Artificial intelligence is being deployed across all departments of companies, with significant productivity gains. But its adoption raises very concrete choices: which data should be entrusted to which actors? What technological dependency should be accepted? What environmental impact and, above all, for what profitability?
“Tomorrow, competitive advantage will no longer come from adopting AI, but from the way companies choose to implement it.”
The choices made at the start of an AI project become all the more important when it scales up: a poorly sized model or infrastructure can increase costs and resource consumption, while a strong dependence on a supplier can limit the possibilities of changing solutions.
Controlling your data and your footprint
The choice of an AI does not depend solely on its performance. When a company uses an external solution, it must assess the data entrusted to the provider and the applicable legal framework. An actor subject to US law may, under certain circumstances, have to transfer data hosted in Europe under the CLOUD Act. For strategic information, local or European hosting can therefore offer greater control.
These choices also have an environmental dimension. A difference in consumption that may seem limited for occasional use becomes significant when usage multiplies. Choosing a model suited to the actual need and an efficient infrastructure makes it possible to limit the consumption of electricity, water and material resources.
Finding the right economic balance
The general-purpose solutions of major providers make it possible to respond quickly to many use cases. But when volumes increase or when AI involves internal data and strategic processes, other criteria become more important: cost per use, control over data and dependence on the provider.
An open-source model hosted on dedicated infrastructure requires more expertise initially, but can offer greater control and more predictable costs. There is therefore no ideal solution: the choice depends on the use case, volume, desired performance and desired degree of control.
How should an AI project be conducted?
Before investing, an assessment of the uses, available data and the company’s needs makes it possible to identify the projects that truly create value. General-purpose solutions such as ChatGPT or Copilot may be sufficient for certain needs; others require a more specific approach. Three steps can structure this process.
- Identify the highest-value use cases. Identify the processes for which AI can provide a measurable benefit: time saved, improved quality, reduced errors or the creation of new services.
- Assess data sensitivity. Distinguish between ordinary data and personal, confidential or strategic information. The more sensitive the data, the more their hosting and processing require safeguards.
- Choose the solution suited to each use case. Compare the options according to four criteria: performance, cost, sovereignty and environmental impact. The most powerful model is not necessarily the most relevant.
Brussels has a card to play
Balancing performance, cost, control over data and environmental impact is becoming a strategic issue for companies. Having local expertise to support them in these choices therefore represents a real asset.
With its network of SMEs, universities, research centres and proximity to the European institutions, Brussels has strong assets for developing this expertise locally. Seeing Brussels-based players specialising in more sovereign and sustainable AI emerge is therefore excellent news for the regional ecosystem.
BRAFT is part of this dynamic, working with companies to determine which uses, which data and which solutions genuinely deserve to be deployed.
By Bruno de Formanoir and Brieuc Thibaut, co-founders of BRAFT, and Sébastien Yasse, director of ALTEO.
