Skip to content
Nominal FR Let's talk about your project

AI in a small business: where to start

AI is already in the company, often without leadership knowing. The question is no longer whether to get started, but how to do so without scattering data or effort.

Driss Lebrun · October 2026 · 6 min read

01

Start from the work, not the tool

The temptation is to choose a tool, then look for what it could be used for. The reverse order works: spot the tasks that take time, repeat and follow rules, then see whether AI can lighten them.

Sorting and summarising incoming mail, preparing a first reply, extracting information from a document, preparing a meeting from previous exchanges: these are modest uses, but they give back hours every week.

02

A tool is only worth what it is connected to

An isolated AI produces text. An AI connected to the company's data and to those who decide produces useful work. That is the whole point: what information it has access to, who reviews what it proposes, and who keeps control of what goes out.

A simple rule avoids most problems: AI prepares, a person approves. Nothing goes to a client, a supplier or a partner without a human having read it.

03

Set the framework before generalising

Three questions are enough to set the framework. Which data can be entrusted to which tool, and where is it hosted? Who has access to what? How do we measure that the use really saves time or improves quality?

Without this framework, uses spring up everywhere, each with its own account and habits, and the company's data disperses into services that nobody controls. With it, each new use adds cleanly to the previous ones.

04

Start small, measure, extend

The right first project is short, visible and measurable: one team, one use, a few weeks. Measure the time saved and the quality achieved, adjust, then extend. A modest, proven success is worth more than a large programme that promises everything.

The systems built in this way must belong to the company: documented, transferable, and independent of any single supplier. That is the condition for AI to remain a lever, and not become one more dependency.

Perspectives

Read next.

Next step

Does this speak to you? Let's talk.

Let's talk about your project