Context

From January 5th to January 9th, 2026, I completed a short-term contract at Salomon, where I previously did my second gap-year internship of 6 months.

In mid-December, I was contacted by Salomon’s Footwear R&D Director (Thomas GAUTHEY), who explained that there was an urgent need related to data management. I had been recommended by a Salomon employee with whom I had made a positive impression earlier in the year.

Accepting this mission meant missing a full week of classes. While it was a relatively light week due to the return from the Christmas holidays, I still missed the start of some courses, notably Reinforcement Learning and Cloud Computing. After discussing the situation with my professors, I decided to take on the challenge. I like challenges and this one really excited me.

Problem Definition

On Monday morning, I met my colleagues, all experts in product development. Together, we analyzed the need and clarified the scope of the mission.

The customs authority of a specific country had recently increased the strictness of its requirements regarding the disclosure of material compositions for all products exported to that country. The main objective was therefore to complete a detailed file to be shared with customs.

Beyond this immediate goal, a secondary (but strategically important) objective was defined: start building a data pipeline that could automate this process in the future. This long-term perspective is why my manager deliberately chose an engineering profile rather than assigning the task to a purely manual role.

Technical and Organizational Constraints

Once the scope was defined and the introductory meetings completed, I was introduced to Salomon’s current Product Lifecycle Management (PLM) software.

Very quickly, several constraints became clear:

  • No available API
  • Outdated user interfaces
  • Significant latency when loading pages
  • Data exports limited to Excel or XML

Later that day, I learned that a file containing most of the required data already existed (50% of required data actually). While this initially raised concerns about redundancy, it quickly became clear that this file needed to be linked with other datasets containing material-level information and detailed material compositions.

Results

I skip a detailed day-by-day account of the five working days and focus instead on the outcomes.

  • The primary objective (completing the customs file) was fully achieved.
  • The secondary objective (building a reusable data pipeline) was significantly more complex than anticipated.

The datasets were highly heterogeneous and originated from multiple sources. Some data was missing and required direct follow-ups with product managers, who had the best knowledge of the products.

Units were manually entered and did not follow any standardized format or naming convention:

  • Examples: CM*M, cm * M, ft2, inch*yard, etc.

While some inconsistencies could be handled using regular expressions, others went far beyond simple unit conversion. I intentionally simplify the reality: the challenges were deeper and more structural.

Key Learnings

This short mission turned into an impactful lesson in real-world data engineering:

  1. Data quality is the foundation of automation
    End-to-end automation is impossible without consistent, well-structured, and reliable data.

  2. Heterogeneous data sources dramatically increase complexity
    Even simple business requirements become challenging when data originates from multiple tools and teams.

  3. Manual data entry is a major bottleneck
    A significant portion of the work still required manual input, validation, and correction.

  4. Traceability is critical
    Being able to follow transformations and manual edits is essential for debugging pipelines and understanding how final outputs are produced.

Conclusion

Although this mission lasted only five days, it provided valuable exposure to the realities of industrial data workflows. It strongly contrasted with the clean, well-defined datasets often used in academic settings and highlighted the importance of robust data foundations before attempting automation (large-scale or not).

This experience reinforced my interest in data engineering and system design in real-world, constrained environments where technical solutions must adapt to imperfect tools, processes, and data.