Adopting AI gradually: where should you start?

If you want to bring AI into your organisation, the best starting point is probably not to make it available to every employee and hope for significant, measurable results in the short term. The key is to identify one or a few initial workflows that are narrow in scope, repetitive and burdensome. To choose them, observe your teams’ actual work and talk to experts across different roles about the recurring or time-consuming tasks they handle every day.

Why starting too big can lead you astray

There are at least two critical points to keep in view when introducing AI: responsibility for decisions and the preservation of knowledge.

For each new use case, both deserve to be questioned. What role do our experts play in this workflow? Who validates the results produced automatically? When is knowledge shared with our experts so they retain command of their domain and the decisions it involves?

A business relies on people who know their work and take responsibility for their decisions. Their skills and knowledge are what allow them to make informed choices.

Another difficulty can arise when adoption is rushed on a large scale: how do you assess the results against the costs involved? “Everyone uses it” does not mean “Everyone works faster” or “Everyone is more productive”. When everyone starts using AI overnight, costs can rise quickly without a clear understanding of what it brings to the teams’ work.

Of course, there will still be many occasional uses that save time without necessarily being measurable. In the same way, performance gains are not always measured after adopting specialised software: if it helps experts in their daily work, that can be enough to justify using it. Identifying recurring workflows, sometimes shared by several people or teams, does, however, make it possible to improve more structured uses and measure their effects.

Start with real work

There are several ways to identify a workflow that AI could speed up, or at least make smoother. You can start with a clearly identified recurring task, but you can also design a new workflow around existing constraints and needs. Knowing what a technology can do also leaves room for creativity!

A simple first approach is to choose a recurring, burdensome task with a limited scope. This is often a task where doing it manually adds little value.

Whatever scope you choose, introducing AI must serve an observed need. It must also answer the two questions raised earlier: who remains responsible for decisions, and how do teams preserve their knowledge?

Keep people at the heart of the solution

When you entrust your personal finances or insurance matters to someone, you expect that person to know their field and be able to advise you. In the same way, handing a critical business activity entirely over to AI, without an expert to supervise it, can be counterproductive or even dangerous.

The risk is that nobody in the business retains enough command of that activity to understand a problem, fix it or prevent it from happening again. AI can then take on a growing role while the skills needed to keep it under control weaken.

This is an important subject and deserves an article of its own!