Two out of three professionals report unregulated use of artificial intelligence in their company, according to the Forvis Mazars–ESSCA barometer published on October 1, 2026. This figure raises a question that goes beyond mere risks: why do so many employees prefer to seek their own solutions rather than use those provided by their employer?
- Two out of three professionals use AI tools without company approval, a practice called Shadow AI.
- Employees turn to these tools because they save time on fragmented, inefficient tasks.
- Faster work means little if the output is not useful, so results must be measured after AI use.
- Companies should rethink their methods and eliminate useless tasks, not just automate them.
This phenomenon is called “Shadow AI”. It refers to a fairly common practice: using ChatGPT or another AI tool without going through the company’s approval channels. IT departments rightly see this as a risk. Confidential data can be exposed, inaccurate results can circulate unchecked, and sensitive information can be entrusted to external providers.
The article, however, invites us to look at the other side of the phenomenon. In a customer service department, an advisor must retrieve the history of a request in one software, consult a procedure in a knowledge base, check information in the ERP (the company’s integrated management software), and then draft their response. This sequence can take several minutes, without adding any value for the customer. If this advisor discovers that an AI tool can save them time, they will naturally be tempted to use it.
Time Wasters That Have Become Normal
The phenomenon is not limited to customer relationship centers. Excel spreadsheets circulating between departments, reports rebuilt manually, approvals whose usefulness no one can really explain: work ends up being organized around system constraints rather than around what needs to be produced. Information searched for multiple times, a file sent to the wrong department, or a request processed twice all consume resources without improving the outcome.
These inefficiencies end up seeming normal. Employees get used to them, managers work around them, and, because they don’t always appear on dashboards, they often escape investment decisions. To gauge their scale, the article takes the example of a 500-person company: ten minutes lost each day per employee represents, over 220 working days, more than 18,000 hours per year. Not all these hours could be recovered or directly converted into savings, but the order of magnitude gives food for thought. According to the same barometer, 58% of respondents indicate that their company plans to increase its investments in AI.
Saving Time Does Not Mean Improving Performance
Speed of execution is often confused with improved performance. A department that halves the time spent on its monthly reports achieves a result that seems excellent. If these reports are rarely used, the real progress, however, lies in questioning their usefulness, not in producing them faster.
In customer relations, a company can reduce its average handling time and observe, a few weeks later, an increase in complaints, as customers have to call back to get a solution. To evaluate the benefits of AI, the article suggests looking at what happens after its use: was the request resolved, did the file require further intervention, did the customer have to make contact again, and was the time saved at one stage simply lost elsewhere?
Reviewing Methods, Not Just Tools
Letting every employee freely choose their tools is not an answer, as risks to data, regulatory obligations, and the need for reliable information require a framework. Between a blanket ban and a laissez-faire approach, however, there is room for a more nuanced policy: asking teams which tasks they tried to simplify, which tools they lack, and which procedures slow them down.
The issue also concerns senior management. For years, companies have invested in systems designed to standardize operations and facilitate management. These investments have brought considerable progress, but they have sometimes multiplied steps, interfaces, and administrative constraints. According to the article, it will also be necessary to agree to eliminate certain tasks, not just automate them. The author poses a final question: “What’s the point of having the most advanced technologies if we continue to make our teams work according to methods that no longer make sense?”
