IDC: EMEA CIOs need to rebuild their data and ROI models to get AI into production

IDC believes enterprise AI deployments in EMEA have stalled not because enthusiasm for the technology has cooled, but because companies are poorly prepared to scale. To move projects out of the pilot stage, CIOs need to rework the economics, the data architecture and the way new tools fit into teams' daily work, all at once.

Why pilots don't grow

Over the past year and a half, companies in Europe, the Middle East and Africa have invested heavily in large language models and classic machine learning, expecting a noticeable operational payoff. Now boards have begun to put the brakes on some programs, scale them back or shift the focus of investment. According to IDC, this is not a loss of interest in AI: the business is demanding tougher financial justification amid competing IT priorities and macroeconomic pressure.

The numbers are harsh: only 9% of organizations in the region have managed to get a measurable business result from most of their AI projects over the past two years. The other 91% have not necessarily run into technical failure, but they are stuck in a gray zone between the demo and real production. Projects don't die loudly. They lose momentum, budget and internal support, and never turn into a lasting corporate function.

Where ROI breaks down

One of the main mistakes is evaluating AI with the old procurement logic, where the value of new software is measured by headcount reduction or direct license savings. For generative models this often doesn't work. Their effect shows up indirectly: through lower risk, faster work by specialists, less downtime and new revenue sources. If a company looks only at obvious costs, a promising pilot almost inevitably loses the fight for budget.

CIOs essentially need to redescribe the economics of deployment and audit not just the model but the whole chain it runs through:

• losses avoided, for example from downtime or errors
• higher team productivity and shorter cycle times
• new revenue scenarios and digital services
• the full cost of inference, storage, integrations and support
• spending on security, compliance and data control

The next barrier appears in the move from a cloud sandbox to the corporate environment. A pilot can be stood up quickly on an API and test data, but production requires constant compute, live pipelines, monitoring and integration with legacy systems. When modern vector databases have to be connected to old Oracle or SAP environments, architectural gaps surface immediately. RAG scenarios need clean, labeled and properly classified data; otherwise answer quality drops and hallucinations increase. At the same time, bills for inference and model fine-tuning are growing, and finance teams now expect them to be explained line by line.

Architecture and people

IDC separately stresses that data protection, cybersecurity and model explainability requirements in Europe do not necessarily hinder scaling. On the contrary, companies that build governance and control rules in from day one move faster. Protection against prompt injection, clear access boundaries, documentation of model decisions and control over data raise a project's baseline cost, but they also make the system fit for real corporate use and strengthen customer trust.

Equally strong resistance arises among employees. An AI solution can work technically and still fail to take hold if it breaks an established process or demands too abrupt a restructuring from the team. So CIOs need to invest not only in models but also in retraining, change management and a rollout designed around how people actually work. Automated contract review makes sense when lawyers spend less time on routine and more on negotiations and complex risks. Against this backdrop the CIO's role is changing: according to IDC, 42% of top executives in EMEA expect the CIO to lead digital and AI transformation, with an emphasis on creating new revenue sources.

What this means

For EMEA, the "let's just try AI" phase is ending. The next stage will be won not by companies with more pilots, but by those that can prove their value, prepare the data and build models into real business processes without a gap between technology, finance and people.