According to a TCS survey 86% of company managers had already deployed artificial intelligence (AI) to improve their existing sources of revenue or create new ones in 2024, demonstrating that the major trend represented by artificial intelligence.
While demonstrations and pilot projects are multiplying, the real challenge lies in companies' ability to move from experimentation to large-scale use. This is what emerged from a discussion between Béatrice Kosowski, President of IBM France, and Christophe Lesur, Managing Director of Cloud Temple.
To watch the whole exchange follow this link to the video of the show
Beyond the GPU: rethinking AI infrastructures
"The GPU issue is often overestimated", says Christophe Lesur. "Some high-performance models can run on modest infrastructures. This pragmatic approach calls into question massive and costly infrastructures. CPUs may suffice if processing time is not critical.
Contrary to popular belief, the race for raw GPU power is not the only way forward. IBM now favours more targeted foundation models, such as Granite 3.0, comprising between 2 and 8 billion parameters, a far cry from the 500 billion-parameter giants. These lighter models reduce resource consumption by 3 to 23 times while maintaining high performance.
The challenge of scaling up
The figures are stark: according to Béatrice Kosowski, 54% of POCs (Proof of Concept) never reach the production phase. This figure can even reach 70% according to certain analyses.
To ensure successful industrialisation, a three-stage methodology is proposed:
- Start with business needs
It is essential to identify the relevant use cases precisely and assess their potential ROI. This first step will enable you to eliminate superfluous processes and concentrate on creating real value. As Béatrice Kosowski points out, "We need to set automation and AI in motion to serve a process that is really close to the company's business."
- Controlling data
Data quality appears to be the main obstacle to AI projects in organisations. Data is often dispersed, compartmentalised and in heterogeneous states. The implementation of effective governance and the use of appropriate tools such as watsonx.data are becoming essential for organising and adding value to data.
- Managing risks
Risk management is essential: risks of drift, hallucinations, bias and regulatory non-compliance. The implementation of strict governance and management tools enables companies to control these risks while ensuring compliance with current regulations.
IBM and Cloud Temple have announced a strategic partnership to deliver IBM's watsonx.ai artificial intelligence solution on Cloud Temple's secure and sovereign cloud infrastructure. This collaboration will enable large-scale AI applications to be deployed while guaranteeing data security and sovereignty, in line with SecNumCloud standards.
watsonx.ai offers a wide choice of AI models, multi-cloud deployment capabilities, and advanced tools for developing and managing AI models.
This combined solution meets the needs of customers wishing to exploit IBM's AI technologies in a trusted cloud environment, qualified as SecNumCloud.
If you would like our team to help you with an artificial intelligence project, please contact our teams.
Key success factors
The involvement of senior management is an essential prerequisite. "The first people who need to be trained are the general managers", insists Christophe Lesur. This training must be accompanied by the creation of dedicated teams capable of understanding the technical and business challenges.
IBM reports concrete results: 3 billion in productivity gains since 2023, with a simultaneous improvement in the customer and employee experience. For example, automating 70% of customer support reduced the time taken to resolve questions by 26%, while increasing customer satisfaction by 25 points.
The recommended approach is to first test use cases with the best performing models on anonymised data, before considering more specialised or lighter models. This gradual approach allows you to quickly validate the relevance of your projects while controlling the risks.
Deploying generative AI on a large scale requires a clear, structured approach that combines strategic vision, solid technical skills and pragmatic implementation. Success depends on the ability to match technology to real business needs, establish effective governance and innovate while managing the associated risks.