Ecommerce ops · Production

Wayfair builds LLM-powered style compatibility labeling pipeline on Google Cloud with Gemini 2.5 Pro

The problem

Wayfair's catalog of tens of millions of SKUs lacked scalable style-compatibility labels: traditional recommendation systems relied on behavioral signals that missed latent stylistic relationships, while human annotation was accurate but too slow and costly to cover the catalog at scale.

Workflow diagram · grounded in source
1
Pipeline pulls imagery and metadata
integration
“The pipeline pulls product imagery and metadata, builds the structured prompts, sends them to the model”
2
Gemini 2.5 Pro classifies pair
ai_action
“The model processes both product imagery and descriptive text (title, class, and romance copy), using structured prompts enriched with a few carefully selected examples to ground its style judgments. It outputs concise JSON containing th…”
3
Results stored at scale
output
“stores the results—enabling rapid, consistent, and scalable style-compatibility labeling that can be used to evaluate and improve recommendation algorithms”
4
Expert holdout evaluation
validation
“We evaluated the system against a hold-out set of expert judgments, treating the human labels as the ground truth”
5
Labels improve recommendations
feedback_loop
“Those labels now help evaluate—and ultimately improve—the quality of our product recommendations”
Reported outcome

The LLM-powered pipeline improved annotation accuracy by 11% over the initial generic prompt and generates style compatibility labels far faster than manual annotation, enabling scalable offline evaluation and future improvement of recommendation algorithms.

Reported metrics
Annotation accuracy improvement11%
Label generation speed vs manual annotationfar faster than manual annotation
Search and recommendation qualitysignificant improvements in search and recommendation quality
Reported stack
Gemini 2.5 ProGoogle Cloud
Source
https://www.aboutwayfair.com/careers/tech-blog/teaching-wayfairs-catalog-to-see-style-an-llm-powered-style-compatibility-labeling-pipeline-on-google-cloud
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The LLM-powered pipeline improved annotation accuracy by 11% over the initial generic prompt and generates style compatibility labels far faster than manual annotation, enabling scalable offline evaluation and future…

What tools did this team use?

Gemini 2.5 Pro, Google Cloud.

What results were reported?

Annotation accuracy improvement: 11%; Label generation speed vs manual annotation: far faster than manual annotation; Search and recommendation quality: significant improvements in search and recommendation quality (source-reported, not independently verified).

How is this ecommerce ops AI workflow structured?

Pipeline pulls imagery and metadata → Gemini 2.5 Pro classifies pair → Results stored at scale → Expert holdout evaluation → Labels improve recommendations.

Related ecommerce ops cases