Companies have never invested so much in artificial intelligence, but the economic results remain difficult to demonstrate. Employees save time, tools are proliferating, and experiments are accelerating. The question remains: who must transform these gains into value for the company.
According to the McKinsey study The State of AI in 2026, 80% of respondents report an improvement in their individual productivity thanks to AI. However, only 37% report an impact, even partial, on their company’s operational results.
Spending, meanwhile, has increased sharply. According to BCG, in September 2026, it doubled in one year to reach 3.3% of the revenue of the surveyed companies. More than 80% of this spending is now outside IT budgets, meaning the AI transformation concerns the entire company and no longer just the IT department (DSI).
Return on investment is not measured by time saved
Take the example of a company that equips its salespeople with an AI assistant capable of preparing their appointments, writing their reports, and automatically updating the CRM, i.e., the customer relationship management software. Each salesperson saves thirty minutes a day, which seems like a success.
What happens to these thirty minutes? If they allow for meeting more prospects, improving the conversion rate, or reducing the sales cycle, the company can begin to measure value creation. If they are absorbed by other administrative tasks, the individual gain does not necessarily translate into economic performance.
The same reasoning applies to software development. AI accelerates code production, but this acceleration can shift bottlenecks to reviews, tests, or pre-production validations. “Individual productivity is a promise. End-to-end performance is the result to measure.” Therefore, before each project, a baseline metric, a measurable objective, and a business manager responsible for verifying that the gains materialize must be defined.
Business units must be accountable for the value produced
Major technological transformations have often been driven by IT departments. With AI, this model is reaching its limits, because an IT department can assess technical feasibility and guarantee the security of an architecture without being best placed to identify the pain points of a sales department or the repetitive tasks of a finance team. The author indicates having structured, within their organization, a working group bringing together representatives from different business departments.
Each department must be able to answer three questions: what problem is it trying to solve, what measurable benefit does it expect, and who will be responsible for the result? The IT department remains responsible for technical coherence, while the business units become accountable for the value produced. The finance department, for its part, must help distinguish theoretical savings from actual, observed benefits.
Once priorities are set, the question of implementation remains. The AI Factory model, a small team of engineers serving the business units, can design solutions quickly, provided it does not limit itself to demonstrations. An application still needs to be secured, connected to existing systems, monitored, and maintained. The prototype proves a solution is possible; production must prove it is useful and profitable.
