European companies are set to invest €150 billion in artificial intelligence over the next two years, yet most projects still fail to deliver value, Mistral AI told asset managers at the ALFI Global Asset Management Conference in Luxembourg on 24–25 March 2026.

Guillaume Bour, head of enterprise Europe at Mistral AI, said the gap reflects a structural problem rather than a lack of investment, arguing that many institutions are deploying AI without sufficient control over how systems operate inside core processes.

He said the problem becomes acute as AI systems move into core business processes, where firms must be able to explain and justify how decisions are produced.

“AI is going to be at the very core of all the internal and business processes of your organisation,” Bour said. “Most models are black boxes today.”

“You know what you give them, you know what you get in return, but you don’t know what happens behind.”

That lack of transparency raises compliance and governance concerns for financial institutions, where decision-making must be explainable and auditable. Firms need to understand not only outputs, he said, but why those results are produced.

Domain limits

Bour said the limitations of general-purpose models are becoming clearer in finance, where performance depends on access to proprietary data.

“General AI do not integrate domain-specific intelligence,” he said.

Models trained on broadly similar public data tend to behave in similar ways, he added, limiting their ability to generate differentiated outcomes. The only way to improve performance is to incorporate institution-specific data that reflects the realities of financial markets and operations.

Industrial shift

The industry is now moving from experimentation to execution, he said, with the challenge shifting from testing use cases to scaling them.

“The real question is not, can we build a use case,” Bour said. “The real question is, how do we bring those dozens or hundreds of tests… to production at scale?”

Mistral AI has structured its financial-services offer around three elements: sector-specific models trained with proprietary data, agentic systems capable of operating within defined processes, and infrastructure that allows firms to retain control over deployment.

“We partner with financial services organisations to train some models with that specific data,” he said, “to bring domain-specific knowledge so the models are way smarter when it comes to talking about finance.”

Operational use

The approach is being applied across banking, insurance and asset management, including in areas such as know-your-customer checks, anti-money laundering and reporting processes.

“An agent is basically a model that is able to take actions,” Bour said, describing systems that can operate within workflows rather than simply respond to prompts.

He also pointed to time-series modelling for market forecasting, where even marginal improvements in accuracy could generate significant value.

Customer-facing systems are also evolving, he said, with models expected to move beyond answering queries to executing transactions on behalf of users.

Sovereign deployment

At the core of the approach is what Bour described as “sovereign” AI: systems deployed in a way that preserves control over data and architecture.

“The only way to have that is to deploy where the data is,” he said.

That requires institutions to build internal capability to operate AI systems. “You have to build your platform,” Bour said. “You have to hire the people who actually can help you build and operate your AI workflows at scale.”

Targeted rollout

Rather than attempting full-scale transformation, Bour said firms should focus on a limited number of high-impact use cases.

“An iconic use case is a use case that is strategically valuable, that is urgent,” he said.

“Identify the top 10 iconic use cases and bring them at scale.”

Operational implications

The ALFI Global Asset Management Conference has this year placed emphasis on the operational implications of artificial intelligence alongside regulation and market structure.

The Mistral session reflected a broader shift in the industry, as financial institutions move from pilot projects towards scaled deployment while seeking to retain control over data, models and decision-making processes.