The world of data science continues to attract professionals from diverse backgrounds, bringing fresh perspectives to an increasingly complex field. In a recent conversation, Yoann Veny, Data Scientist at Agilytic, shared his unique journey from sociology to artificial intelligence, offering valuable insights into the current challenges and opportunities facing organizations as they navigate the generative AI revolution.

An Unconventional Path to Data Science

Yoann's story begins with what he calls an "atypical" background for a data scientist. Originally pursuing sociology at university specifically to avoid mathematics and sciences, he discovered an unexpected passion for statistics during his social sciences studies. This revelation led him to complete a master's degree in statistics for social sciences at KU Leuven, setting the foundation for his future career.

When I had statistics courses, I don't know why, but there was a click that said 'Ok, this can be used to do this... this kind of thing.' It was really a revelation.

This interdisciplinary foundation has proven invaluable throughout his career. After transitioning from academia to industry, Yoann spent a decade at a telecommunications operator, where he experienced firsthand the power of data-driven insights. A particularly memorable project involved combining marketing and technical network data to predict customer churn – a perfect example of how different data silos within organizations can be united to solve complex business problems.

The Consulting Advantage: Diversity and Methodological Rigor

After ten years in a large corporation, Yoann joined Agilytic, embracing the dynamic world of data science consulting. The transition highlighted key differences between in-house and consulting roles, particularly around diversity of experience.

"The main change that interests me a lot, because I have a very curious mind and am always in continuous learning mode, is diversity," Yoann explains. "We can work in retail, pharmaceuticals, financial institutions, the public sector. I find this really interesting because it's one of the things that made me think working in data and statistics was interesting – these are problems that repeat from one sector to another, with skills that can be transposed to problems sometimes of very different natures."

This cross-sector experience has reinforced a fundamental truth about data science: while business contexts may vary dramatically, the underlying methodological approaches remain remarkably consistent. Whether predicting churn in telecommunications, healthcare, or retail, the statistical foundations and analytical frameworks transcend industry boundaries.

Navigating the Generative AI Hype: Beyond the Chatbot Request

Perhaps nowhere is methodological rigor more crucial than in the current generative AI landscape. Organizations across sectors are approaching consultants with requests that often begin with "I want a chatbot," reflecting both the excitement and confusion surrounding these new technologies.

"The real challenge today is guiding businesses as they define what they want and could do with generative AI," Yoann observes. "Companies often tell us 'I want a chatbot.' Technically, we can build one—no problem. But the bigger challenge is sitting down with these stakeholders and taking the time to define the concrete problem they're trying to solve, and whether a chatbot is actually the best solution."

This challenge has fundamentally changed Agilytic's approach to client engagement. While traditional machine learning projects could often be scoped in one or two workshop sessions, generative AI initiatives require deeper diagnostic phases lasting five to ten days. This extended discovery process addresses several critical questions:

• What specific business problem needs solving?

• Is generative AI actually the right solution?

• What technical infrastructure and data requirements exist?

• How will the solution scale from proof-of-concept to enterprise deployment?

The Data Quality Imperative: Why GenAI Projects Remain Data Projects

Despite the apparent simplicity of deploying modern AI tools, Yoann emphasizes that generative AI projects remain fundamentally data projects. This perspective, rooted in a decade of experience at Agilytic, challenges the common misconception that these are primarily IT initiatives.

"We have the extremely strong conviction that, despite all the facilitations we have today to deploy GenAI solutions, and it's becoming easier and easier, these projects are not IT projects as many believe," he states. "We have the conviction at Agilytic that GenAI models remain above all data projects."

The implications are significant. Organizations rushing to implement RAG (Retrieval-Augmented Generation) systems with poorly maintained document bases will inevitably fail. The same data quality principles that govern traditional machine learning – clean, well-organized, regularly updated information – apply with equal force to generative AI systems.

This creates new challenges as organizations must now consider unstructured data sources that may have never been subject to rigorous data governance. Legal departments maintaining policy documents, for instance, must now understand that their work feeds into automated processes requiring consistent quality and maintenance standards.

Democratizing AI Innovation: Public Support for Enterprise Transformation

One encouraging development is the growing public sector support for AI adoption among businesses. Both the Walloon and Brussels regions offer substantial financial assistance to help organizations overcome the initial barriers to AI implementation.

Programs like "Start IA" support diagnostic phases and roadmap development, while initiatives such as the Walloon region's "Tremplin IA" can finance up to €40,000 in proof-of-concept development, with 70% public funding. For Brussels-based companies, Innoviris offers specific support for generative AI projects.

"Often companies aren't even aware that this exists," Yoann notes. "It's often us who spread the good word to certain prospects or clients. But I think this really helps get over that hurdle of de-risking the POC, because sometimes certain companies can be reluctant about this kind of inherently risky POC."

These programs recognize a fundamental truth: the biggest barrier to AI adoption isn't technical capability but rather the financial and organizational risk of initial experimentation.

The Future of Enterprise AI: Beyond Technical Implementation

Looking ahead, the success of generative AI initiatives will depend less on technical sophistication and more on organizational readiness. This encompasses several dimensions:

• Cultural Adaptation: Teams must understand how their work contributes to automated systems and adjust their processes accordingly.

• Scalability Planning: Solutions that work for five test users must be architected to handle hundreds of concurrent users in production environments.

• Framework Selection: Rather than building everything from scratch, organizations benefit from methodical evaluation of open-source frameworks across multiple dimensions – from OCR capabilities to maintenance requirements.

Yoann's team has developed a comprehensive evaluation matrix for RAG frameworks, assessing approximately ten different open-source options across various criteria. This approach ensures that solution recommendations align with both technical requirements and organizational capabilities.

Personal Perspectives: AI in Daily Life

Beyond professional applications, Yoann's personal use of AI technologies offers interesting insights into broader adoption patterns. Like many professionals, he uses ChatGPT for everyday queries, noting how it has become particularly engaging for children due to its conversational interface.

More uniquely, his musical hobby has benefited from AI-powered audio separation tools that can isolate individual instruments from complete recordings. "Starting from a single track, separating it between different tracks – drums, bass, vocals – is quite incredible," he explains. "It's nice to be able to take a song, remove the guitars from the song to leave the drums, bass and possibly the vocals, and play the guitar yourself over it to join Metallica or any other music group."

This personal application illustrates AI's potential to enhance creative pursuits, suggesting broader implications for how these technologies might integrate into various aspects of daily life.

Conclusion: Methodology Over Hype

As organizations continue to navigate the generative AI landscape, Yoann's experience offers several key lessons:

1. Take time for proper diagnosis: The complexity of GenAI implementations requires more thorough upfront planning than traditional ML projects.

2. Remember data fundamentals: Despite technological advances, data quality remains the foundation of successful AI initiatives.

3. Leverage available support: Public funding programs can significantly reduce the financial risk of initial AI experimentation.

4. Focus on business problems: Technology should serve clearly defined business objectives, not drive them.

5. Plan for adoption: Technical success means nothing without organizational adoption and cultural change management.

The generative AI revolution is real, but its success will be measured not by the sophistication of the underlying models, but by how effectively organizations can integrate these tools into their operations while maintaining the methodological rigor that has always defined successful data science initiatives.

As Yoann's journey from sociology to AI demonstrates, the most valuable practitioners often bring diverse perspectives and systematic approaches to complex challenges. In an era of rapid technological change, this combination of methodological discipline and cross-functional thinking may prove to be the most important competitive advantage of all.