95% of AI projects never make it.Here is why — and what we do differently.

of generative AI projects never reach production

It is not the technology that fails: projects remain pilots, with no integration into business processes and no governance.

of agentic AI projects will be canceled by the end of 2027

Three causes cited by Gartner: escalating costs, unclear business value, and insufficient risk controls.

agentic AI vendors deemed real, out of thousands

Gartner names the practice: "agent washing" — rebranding existing chatbots or RPA without real agentic capability.

of respondents see individual productivity improved by AI

But only 6 % of organizations draw a significant impact on their bottom line (EBIT). AI benefits individuals before it benefits the company — that is exactly the gap to close.

of organizations are held back by the cost of AI

Operating costs, including the cost of tokens.

of executives anticipate AI-related headcount reductions

Versus 32 % a year earlier. But half of the cuts announced last year never materialized.

of the economic value of AI captured by 20 % of companies

Across 1,217 executives surveyed in 25 industries: value is not spread out, it concentrates.

more financial performance for the companies most mature in AI

Performance measured as the revenue and efficiency gain attributed to AI, relative to the industry median.

more performance when usage increases WITH the foundations (data, processes, governance)

It is not the model that delivers ROI: it is what you put around it. Increasing usage without the foundations halves the return.

more automated decisions without human intervention — and more governance

The best automate more AND control more. These are not two opposing goals.

more likely to rethink processes, rather than stack up tools

Adding a tool without rethinking the work it replaces produces a lower return.

of companies are genuine leaders in AI adoption

An "outside-in" measure applied to more than 600 listed U.S. companies, to avoid self-assessed statements.

of shareholder return above the median, over three years

This outperformance comes entirely from fundamentals (revenue, margins), not from a stock-market fad.

of headcount growth — they hire faster, they do not lay off

The dominant narrative presents AI as a tool for cutting headcount. Among the leaders, it is the opposite.

of executives say AI underperforms because of failing controls

"Don't blame the technology": what holds AI back is not the tech, it is governance.

are not certain they would pass an AI governance audit within 90 days

In other words: they cannot demonstrate what their AI does.

only have a truly developed and deployed AI strategy

51 % of executives cite strategy as the first factor for return — and 22 % have one.

of executives believe their teams are truly ready to adopt AI

Technical deployment has outpaced the teams' ability to use it.

trust and governance stand out as the real competitive edge

Despite massive adoption, very few organizations convert their AI activity into lasting results. The gap comes down to governance, not tooling.

of employees use AI tools at work (Salesforce, 2026)

The usage is already here, massively — whether the company decided it or not.

only of organizations can name ALL the AI tools in use (Gartner, 2026)

The gap between what is used and what is known is the very definition of unmanaged risk.

use tools their employer did not provide (Microsoft)

More than one employee in two goes around the official tooling.

of those who use unauthorized AI shared sensitive data

This is not a hypothetical risk: it is a finding.

of generative AI users do so without their employer knowing (Deloitte UK, 25,000 employees)

Deloitte even measured that British employees spend close to a billion pounds out of pocket on these tools.

of annual gap between two companies with an identical procurement budget

On a base of $2B in procurement and $10M in identical budget: one generates $96.8M in gains, the other $50.6M. The gap does not come from the budget, but from the ability to steer.

return for companies that deploy AI in depth

Compared with those stuck in pilots. Only 17 % of the organizations surveyed moved past the testing stage.

median ratio of annual gain to investment

A €10,000 investment generates about €16,000 in measurable gain in the first year.

of deployments reach production

of go-live when human supervision is planned (versus 73.9 % without)

177 of the 200 deployments included a documented human supervision mechanism.

median time before a positive return

Median investment observed: €61,874.

of French companies use at least one AI technology

8 points more than in 2024; the rate has tripled since 2023.

adoption in companies with 250 or more employees

The gap with the smallest organizations remains very pronounced.

cite a lack of expertise as a brake

More than half of the companies that use AI report being held back by a lack of expertise — even though they have already started.

incidents handled by ANSSI in 2025

data exfiltrations compared with 2024

Attackers now aim at data theft rather than simple encryption.

ransomware compromises reported

of French companies hit by ransomware in 2024

The highest rate in Europe.

of victims take more than a month to recover

Versus 24 % in 2023: recovery takes longer, not less.

of victims recover thanks to their backups

Backup is the first line of defense — which also makes it the attackers' first target.

general application of the regulation — enforcement has begun

machine-readable marking requirement for generated content

"high-risk" regime for autonomous systems (postponed by 16 months)

of French small and mid-sized businesses use at least one AI solution

This figure doubled in one year.

average return per euro invested in AI

Up to 10.3x for the most advanced organizations.