AI Transformation Will Be Driven by the Business Units (or It Won’t Happen)

TECH & AI

AI will serve the customer and the business functions that serve them, or it will not happen at all.

Too often, the IT department deploys AI without sufficiently involving the business units.

The AI execution gap: a misalignment between profitability expectations and actual deployment.

Nearly two-thirds of CEOs expect strong profitability from AI. According to BCG, 62% of them are confident about AI’s return. The real question is one of time horizon and value created. Since AI saves time, productivity should follow, and automation should bring costs down. Yet in many companies, while the time savings on automatable tasks are real, they have not yet translated into significant, large-scale cost reductions.

Still, there is a clear pressure for AI to deliver profitability across the executive committee, and it mostly takes the form of cost reduction.

In many companies, the IT department not only deploys AI operationally but is also, more often than not, in charge of the strategic vision. According to Forrester, more than 50% of AI deployments are led by the Chief Information Officer.

Business functions (marketing, CX, insights, customer service, sales leadership, and so on) have a voice but are not sufficiently involved in the process. Work does not start from their needs, let alone from those of the customer, nor from the way they work or their processes. Budgets and roadmaps too often sit with IT. The result: according to Gartner, only 30% of Chief Marketing Officers say they are ready or mature enough to deploy AI at scale.

This is clearly a mistake.

Of course, AI technologies are complex, evolve very quickly, and carry significant legal and security risks. No company wants to be caught in a scandal where its customer data or source code is exposed.

Of course, the goal is to avoid so-called “shadow AI” and the use of unsecured tools. Of course, AI tools must also be integrated into the company’s information system and be scalable and industrialized.

But AI changes the way we work. Rather than industrializing an existing process, you first need to simplify and rethink it for the AI era, before trying to automate it. This requires a detailed understanding of how work actually gets done and of operational processes, and it demands strong involvement from users and business leaders.

The most advanced players are, in fact, those whose AI budget is managed by business units against clearly defined KPIs from the outset. Otherwise, the risk is discovering that the “token” budget has been massively consumed without IT knowing exactly who used it. This is what we have seen since this summer: a sharp tightening of AI spending.

Worse still, I hear more and more business leaders forcing the hand of their IT departments. A head of e-commerce recently told me: “IT wanted us to use the hundreds of Copilot licenses they can’t make pay off. In less than a month, I worked under the radar with a few dozen Claude licenses and we built real-time inventory tracking dashboards and a CRM tool for our marketplace, which would have taken IT months to implement...”

From individual use of generative AI to collective use of agentic AI.

While AI adoption is massive, driven by shadow AI, the challenge is to move from individual use centered on office productivity tools to collective use centered on business processes and the company’s core business.

Too often, IT departments have secured an environment built on one or several LLMs so that each employee can use AI tools and boost their individual productivity. But the real need is to address the specific requirements of a function (marketing, customer service, logistics, HR, etc.), and that calls for purpose-built tools.

This is the key difference between generative AI, which more than 40% of French people say they use at least weekly (according to IFOP/Converteo), and agentic AI, which only a minority of companies have truly deployed.

“Just because we’ve deployed agents here and there doesn’t mean we have a real agentic process,” the Chief Data Officer of a large French group recently told me.

Real value often lies at the intersection of business units, roles, and functions.

To get the most out of AI, companies need to rethink how internal teams and external partners, such as advertising agencies, work together. This is especially true for content personalization: AI requires rethinking the entire content value chain, from creation to production to distribution. This involves many functions, internal and external, from legal to creative teams to IT.

When such projects are deployed as part of a multi-year strategy, the value created is significant, sometimes exceeding €150 million for large-scale projects whose business plans I have been able to review. This is the case for major consumer goods companies such as Mondelez, which launched a far-reaching overhaul of its content strategy across a portfolio of brands. Other consumer brands, such as Nestlé and L’Oréal, are further ahead.

Ultimately, it comes down to rethinking business processes and workflows. AI transformation is therefore not just a technology matter; it is first and foremost a matter of transforming how we work. What can be simplified, automated, and delegated to an agent? What control and validation steps must be put in place?

This kind of transformation requires strong governance and spans several years, much like the digital transformation did. Moving from a Chief Digital Officer to a Chief AI Officer will not be enough. AI is invisible and seeps into everything, so it must be embedded across the entire organization.

We must not repeat the mistakes of the digital transformation: the business model cannot rest solely on cost reduction; it must first contribute to improving the customer experience. To succeed, the business units, which serve the customer, must be far more involved in the AI strategy and its deployment.

By Thomas Husson, independent analyst.

I believe in quality over quantity so don’t expect daily or even weekly updates, but rather bi-monthly or monthly news – assuming I’ll only write when I have something meaningful to share.