Leroy Merlin (Adeo)
genAI enrichment of product attributes and pages
With Gemini and Artefact, Adeo (Leroy Merlin) raised the attribute coverage of its product pages from 27% to 87%, with over 96% accuracy and 63% of products above 96% precision.
Key points
- Automated enrichment of product pages with generative AI, together with Artefact.
- DistilBERT classification then attribute extraction by Gemini with self-verification.
- Attribute coverage rose from 27% to 87%, with over 96% accuracy.
- Evidence level B, live status confirmed in production.
Objective
Make product pages complete and reliable at scale, to avoid losing sales because of thin product content and to improve search and filtering on the site.
The deployment
Adeo, the parent company of Leroy Merlin, industrialized the enrichment of its product pages with generative AI, in partnership with Artefact and on Google Cloud. The catalog covers around 3,600 categories, with 50 to 60 attributes on average per product and a library of 11,000 attributes. A DistilBERT classification model sorts the products, then Gemini extracts the attribute values with a built-in self-verification step. Attribute coverage rose from 27% to 87%. On a measured batch, around 32,000 predictions were made, over 20,000 of them fully automated, with over 96% accuracy and an error rate of 3.6%, versus roughly 8% for a human. 63% of products reach precision above 96%.
Results Proof B
Quantified case study published by the integrator Artefact and by Google Cloud (official interested sources), with measures of coverage, accuracy, and error rate on a production batch. Figures come from the project partners, not from a third-party audit, which places it at B.
How it works
Documented architectureThe stack in detail
- llm Google Gemini Google's Gemini models used to extract product attribute values, with a built-in self-verification step.
- llm DistilBERT Open source classification model (a distilled version of BERT) that sorts each product into one of the 3,600 reference categories.
- plateforme Google Cloud Cloud foundation on which the Adeo catalog enrichment pipeline runs.
- integrateur Artefact Implementation partner for the classification and attribute extraction pipeline.
- infra Pipeline custom Adeo In-house orchestration: a reference of 11,000 attributes, confidence thresholds, and handoff to human review.
How it runs, concretely
For ops teams-
1Classification AI
DistilBERT sorts each product into one of the 3,600 reference categories.
-
2Attribute extraction AI
Gemini extracts the attribute values from the product documents and labels.
-
3Self-verification AI
A built-in verification step filters out low-confidence predictions before automation.
-
4Review and handoff data team
Confident predictions are automated, uncertain cases go to human review before publication.
The quality of the product sources (supplier documents, labels) and the attribute reference. If the source is thin, extraction degrades and the error rate rises.
How your customers perceive this type of use
Sourced studiesUn ecart net separe les annonceurs des consommateurs : 77% des annonceurs voient l'IA positivement contre 38% des consommateurs (Yahoo/Publicis, 2024). Les mesures implicites confirment le rejet declare : en EEG, les pubs generees par IA produisent une activation memorielle plus faible que les pubs traditionnelles et sont decrites comme agacantes, ennuyeuses et confuses (NIQ, 2024). La disclosure a un effet ambivalent : elle augmente fortement la confiance quand elle est remarquee (Yahoo/Publicis), mais 27% des jeunes consommateurs disent faire moins confiance a une entreprise dont la pub est creee par IA (IAB, 2024).
Acceptance conditions
- Une disclosure visible : quand la mention IA est remarquee, la confiance globale envers l'entreprise augmente de 96% (Yahoo/Publicis 2024)
- Une qualite visuelle suffisante : les visuels IA de basse qualite augmentent l'effort cognitif et distraient du message (NIQ 2024)
Red lines
- Le contenu IA non declare puis identifie : 72% des consommateurs disent que l'IA rend l'authenticite difficile a etablir (Yahoo/Publicis 2024) et les marques utilisant des pubs IA sont plus souvent jugees inauthentiques ou non ethiques par les consommateurs que par les dirigeants (IAB 2024)
- Les mannequins et personnes generes par IA : 46% des consommateurs n'en veulent pas dans la publicite, l'inquietude premiere etant les standards de beaute irrealistes (Attest 2025)
Sources: Yahoo / Publicis Media (terrain Ebco) 2024 · IAB (avec Attest) 2024 · NIQ (NielsenIQ) 2024 · Attest 2025
How to replicate
Inference, not sourcedData prerequisites
- Attribute reference and category taxonomy
- Usable supplier product documents and labels
- Validation set to measure precision and error
Org prerequisites
- Rule for automatic handoff to human review on uncertain cases
- Catalog team to adjudicate edge cases
- Coverage and error metrics tracked
Possible stack
- Product classification model
- Generative LLM for attribute extraction
- Self-verification step and confidence threshold
The plan, step by step
- Step 1Pick a high-volume category with thin pages, audit the attribute reference and the quality of supplier sources.Deliverable: Pilot scope defined plus annotated validation set
- Step 2Set up product classification (DistilBERT-type model) on the category reference.Deliverable: Classifier evaluated on the validation set
- Step 3Connect LLM attribute extraction with self-verification and a confidence threshold.Deliverable: Extraction pipeline with error rate measured against human entry
- Step 4Wire the handoff: confident predictions automated, uncertain cases sent to a human review queue.Deliverable: Tooled review queue and documented handoff rule
- Step 5Move the pilot category into production, track coverage and error rate, decide on extension.Deliverable: Coverage/precision dashboard and extension plan for the next categories
First step: Pick a high-volume category with thin pages, connect classification then genAI extraction, and set an automation threshold with review of uncertain cases.
The same pattern, proven in other industries
Intuit
done-for-you genAI assistant integrated into the product to reduce friction and convert
L'Oreal (marques ModiFace)
AR virtual try-on (synthetic makeup rendering on the face)
Gatorade
self-service genAI product personalization on a D2C site
Sources
- S1 ADEO: Improving product referencing speed and quality with AI Interested party archive pending
- S2 ADEO case study - Google Cloud Interested party archive pending
- S3 Leroy Merlin entraine ses vendeurs avec une IA conversationnelle Secondary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.