# Carrefour: B2 Paris Students Optimize the Pet Food Assortment

**Authors:** Albert's Deep Dive
**Categories:** Business Deep Dives
**Tags:** Carrefour, assortiment, Retail, LightGBM, B2, Data science
**Last Updated:** 2025-11-05T15:57:34.967Z
**Reading Time:** 1 min read

---

## Summary

In front of Carrefour teams, B2 students presented a data-driven (LightGBM) Pet Food assortment approach, tailored by store format, to maximize sales, satisfaction, and space.

---

Last Friday, the B2 students presented their analyses to Carrefour, aiming to optimize the on-shelf product assortment for customer satisfaction, sales, and space efficiency. After a 100% advanced-math final, Enzo Natali, Nathan Sarfaty, and Sacha Nardoux were the jury’s and audience’s favorites.

“I am very happy to have won the Carrefour Business Deep Dive, an exciting challenge that allowed us to put our data analysis and commercial strategy skills to the test.” The winning team worked on a solution to optimize the Pet Food assortment, leveraging LightGBMRanker and LightGBMRegressor, with recommendations tailored by store type (Hyper, Market) and management of cannibalization.

The jury praised a realistic approach aligned with business objectives. The model forecasts an estimated 0.4% increase in revenue per store, with an interactive dashboard to visualize the impact of the recommendations.

## Key Takeaways

1. B2 Paris team used **LightGBMRanker** + **LightGBMRegressor** to optimize Carrefour’s **Pet Food** assortment with **store-format** (Hyper/Market) tailoring and **cannibalization** control.
2. Projected **0.4% revenue uplift per store**, designed for business realism and ready for validation via **A/B pilots** and backtesting.
3. An **interactive dashboard** turns model outputs into store-level actions, boosting **space efficiency** and decision speed.
4. Assortment rules balance **customer satisfaction** and **sales** by ranking SKUs, right-sizing breadth vs. depth, and preserving **core/premium/value** roles.
5. Framework is **scalable** to other categories with the right **data stack** (POS, price/promo, attributes, space) and category-specific constraints.

## Frequently Asked Questions

### What is retail assortment optimization and why is it important for Carrefour’s pet food category?

Assortment optimization uses data to choose the right mix of SKUs for each store, balancing customer demand, sales, and shelf space. In pet food, it reduces out-of-stocks, cuts duplication, and improves space productivity while maintaining choice.

### How do LightGBMRanker and LightGBMRegressor improve pet food assortment planning?

LightGBMRanker ranks SKUs by expected contribution so category managers can prioritize what to keep or add. LightGBMRegressor predicts sales and revenue impact, enabling space-aware, store-specific recommendations that outperform rule-based methods.

### How do you handle product cannibalization when optimizing a pet food category?

Model cross-effects between similar SKUs (brand, pack size, price tier) and constrain the final mix to preserve unique roles. Simulate add/remove scenarios to ensure gains from new items aren’t just offset by losses from near substitutes.

### What data do retailers need to replicate this Carrefour approach?

You’ll need 12–24 months of POS sales, price and promo history, product attributes (brand, size, flavor), store attributes (format, space), and planogram constraints. Optional loyalty or basket data improves cannibalization and cross-sell insights.

### How should assortments differ between hypermarkets and markets?

Hypermarkets benefit from broader ranges, premium tiers, and niche needs because of larger space and diverse shoppers. Markets should focus on a curated core, high-velocity SKUs, and value tiers, minimizing duplication to protect space efficiency.

### How can Carrefour validate the projected 0.4% revenue uplift per store?

Run controlled A/B pilots with matched control stores and monitor revenue per store, margin, and space productivity. Complement with backtesting on historical periods and phased rollouts to de-risk scale-up.

### Can this machine learning framework scale beyond pet food?

Yes—apply the same ranking and regression setup to other categories, tuning features for seasonality, pack-sizing, and perishability. Add category-specific constraints (e.g., freshness windows) to keep recommendations realistic.

### What KPIs should category managers track after the assortment change?

Track revenue per store, gross margin, space productivity (sales per meter), on-shelf availability, and substitution rate. Also monitor SKU churn and customer satisfaction signals like complaints or NPS for early warnings.


---

*Article from [Albert's Deep Dive](https://deepdive.albertschool.com) - Albert School's Journal*