New:The AI Integration Day: one day at your office to put AI to work.The AI Integration Day

Discover

Where should an SMB start with AI?

Start with a problem, not a tool: a task that comes back every week, takes time and has a result that's easy to check. Here's how to choose that first project, what you can have in hand quickly and how to avoid the pilot project that goes nowhere.

Par où commencer avec l'IA dans une PME?

Everyone is talking to you about AI. Your software vendor, your accountant, half your LinkedIn feed. You know you should do something. The real question is what.

You may have already seen the scenario we want to avoid: a pilot project that runs for a few months, a nice demo at the end, polite applause. Then nothing. The pilot sleeps in a corner and the team goes back to its Excel files.

Good news: in 2026, getting started with AI doesn't take a big budget or a team of in-house specialists. Mostly, it takes choosing the right first problem.

Where should an SMB start with AI?

With a problem, not a tool. Rather than looking for where to put AI, look for the task that costs you the most time for the least value. Then ask yourself whether AI can take it on.

A good first problem ticks 3 boxes: it comes back often, it takes time and its result is easy to check. If a person can look at AI's work and say within seconds “that's right” or “that's not right”, you have a candidate.

One example: preparing quotes. They come back every week, they take time to put together and the estimator can tell right away whether a price makes sense.

Not everything deserves to be automated, and not everything needs AI. Sometimes the right answer is a simple form or software that already exists. That's fine too.

Fictional example: 4 tasks run through the 3 criteria. The right first project is the one that ticks all 3 boxes.

How do you recognize a good first AI project?

It's often boring. The best first project impresses nobody in a demo, but it removes a chore your team does every week.

Save the big projects for later: the platform that will connect everything, the assistant that knows the whole company. They'll come much more easily once the team has seen a first project actually work.

Here are the signs that don't lie.

  1. Someone bridges 2 tools

    A person retypes information from an email into software, or from one piece of software into another. That's exactly the kind of handoff AI does well.

  2. The task comes back every week

    A task done once a year isn't worth the effort. A task done dozens of times a week is.

  3. The rules already exist

    The estimator knows how they build a price. The person at reception knows where to send each request. If the rules can be explained, they can be handed over to AI.

  4. Someone has had enough

    The best source of ideas is the person who does the task. If they sigh while explaining it to you, take notes.

What can you have in hand quickly?

Much more than a few years ago. With AI, a prototype takes a few days instead of a few months, and a simple app costs from $5,000.

For a first project, that means concrete results, fast. An automation that prepares quotes in your usual software. An assistant that answers the team's questions from your documents, with sources. A small tool that replaces the shared file everyone edits at the same time.

What takes time is no longer writing the code. It's choosing the right problem, and that's exactly the part that's up to you.

Fast doesn't mean sloppy. AI speeds things up, engineering keeps the software solid: access, data security and the link with your systems are planned from the start, even for a small project. For pricing details, see How much does custom software cost?

Illustrative: AI shortens the path to the first trial. The work that makes the tool reliable doesn't disappear.

How do you avoid the pilot project that goes nowhere?

By running it in your real tools, on your real data, with the people who will use it. A pilot that doesn't touch your real tools proves nothing: it shows that AI works in general, not that it works for you.

Conversely, a good pilot is small, connected to a single real process, owned by someone on the team and measured before and after. Less impressive in a demo. Much more useful on Monday morning.

Pilots that end up in a drawer often look alike.

  1. The demo on test data

    Everything works with 10 clean examples. Real files are incomplete, poorly sorted and full of exceptions.

  2. The tool beside the work

    If the team has to open a new screen and retype information into it, the pilot adds a step instead of removing one.

  3. Nobody owns it

    A pilot that nobody on the team carries fades out as soon as the next project comes along.

  4. No starting measure

    Without knowing how long the task took before, there's no way to tell whether the pilot succeeded.

  5. Too big to finish

    A pilot that tries to solve 5 problems at once solves none. One process, from start to finish.

How do you know if the first project worked?

By measuring before you start. If you don't know how long the task takes today, you'll never know what AI saved you.

Write down the time spent, the number of files per week, errors and rework, the delay before a customer gets an answer. For example, if preparing a quote takes 45 minutes and the team prepares 15 a week, that's more than 11 hours a week. It's a hypothesis: redo the math with your real numbers, as we explain in 5 steps to automate processes in an SMB.

After a few weeks, measure again with the same questions. Hours saved count, but not only: a quote that goes out the same day, fewer errors to fix and a team that can breathe count too. Keep those numbers: they'll also help you choose the second project.

Hypothesis: you write down the starting point before connecting AI, then measure again with the same questions.

What if the first project happened in a single day?

That's the idea behind the AI integration day. We come to your offices for a day, we train your team on its real files and we don't leave without 2 automations working in your tools.

We choose the 2 tasks with you, using the same criteria as in this article: what comes back, what takes time, what's easy to check. They're tested on your real data before we leave. At the end of the day, we take stock: what works, what's left to do and the possible next steps.

We don't promise to transform your business in a day. We promise a concrete first step that runs and a clear idea of what comes next.

It's not a conference. It's a working day, and your first AI project runs the very next day. All the details are on the AI integration day page.

What's left at the end of the day: a trained team and 2 automations running in your tools.

Frequently asked questions.

Where do you start with AI when you know nothing about it?

With a specific task in your business, not with the technology. Look for what comes back every week, takes time and is easy to check. You don't need to understand how AI works to know where it would save you time.

How much does a first AI project cost?

It depends on what it has to do and which tools need to be connected. With AI, a simple app costs from $5,000, and a prototype takes a few days instead of a few months. The best approach is to start with a small project whose effect you can measure.

Do you need perfect data to start with AI?

No, but you need to know where it is and that it's up to date. AI that reads incomplete or contradictory files makes mistakes, and faster than a person. Often, the first project starts by putting a bit of order in the information.

Do you need to hire an AI specialist to get started?

Not to get started. A first project can be done with an outside team, while yours learns to work with AI on its real files. What matters most is someone on your team who knows the task well and owns the project.

What is the AI integration day?

It's a day when our team comes to your offices. We train your team on its real files and we don't leave without 2 automations working in your tools, tested on your real data. At the end, we take stock of what works and of the next steps.

The right first AI project isn't the most impressive one. It's the one that removes a real chore in your real tools and whose effect you can measure.

Start small, measure, then move on to the next one. That's how you move forward without ending up with a pilot sleeping in a drawer.

Read next

An idea in mind?

Let’s chat!

By submitting this form, I agree to the privacy policy.