Artificial intelligence in business: feedback

The dominant discourse presents artificial intelligence as an almost miraculous solution, capable of instantly transforming businesses. However, between marketing promises and operational reality, the gap often remains significant. To measure it, nothing is better than concrete experiments, anchored in the field.

This is precisely the objective of a pilot project carried out by a supermarket located in a coastal area, grappling with a serious problem: management of the strong seasonality of activities. The challenge was not to innovate for its own sake, but to respond to a very real economic constraint.

Beyond the story, this experiment highlights four strategic lessonssometimes counterintuitive, about the real and relevant use of AI in businesses.

Lesson no. 1: The real challenge of artificial intelligence is not futuristic, it is operational

Contrary to popular belief, the most critical application of AI in this project was neither customer experience nor predictive marketing, but a much more concrete topic: optimization of team planning.

The goal was clear: improve the accuracy of business forecasts by consolidating several years of sales data with complex contextual variables:

• weather conditions,
• school holidays by area,
• public holidays,
• promotional operations.

This approach illustrates an often overlooked reality: the biggest performance improvements rarely come from the most visible usesbut by the rigorous optimization of fundamental business processes.

A case in point: an improvement in solo 0.10% of the paycheckachieved through better management of the volume of hours worked for seasonal reinforcements, generates a significant impact on overall profitability.

In this context, the value of artificial intelligence is not science fiction, but its ability to solve essential management problems.

Lesson #2: When a tech giant fails to meet expectations

The heart of the experiment consisted of comparing different artificial intelligences on a specific analytical task: the exploitation of Open Data for over 20 years (weather, school calendar, holidays) in order to identify the critical periods, in particular the famous «bridges» outside the school holidays.

The result was unexpected.

While some competing AIs have successfully identified these complex patterns correctly, repeatedly, regardless of the data format (spreadsheet, spreadsheet, or CSV).

The observation is clear: where several solutions provided exact and actionable answers, this tool was unable to produce a reliable analysis.

This situation is all the more striking given that the solution in question was already integrated into the internal ecosystem and was the subject of dedicated training.

Lesson no. 3: Powerful AI isn’t magic, it knows how to produce reliable code

This failure is not anecdotal. It reveals an essential truth for leaders and decision makers: the performance of an analytical AI is not measured by its conversational fluencybut to its ability to translate a business problem into a robust computer program.

In fact, language models do not “compute” directly. They generate code, most often in Python or JavaScript, responsible for performing the processing.

In this project, the faulty tool:

• did not systematically generate a program,
• or produced code that introduces conversion errors,
• leading to inconsistencies and confusion in data interpretation.

The lesson is strategic: Choosing enterprise AI doesn’t mean choosing abstract intelligencebut an engine capable of producing code that is reliable, understandable and suited to the specific challenges of the organization.

An invisible criterion, rarely proposed, but absolutely decisive.

Lesson n.4: Maturity also means knowing how to say no to AI

This experiment not only allowed us to identify relevant use cases. He also highlighted the importance of voluntarily give up certain usesfor human, ethical or simply pragmatic reasons.

Two paths were therefore excluded:

• Seasonal recruitment : the automated analysis of student CVs was judged to be of little relevance (scarce data) and too sensitive from an ethical point of view.
• Theft prevention through video analysis : although technically possible, this option was abandoned, because it was perceived as intrusive and potentially counterproductive in a collaborative transformation process.

This ability to say no was also expressed in the definition of project priorities. Other uses, such as the generation of marketing content or post-campaign analysis, have been deliberately relegated to the background, because they are considered less critical in terms of return on investment.

Conclusion: true intelligence is strategic clarity

This experiment demonstrates one essential thing: true intelligence in digital transformation is not artificial, it is strategic.

Success does not depend on having the most hyped AI, but on a clear approach:

1. identify the business problem with the greatest impact,
2. choose the truly suitable tool (AI or not),
3. accept technical, human and organizational limits.

This clarity even led teams to recognize that some challenges, such as managing complex promotions, were more a matter of NoCode tools than AI, confirming that a mature innovation strategy remains technology agnostic.