Making data useful: a decade of lessons with Sophie Buresi

10 years in data, from hype to impact. Sophie has led data transformations at Artefact, LVMH, Etam, and most recently as a contractor at Pernod Ricard.

Consulting, end client, freelance: she's tried all three angles, with one consistent thread — creating business value through data. In this interview, we talked architecture, AI, hiring, internal politics, stack, cost control, and strategic vision.

Consulting, client, freelance: three angles, three realities

What Sophie takes away from her time in consulting is the ability to switch between strategic and operational subjects with no transition. "You're trained to do the splits: talking data roadmaps with a CDO in the morning, then unblocking a SQL script in the afternoon."

In-house, the reality changes: you manage a P&L, you arbitrate budgets, you measure your vision against political cycles. "You start to understand why things don't move as fast as you'd want from the consulting side: you have to convince, align, decide."

Today as a freelancer, she navigates between those two worlds: speed of execution on one side, an understanding of internal mechanics on the other. A rich balance, but a demanding one: "You no longer hold the reins — you coordinate, you influence, but you don't get to decide."

Artefact: learning to stand on your own at 25

"I was a director at 25. I was managing accounts worth several hundred thousand euros for CAC 40 companies. Honestly, I didn't have the professional maturity to handle everything perfectly. But I learned. And what held it all together was this team of warriors that had been hired."

At Artefact, she learned to structure, to ship fast, to manage young. The processes were solid, the expectations high, the overall level impressive. "The hiring process was demanding, involving the teams. It produced a real Artefact mold — a guarantee of quality."

But that mold has its limits. "When you hire people who all look alike — with strong targeting on top business or engineering schools — you gain speed, but you risk creating a culture that's too homogeneous. It works at the start. But to scale, you have to open up. Your customers don't necessarily look like you. And even less so internationally."

Data & AI at ETAM: the field, not the theory

When Sophie joined ETAM, the brief was clear: build a data function from scratch. Across three streams:

1. Move from reporting to steering: "Every company surfaces numbers. Very few actually make decisions with them."
2. Optimize the business with targeted AI use cases
3. Explore the potential of GenAI: marketing asset generation, AI editorial photos, optimizing e-commerce shoots

The use cases are concrete, ROI-driven:

• 18-month sales forecasting
• Smart inventory allocation across stores (factoring in real foot traffic, nearby construction, etc.)
• Promo campaign optimization (less linear, more personalized)
• Product recommendations on the e-commerce site
• Marketing mix modeling (media budget allocation)
• AI-generated assets for product visuals

The strategy? Don't do everything — focus on what's most core to the business and most specific. "We focused our internal effort on core business topics, like supply. The rest, we outsourced."

Structuring a data team in a retail company: concrete challenges

At ETAM, the Data Factory team grew to 20 people, split across 4 pillars:

• Data Tech (platform, ETL, APIs)
• BI (the legacy team)
• Data Value (data scientists, PMs, analytics engineers)
• Data Governance (an embryonic team, but essential)

A hybrid model: in-house plus freelancers. "But careful: if your run depends on freelancers and your capex budget gets cut, the whole thing collapses. You always have to think about insourcing."

What AI really changes

For Sophie, GenAI has a double effect:

1. A new playing field for generating value: automation, creativity, speed.
2. A new risk if used badly: "Anyone can prompt an LLM and pull out 'insights.' But without data culture, you create analytical errors."

The key? Don't slow down adoption — frame it. "You have to support, train, put in place light but real governance. And you need people capable of challenging the prompts, the results, the interpretations."

She also flags a point that's often underestimated: the organizational cost of scaling. "POCs are fun, but to scale you need infrastructure, solid profiles, run, real governance. Otherwise, it stays a fad."

Hiring in data: the real challenge is finding senior, curious, autonomous profiles

Sophie says it plainly: "It's not the role that's hard to hire — it's the level. The good profiles are expensive, already taken, or very demanding."

What she needs? Hybrid profiles, senior, but who keep their hands dirty. "You can't afford to hire one profile per technology. You need people who can cover broader ground, without losing the bar on quality."

And to keep them? "Give them meaning, feed them intellectually, give them autonomy, don't manage top-down. And be a decent person — that simple."

Her view on management: clear the path, don't control everything

Her approach: "curling-style" management. "My job is to sweep ahead so the team can move in the right direction."

She describes herself as demanding but caring. No constant control, but a clear expectation of quality, commitment, and transparency. "I love when people make decisions on their own. But I'm there when needed."

And now? Toward startups… or somewhere else

Sophie doesn't just want to "go back to a permanent role" — at least not in the same scope. She's exploring. Startup studio? VC fund? Shadow CDO? Data-first COO? "I want to try something different. Try a different way into the data and AI market. And there are a thousand ways to enter the startup ecosystem without necessarily being a founder."

Her advice to companies launching into AI:

• Build data governance from day one: "Catching up is always more expensive than starting on the right foundation."
• Make AI a product brick, not a styling effect
• Use consultants… in moderation: "Things move too fast to do everything in-house, but you have to keep the skills in-house for the long term."

Tools, stacks, costs: frugal rigor

Sophie recommends classic cloud stacks (the modern data stack), but warns about cost. "More and more startups are coming back to self-hosted infrastructure once they've proven the value. To control costs, one option is to take back ownership of your infra."

What she's reading and listening to

• 📚 Medium / LinkedIn
  • Chad Sanderson
  • Bar Mooses
  • Nicolas Dessaigne
  • Jean de la Rochebrochard
  • …
• 🎙️ Podcasts:
  • Another Podcast by Benedict Evans
  • Data de Robin (professional, precise, accessible)
  • GDIY (less data, but always inspiring)
  • Le Podcast by Pauline Laigneau (less data, but inspiring profiles, personal development)
• 👥 Regular peer exchanges to benchmark tools, orgs, stacks

Working with Sophie?

Sophie is freelance for now. She takes on engagements around structuring, organizing data/AI functions, or leading projects.

Her day rate ranges from €1,000 to €2,000, depending on the scope, duration, and mutual interest.

Conclusion

This kind of conversation reminds me why I love this job: discovering paths you couldn't make up, visions built on experience rather than buzzwords. Sophie checks every box you'd want in a head of data: strategic vision, hands-on rigor, a sense of the collective. And she does it without ever putting on airs. Just by doing the work well.

Thanks Sophie!