Predictability Pulse: Supply Chain’s Data Revolution Is a Training and Strategy Problem

A conversation with Rémi Alexandre — Independent Supply Chain & Forecasting Consultant and CentraleSupélec alumnus — on bridging the gap between machine learning models, corporate operations, and the AI-assisted upskilling of non-technical teams.

By Louise Servoin · 2026-08-04 · 12 min read

## Rémi Alexandre – Supply Chain & Forecasts Consultant

Expert in S&OP, data architectures, and automated predictive workflows. Alumnus of CentraleSupélec and Lycée Descartes. Former Demand Planner at Guerlain (LVMH), Product Owner Assistant at L'Oréal, and BI Analyst at Amazon. Based in Luís Eduardo Magalhães, Bahia, Brazil.

Some professionals understand database design. Others understand the physical realities of global inventory. Rarely do you find someone who can navigate both while operating across high-stakes international luxury networks and agile mid-sized enterprises.

Rémi Alexandre’s trajectory — from the rigorous mathematics and physics prep school of Lycée Descartes to CentraleSupélec’s premier engineering program to managing S&OP for a €600M cosmetics portfolio at Guerlain — places him at a unique intersection of operations research and digital transformation. Now, as an independent consultant, he is solving a critical paradox: how to take the advanced forecasting models of multinational giants and make them accessible, actionable, and educational for smaller, non-technical teams.

We sat down with Rémi to discuss why predictive tools fail without proper human integration, how low-code automation and AI can act as training co-pilots for operations teams, and why the ultimate metric of any supply chain is not algorithm complexity, but team autonomy.

## The Pragmatic Architect

Rémi, you spent years inside corporate giants like Amazon, L'Oréal, and LVMH's Guerlain before launching your independent consulting practice. What made you decide to transition to helping mid-sized enterprises (SMEs)?

What I enjoyed most in the very mature companies I worked in wasn't running an already-optimized process, it was setting up something new: a practice, a team, a way of working that didn't exist yet, inside business intelligence or supply chain. That's where I learned the most, and where I felt the most useful.

At some point I realized I wanted more of that specifically, more contact with the moment where a company is still deciding how it's going to structure its data, its forecasting, its planning, rather than just running what's already in place. Consulting, with the experience and training I'd built at Guerlain, L'Oréal and Amazon, was the natural way to stay close to that kind of work and to do it for more than one company at a time.

What surprised me is that the friction points turn out to be nearly identical whether it's a €600M portfolio or a €40M company: forecasts built on gut feeling rather than history, no shared visibility on stock targets, S&OP meetings that end without a real decision. What's different is resources : a multinational has a team to build that structure. An SME usually has one person doing it on the side of another job. That gap is exactly where I try to be useful.

## The Foundations of Analytical Rigor

Your educational foundation began with theoretical physics and mathematics at Lycée Descartes before you specialized in Supply Chain engineering at CentraleSupélec. In an era where AI is often treated as a "black box" that solves everything, how important is that deep mathematical background?

It matters less for building the model than for explaining it. Every forecasting tool produces outliers and outputs nobody expected, if you don't understand why, you can't get a business stakeholder to trust it. At Guerlain, some orders were distorting the forecast because of specific rules in how they were entered. Understanding the mechanics is what let me isolate and fix that, rather than just flag "the model looks wrong."

At Amazon, that same instinct to look under the hood : mapping inventory flows directly in SQL rather than trusting a report, and it led to recommendations valued at close to €2 million in annual savings.

The prep school and engineering years gave me the habit of always looking under the hood rather than trusting a tool at face value.

## The Implementation and Adoption Gap

You have contributed to deploy software tools globally—including L'Oréal's project of stock visibility platform used by 160 key users across 120 countries. Why does the "bridge" between great data and day-to-day operational execution break down so frequently?

It almost always breaks at adoption, not at the model. A clean model that a planner doesn't trust gets ignored and they go back to their own Excel file.

At L'Oréal, I helped deploy a stock visibility platform that reached 160 key users across 120 countries. What made the difference wasn't the dashboard itself, it was building it around how the team actually worked day to day, one reliable dataset instead of five conflicting ones. When the tool matches the real workflow, adoption isn't something you have to push.

## AI-Assisted Enablement & Modern Training

How can we leverage AI or AI-assisted training to ensure that non-technical teams do not become passive, but instead become highly autonomous managers of these supply chains?

This is the interesting shift. Managing a forecasting system used to require knowing databases and statistics. Low-code tools and AI change what "training" means. Instead of teaching someone to extract and clean data, you can have the system surface the anomaly directly and explain it.

Using tools like Power Automate to eliminate manual compilation work, and giving teams a clear read on WAPE reports, shifts the job from processing data to interpreting it.

The goal isn't a team that watches a dashboard. It's a team that understands what a forecast error means for their stock and their cash, and can act on it without waiting for someone else to interpret the number for them.

## The Irreplaceable Human Element

As AI continues to commoditize data processing and routine statistical forecasting, what is the one human capability that an algorithm will never replicate in S&OP?

Getting people to agree. A model predicts baseline demand well. It can't sit sales, finance, and operations in the same room and get them to align on a plan they'll all actually commit to.

Sales wants to over-forecast to avoid stockouts. Finance wants to under-forecast to protect working capital. Operations wants stable, predictable runs. None of them are wrong, they're optimizing for different things. Reconciling that into one number everyone will act on is the real work of S&OP, and no algorithm does that for you.

## The Golden Thread

We often see a gap between what technology can do and what professionals actually adopt. How do we ensure AI becomes a 'co-pilot' that elevates human judgment rather than a tool that encourages intellectual passivity?

By keeping the human decision visible. The risk isn't that AI replaces judgment, it's that a dashboard becomes so smooth people stop asking why. My rule is simple: if a tool can't explain its recommendation in terms a non-technical stakeholder understands, it's not ready to be trusted, no matter how accurate it is.

Tags: Supply Chain, S&OP, AI in Logistics, Power BI, CentraleSupélec, LVMH, Guerlain, L'Oréal, Amazon, Dr. Beckmann