Where should an SME start with AI? The five-step method

More than half of French very small and small businesses already use generative AI, but only one in six uses it regularly (Bpifrance Le Lab, January 2026 barometer). The gap has nothing to do with technology: it comes down to the starting point. Here is a simple method, which we apply to ourselves.

Step 1 — Start from lost time, not from the tool

Do not look for "what AI can do". Look for what costs you hours every week: entering invoices, chasing customers, sorting emails, preparing quotes, answering the same questions.

Make the list with the people who do the work. Urgent tasks are not necessarily the heaviest: a monthly close that takes a whole day weighs more than a ten-minute emergency.

Step 2 — Put a figure on it, even a rough one

For each task, two questions are enough: how often, and how long each time?

Example: a fifteen-minute customer reminder, done 20 times a week, adds up to 5 hours a week — more than half a day.

This figure has two uses: it points to the most profitable candidates, and it sets the starting point without which no gain can be measured.

Step 3 — Choose the first task with five criteria

The best first task is not necessarily the heaviest. It is the one that ticks these boxes:

1. It comes round often, in a comparable form.
2. The information is accessible in a tool.
3. An occasional mistake is not serious, or is easy to put right.
4. The gain can be measured: we know what we will count.
5. We can go back without damage.

We go through these criteria in Which processes should you hand to an AI agent first?

Step 4 — Hand it over, then measure at 30 days

A simple task — a trigger, an action, a result — goes live in one to two weeks, in your current tools. Agree on the indicator before starting, record it, then record it again thirty days after going live.

That is the format of our Quick Start: a fixed price, a task, a measured gain.

Step 5 — Expand only if the result justifies it

If the first task delivers, look at the process it belongs to: from the invoice received to payment, from the customer request to its resolution. That is where the bigger gains are. If it does not, you have learned at little cost.

The three most common mistakes

• Starting with the tool: buying licences before knowing what work to hand over.
• Aiming too high: a six-month project that shows nothing until the end.
• Measuring nothing: without a starting point, there is no way to know whether the agent pays off.

And if you do not know where to start?

That is exactly what the Flash Diagnostic is for: a ten-minute questionnaire, a one-hour interview, then within 72 hours a written report with 3 to 5 leads quantified in time saved and one recommendation, a single one. It is free, and it may conclude that an off-the-shelf tool is enough for you.

Sources

• Bpifrance Le Lab, January 2026 barometer of French very small and small businesses: 55% use generative AI, 17% regularly.

Frequently asked questions

Do we need an AI strategy before starting?

Not for a first step. A well-chosen task, measured before and after, teaches you more than a strategy written in advance. Strategy comes later, when several agents have to work together.

How long does a first result take?

For a simple task, one to two weeks from scoping to going live. For a complete process with its decisions, allow about three weeks.

Is our data ready?

Often enough for a first task. If the information is scattered or only in someone’s head, it is better to start elsewhere, or to gather the data first.