AI is now accessible to almost everyone. Opening a ChatGPT or Claude account for an entire team takes a few minutes. And when people talk about transitioning to AI, this is almost always what SMEs do. Yet giving this access is not enough to transform a team’s productivity. Three weeks later, two people are genuinely using it and the others have gone back to their old reflexes.
The gap remains wide, even at the scale of an entire continent. According to Eurostat, in 2024 only 13.5% of European companies with at least ten employees were using AI technology. The tool is spreading fast, but a figure like that recalls one simple thing: having access does not mean someone knows how to work, or think, with it.
At ScaleMyCrew, we train our contributors in Madagascar in the practical use of AI from the moment they arrive. What we drew from it holds in one idea that changed our way of training. Training a contributor in AI does not consist of teaching them to use a tool. You have to teach them to rethink their way of working. The real goal is for them to think with the tool.
Using AI is not yet knowing how to work with AI
A contributor who occasionally asks ChatGPT a question uses AI. They are not yet working with it. The difference seems thin, but it decides almost everything. Occasional use boils down to opening a window when stuck, reading the response, then closing it without changing any habits. AI remains an accessory that is displayed, never an additional position doing its share of the work.
Integrating AI into your work is something else. The contributor identifies the moments in their week where the tool genuinely changes the result and calls on it without being reminded. An assistant told “you can use AI for your emails” stays at the first level. The same assistant shown how to prepare in one minute the response to a recurring complaint, then anticipate the next ones instead of being caught off guard, moves to the second. They have seen the gain on a recurring task in their week. They have changed their way of handling this task, not just gained one more tool. This observation holds everywhere, whether the contributor is in your office or in a remote offshore team.
Why some AI training produces few results
Many training programs start from the wrong point. They start with the tool: the tour of menus, then a long list of features and prompts to copy. The contributor comes out impressed and applies two or three tips that change nothing in their real tasks. They were shown software, nobody connected it to their real work.
The perfect prompt is the best example of this false start. You see entire training programs built on phrasing recipes, as if the result lay in the magic of a sentence. A contributor who recites a learned prompt gets stuck the moment their task moves one centimeter from the template. The one who understood what the tool does well and where it goes wrong adapts to any situation.
What is missing from these programs is processes. The real tasks the contributor executes every week, on their real files, rather than tool features taken out of context. Training that starts from there installs a skill that is reusable, because it attaches to a known action.
Learning to “think with” AI
Thinking with the tool is something that is learned. It is even the heart of training. A contributor who thinks with AI starts by breaking down their task. Faced with a complex request, they do not launch a vague query hoping for a miracle, they separate what can be entrusted to the tool from what requires their own thinking. They then give context, because the tool knows neither the client nor the house’s habits. The less context given, the further the response strays from what is expected.
Then comes interaction. A first response is rarely the right one. The contributor who thinks with the tool relaunches and refines it until they get what they are looking for, instead of settling for the first draft. Then they verify, always. An AI tool makes mistakes, like anyone: it produces a fluent and well-turned response, sometimes wrong, with nothing in the tone signaling the error. Never letting a figure go out without cross-referencing it with its source becomes a reflex installed before everything else.
Judgment remains, which belongs only to the human. Deciding whether the response holds and whether it can go to the client, no tool does this in place of the contributor. This is what is meant by thinking with the tool: the person keeps control, AI accelerates.
Training a contributor to identify what can be augmented or automated
A contributor who thinks with the tool ends up asking themselves a question on their own: in everything I do, what could AI take on? Training in AI also means training this way of seeing. The rule transmitted holds in one sentence: you automate what is repetitive and verifiable, you keep in hand what requires understanding a context or engaging the relationship.
From there, the use changes in nature. The contributor stops treating AI as a one-off fix and starts building genuine sequences, what are called workflows. Taking invoice information to fill a table, then preparing a first report from these figures: the sequence triggers almost on its own and the contributor checks the result at the end. Recent tools go further with agents, capable of holding a chain of tasks without someone stepping back in at each stage. Teaching a contributor to spot these opportunities is often worth more than teaching them ten features.
From AI as assistant to AI as process component
This is the real shift: AI becomes a link in the process instead of an assistant occasionally called upon. A few examples show this better than a definition. In email processing, it sorts messages and summarizes long threads, while the contributor adjusts responses before sending. For prospect qualification, a sales assistant entrusts it with list sorting and public information search, before taking back control to write the message that counts. On the reporting side, already-entered figures are transformed into a readable dashboard without re-entry. In documentary research, it finds in a few seconds what used to take half a day of reading.
One very telling case. An e-commerce contributor managing a site’s product sheets lets AI rename images and write a first draft of descriptions, while she reviews and corrects what touches on price or the commercial promise. The tool handles the volume, she keeps the sensitive points. At this stage, AI is no longer placed alongside the work, it has entered it.
The challenge for SMEs: creating teams capable of evolving with AI
For a European SME, all of this comes down to one simple question: can my team still progress on its own? A solid team must not only execute tasks. It must be able to evolve and integrate AI into its functioning. A team that executes fixed tasks ages quickly. One that has learned to think with AI absorbs a new tool without redoing all the training and identifies on its own the tasks to optimize the following year.
A team does not become this by chance. It rests on four supports that training must establish together: skill development of contributors, documentation of ways of doing so that a use found by one benefits the others, clear processes on which to connect the tool, without forgetting agents when they genuinely save time. An SME that installs these four supports gets a team that grows with AI rather than being subject to it. The first two play out mainly internally. On the last two, this is precisely where ScaleMyCrew supports SMEs.
How ScaleMyCrew integrates this logic into its dedicated teams
This is exactly the logic we apply in our dedicated offshore teams. Our contributors in Antananarivo do not all arrive mastering AI. We train them from their arrival, starting from their real tasks. We take the files they actually handle and show them where the tool saves time and where they need to keep control. They learn to verify what it produces before trusting it, instead of copying a response because it looks right. This is the way of working we pass on to them: thinking with the tool, not just using it. A European account manager keeps the thread over time: he monitors each person’s progress and documents with the client the validation rules as well as what works. The one to two hour time difference with Europe allows direct exchanges on Slack, like with a colleague down the corridor.
The result is a dedicated offshore team that does not merely execute, it evolves. When a contributor finds a use that saves time, they document it. The rest of the team picks it up immediately. This is what transforms a dedicated offshore team in Madagascar into a device that progresses instead of stagnating.
FAQ: what executives ask us about AI training
Thinking with the tool, the real skill to pass on
Training a contributor in AI requires more than giving them access and crossing your fingers. It requires starting from their real tasks and teaching them to verify what the tool produces, before leading them to identify what can be augmented or automated in their day. At the end, they think with the tool. Their team progresses with them.
This is what we build every day with our contributors in Madagascar, then with our clients’ teams. Are you looking to develop a dedicated team in Madagascar while integrating more AI into your processes? Let’s discuss the tasks that could be optimized or automated in your organization.
Publié le 14/08/2026