AI in Business: The Real Challenge Is No Longer Time Saved

Companies are investing heavily in AI, but productivity gains do not always translate into financial results. The challenge is shifting towards governance, ROI measurement, and business accountability.

Companies are accelerating their adoption of artificial intelligence, but the gap remains clear between time saved and value created. According to McKinsey, 80% of respondents to The State of AI in 2026 report an improvement in their individual productivity thanks to AI, while only 37% observe an impact, even partial, on operating profit. At the same time, BCG indicates that by September 2026, AI-related spending had doubled in one year to reach 3.3% of the revenue of surveyed companies. The issue is therefore no longer limited to tool adoption: it becomes an economic management problem.

Le temps gagné ne paie pas les factures tout seul

An AI assistant can prepare meetings, write reports, or update a CRM in seconds. Very good. The question remains what the company does with this freed-up time. If it is used to handle more prospects, shorten the sales cycle, or improve conversion, the value becomes measurable. If it disappears into other administrative tasks, the gain remains confined to the individual level.

The reasoning also applies to software development. AI speeds up code production, but bottlenecks may simply shift to reviews, testing, or validations before deployment. Producing more does not mean delivering more. Individual productivity is a promise; end-to-end performance is what counts.

Qui porte la valeur : la DSI ou les métiers ?

The question becomes central as uses extend beyond the IT perimeter. An IT director can frame security, architecture, or interoperability. However, they are not necessarily best placed to identify the pain points of a sales team or the repetitive tasks of a finance department.

It is in this context that a cross-functional approach gains weight. The right question is not only "what can AI do?" but "what problem do we want to solve, what benefit do we expect, and who is responsible?" According to the logic described in the source, the IT department remains responsible for technical coherence, the business units become accountable for the value produced, and the finance department must distinguish theoretical savings from actual gains.

One detail matters here: AI investments now largely exceed the IT budget. BCG notes that over 80% of spending is now outside the IT department. The center of gravity shifts without warning the organizational charts. This rarely pleases dashboards, but dashboards were not consulted for their opinion.

De l’usine à prototypes à la chaîne de valeur

Recent advances in models allow small engineering teams to quickly create internal applications, agents, and automations. Hence the idea of an AI Factory, a tight-knit team serving the business units, capable of designing quickly and then industrializing. The concept makes sense, provided demonstration is not confused with production.

A solution developed in a few days still needs to be secured, connected to existing systems, supervised, and maintained. It also needs a business owner to ensure adoption. Otherwise, the prototype remains a prototype, with a shinier veneer than its real impact.

The source also highlights the limits of two extreme models: an AI Factory that is too centralized can become a bottleneck; engineers scattered across each business unit can multiply isolated developments, costs, and security risks. The hybrid model therefore appears more robust: proximity to users, common technical standards, and pooling of reusable components.

Mesurer l’IA comme un portefeuille, pas comme une collection d’idées

The ROI of an AI program does not depend on the number of projects launched, but on the ability to select the right ones, track results, and stop those that do not deliver on their promises. Before launch, it is necessary to document the cost of the current process, expected gains, necessary investments, and risks. During experimentation, it is better to measure on a limited scope, comparing it to the initial situation. After deployment, all costs must be included in the equation: development, licenses, model consumption, supervision, maintenance, and change management.

At this stage, a company must also accept a simple rule: AI can save time while increasing verification or maintenance costs. The calculation is therefore not limited to hours saved. It must include cost reductions, revenue increases, service quality, or the capacity to absorb more activity.

A clear AI policy completes the whole: authorized tools, data protection rules, responsibilities, and levels of human validation. Governance is not meant to slow down teams. It is meant to accelerate good projects and limit bad ones.

Investir sans ROI immédiat, mais sans brouillard

Demanding immediate profitability on every initiative would be a mistake. In 2026, models are advancing rapidly, and a company that stopped experimenting would take a strategic risk. Some approaches must remain exploratory, with a defined budget, duration, and learning criteria. But they must not drift in the vague.

The distinction between operational projects and preparatory experiments has become decisive. The former must create value today. The latter serve to test new possibilities or prepare tomorrow’s skills. Maturity is therefore not measured by the volume of AI deployed, but by the ability to transform technical possibilities into economic results, then to reallocate investments accordingly.

To read