RAISE Summit Paris: 3 lessons on AI in the enterprise, from data to execution

On July 9, the DATASOLUTION teams attended the RAISE Summit in Paris, one of Europe’s leading events dedicated to artificial intelligence for business.

On stage: an executive from one of the most heavily regulated banks in the world and the CEO of a young European company specializing in AI agents. On paper, two very different worlds. In reality, the same conclusion: AI creates value only when it is connected to well-managed data, real-world processes, and a clear measure of its impact.

Without that, even the greatest potential remains just a promise.

At Goldman Sachs, AI is already impacting many aspects of the organization: software development, internal processes, business teams, partners, and clients. The company has observed a net productivity gain of about 20% among developers, but Marco Argenti emphasizes a key point: this figure is meaningful only because AI is built on a solid data foundation.

AI is therefore not simply a layer added on top of existing tools. It requires a reliable, connected, and well-governed data infrastructure. In a regulated company, this requirement is not a hindrance—it becomes an advantage.

And that is exactly the message we convey at our Data Stories events: the AI Act is moving in the same direction. Long viewed as a constraint—or even a barrier to innovation—it can, on the contrary, become a structuring framework. By requiring that uses be tracked, data be documented, access be controlled, and humans remain in the loop, it compels us to build the very foundations that make AI reliable and deployable at scale. In other words, regulation is not at odds with productivity: when properly understood, it reduces risks and creates the conditions for industrializing AI with confidence. What applies to an American bank like Goldman Sachs also applies, on its own scale, to any European organization that will have to comply with the AI Act.

Data is not a technical prerequisite. It is what determines whether AI actually delivers value.

This prototype is not put directly into production. It serves to make the concept much clearer. Engineers can then improve its reliability, enhance its security, and make it suitable for mass production.

At H Company, Gautier Cloix takes this line of thinking a step further. In his view, people are still too often tied up with routine tasks: clicking, searching, copying and pasting, re-entering information, switching between multiple tools, or coordinating software programs that don’t communicate well with one another.

However, their true value lies elsewhere: setting a goal, understanding a client, devising a service, weighing options, making decisions, and guiding a strategy. The challenge, therefore, is not merely to complete the same tasks more quickly, but to free teams from the operational level so they can refocus on the bigger picture.

This is precisely where agents and computer use become interesting: not because they put on a nice demonstration, but because they can operate within existing software and handle concrete operational tasks.

The real breakthrough isn’t just automation. It’s the ability to turn a business objective into actual execution.

From Design to Production: The Cycle, Security, and ROI Make All the Difference

Another key takeaway from the RAISE Summit: You can’t create sustainable value by moving directly from an AI model to production.

Marco Argenti explains that a model’s outputs should be treated like the work of a junior employee: useful, fast, and sometimes highly relevant, but they need to be reviewed, tested, validated, and integrated into a secure workflow. In a corporate environment—and even more so in a regulated sector—AI must be governed by safeguards: testing, human validation, data control, development pipelines, security, and oversight.

Gautier Cloix adds another criterion: results. In his view, an AI project should not be evaluated based on the number of prompts, tokens used, or tools deployed. These metrics may measure activity, but not value.

Good indicators are more concrete:

• savings achieved;
• revenue generated;
• time saved on critical processes;
• reduction in operating costs;
• improvements in quality or speed of execution;
• creation of a strategic asset for the company.

In particular, he gives the example of optimizing long-tail purchases within a large corporation. Major purchases, such as steel, concrete, or heavy equipment, are often well negotiated. In contrast, secondary purchases—cables, light bulbs, fire extinguishers, and small equipment—are rarely optimized with the same rigor, as there are too many of them and they are too scattered.

Agents can then review contracts, invoices, internal systems, and the web to identify opportunities: unused discount thresholds, cheaper suppliers, purchasing consortia, potential renegotiations, or more effective local practices. For certain categories of purchases, this type of approach can yield significant savings, with a measurable ROI within a few weeks.

Behind every AI system that truly delivers, there is the same process: act, measure, correct, and iterate.

In other words, the data must not only be available at the outset. It must feed back into reality, drive continuous improvement, and enable the system to evolve over time.

Our Book Club at DATASOLUTION

Whether AI is regulated or agent-based, whether it’s adopted by a bank with 45,000 employees or by a startup deploying agents, the bottleneck remains the same: data.

Large organizations highlight the challenge through their scale, constraints, and security requirements. Everyday data and AI projects make it crucial: without a clean, structured pipeline that is grounded in reality, even the most impressive demonstrations will remain just that—demonstrations.

AI in business isn’t just about choosing a model or deploying a tool. It involves connecting the right systems, ensuring data reliability, securing access, measuring impact, monitoring results, and integrating models into actual business processes.

It is precisely at this intersection that our expertise lies: structuring, ensuring the reliability of, and leveraging data so that models deliver on their promises, from prototype to deployment.

And to wrap up with a quote from Gautier Cloix: Next year, we may talk less about AI itself and more about what it has made possible.

To put it simply, the real work isn’t in the hype surrounding the models. It’s in the data, reliability, integration with processes, and deployment.

Thank you to the RAISE Summit for the fruitful discussions.

Do you work on data and AI projects, from prototyping to deployment?

Let’s talk about it.