Smile news

The next challenge for Gen AI projects? Moving from the pilot phase to industrialisation!

  • Date de l’événement May. 29 2024
  • Temps de lecture min.

Guillaume Lanthier, Business Unit Manager, AI Factory at Smile, answers 5 questions about the life cycle of GenAI projects

1/ Let's start from the beginning: what is the main value of Gen AI? 

Gen AI is a transformative technology. I'm a big believer in this notion of transformation: Gen AI is profoundly changing the trajectory of business transformation. By its very nature, AI gen manifests itself almost everywhere: in everyday tools and applications (chatbots, editorial assistants, translation, messaging, graphic design, etc.), but also in processes and in the way businesses are approached. 

Gen AI is of great interest to all types of companies and organizations. If you don't, your competitors will be well ahead… Most of them still need time and support, and launching a pilot project is not always easy (at Smile, we have prepared a document to help them define their needs, which you can download here!) Productivity is obviously a key issue, and there is much promise and excitement in what can be achieved by using these technologies. The expectations will vary from one sector to another! 


2/ What challenges will they face?


There are plenty of challenges. When I say that AI is and will be essential for companies to remain competitive, I'm only stating the obvious! I think the breaking point is industrialisation. Let me explain: AI will be a performance driver for companies that are able to move from experimentation to industrialisation. This may seem like a big step for directors and managers, but by accurately identifying the obstacles, they will be best placed to remove any barriers to AI adoption and scale. 


3/ What about users? How can they be encouraged to adopt Gen AI tools?

I mentioned above that AI promises to improve productivity and performance. AI technology has the potential to increase user interest and trust, especially when it specifically addresses their business needs. The adaptation of the famous LLM models is a driving force for adoption by employees, who will recognise their own value as specialists. In particular, RAG (Retrieval Augmented Generation) adaptation techniques are used to train the models on specific data, giving them a specific context. Of course, as in any technology or tool adoption, the proposed user experience is critical


4/ What can companies do to scale up? 

Adopting a data-driven approach, defining the technical and technological path, sharing the product vision, involving all business stakeholders (CTO, IT, Marketing, Purchasing, Sales, etc.), considering processes, promoting change management, deciding on governance, working in sprints, closely monitoring impact and controlling costs are all necessary steps for successful & industrialized deployment of Generative AI. 

Here's what I would recommend if I had to choose 3 priorities: 

  • Encourage change management: Setting up a core team will lay the foundations for the project. This team is in a position to define the complete framework (legal, methodology, technology, operations...).
  • Identifying and implementing the right tools and methodologies for these specificities is perhaps the most complex, but also the most important step in managing scalability and performance, monitoring (and therefore ensuring longevity), and managing costs. 
  • Finally, the adoption of a 'product' vision for the evolution of applications over time. By providing a clear, understandable direction that is adapted to the needs of users, the generative AI project will be able to endure over time and meet the business objectives in its particular environment.


5/ What Gen AI technologies should they adopt? 


I say open source because most open source generative AI solutions are more adaptable, and I'm more confident that these technologies will be able to adapt to the specific needs of each company, regardless of their size, industry, strategy, maturity or market positioning. Open source models also have the advantage of being available in the cloud. This speeds up their time-to-market! It's also a matter of philosophy: By relying on open models, fed by people from all walks of life, we limit bias and can look critically at the answers we come up with. 


Would you like to find out more and/or contact us?