BrazilMetyis partner Louis de Cointet on bringing Generative AI to life
Louis de Cointet is a partner with Metyis, specialized in IT strategy and data transformations. As a key member of the Generative AI practice at Metyis, we asked Louis five questions about how businesses should go about adopting this rapidly growing technology.
What are the biggest benefits that you have seen GenAI drive for your clients?
When Generative AI is tailored to the specific needs of a customer, it can be deployed far more rapidly than traditional AI approaches, especially in areas like complex text analysis, message classification, and, of course, content generation. This enables the creation and implementation of conversational solutions, such as chatbots, with remarkable speed, delivering a level of response quality that far surpasses what was possible with earlier tools.
Generative AI can also power all kinds of improvements to processes, in areas such as marketing, sales, customer services, engineering, software development, and much more.
In projects that we conducted, we have seen productivity gains ranging from 30% to 40% and significant error reduction, with up to a 40% decrease in entry errors.
Kickstarting GenAI starts with selecting the right Large Language Model (LLM). What are some of the key things organisations should consider?
The starting point must always be the functional use case, the business objectives (including ROI), and the client’s context, such as data and technical environment. Large Language Models (LLMs) can be an ideal fit for certain needs, but they may not always be the best solution.
It is crucial to maintain a critical perspective, especially as it’s becoming increasingly easy to believe that a solution can be delivered ‘turnkey’ through an API, which isn’t always the case.
This is especially true with LLMs, where the rapid evolution of models on the market means that cost and performance can vary significantly depending on the specific use case. Selecting the right model is essential to achieving optimal results. I have in mind a project where we observed a 1-to-10 cost ratio between two LLM models designed to address the same need.
When it comes to developing a GenAI solution, what is the process that should be followed?
Always begin with the use case – whether it’s content creation assistance, customer chatbots, or text corpus analysis, to name a few.
Once the functional requirements are clearly defined and validated, identify the key constraints, such as data, architecture, security, intellectual property, legal/compliance, user interface, and API integration. This approach allows for the creation of a detailed specification that outlines both the technical architecture and the AI pipeline, paving the way for effective development.
And, of course, iteration is required to ensure solutions are tested and the right one is found, with the help of continued testing and feedback from users. Keep in mind that 80% of an AI project's success depends on the human factor.
A key part of using AI is being responsible. How does Metyis balance the need for innovation with the need for safety?
First and foremost, it is important to remember that while LLMs have revolutionary aspects, they come with significant limitations. Notably, they lack true understanding and reasoning capabilities, which can result in what is known as hallucination. Moreover, the costs to run these models can quickly become difficult to bear if businesses are processing or generating large amounts of complex data, not to mention image and video.
To address this, it’s essential to design systems where performance measurement – whether through statistical tools or human oversight – is central to the approach.
It's also crucial to remember that the EU AI Act, which came into effect last August, introduces a host of new constraints that will significantly impact all AI ecosystem and the companies that use it.
Finally, what advice would you give to organisations looking to adopt GenAI solutions, particularly those that are new to AI?
AI, especially Generative AI, is not magic, but it can deliver substantial benefits in many domains, and it can do so rapidly compared to traditional approaches. However, to fully harness its potential, it’s crucial to have the support of expert “doers” who have successfully implemented these solutions before – something we, Metyis, bring to the table.
