes a sequence of decisions tied to the direction of the business, rather than a list of tools acquired in case they prove useful.
It also produces something most AI advice leaves out, which is a clear view of what to ignore. Knowing where not to spend is not caution for its own sake. In a market this loud, it is most of the value.
AI is a support, not a goal. Start from the thing it is meant to support, and the rest of the decisions get simpler.