SIZE
AI-Powered Cataloguing for 10,000+ SKUs
80%
Reduction in manual tagging
6 weeks
Delivery Time
Streetwear & Sneaker Retail
Industry
size.co.uk
Client
Overview
SIZE is one of the UK's largest streetwear and sneaker retailers with over 10,000 active SKUs across their online store. Their product cataloguing process was entirely manual — a team of 4 people spent an average of 12 hours per week tagging products with categories, attributes, and search keywords. Errors were frequent, inconsistent tagging hurt search relevance, and new product drops were delayed by the bottleneck.
The Challenge
The existing workflow relied on spreadsheet-based tagging with no standardisation. Product images arrived from suppliers in inconsistent formats. The team needed a system that could auto-classify products from images and supplier data, apply consistent taxonomy, and integrate directly into their Shopify-based storefront — without disrupting the existing product pipeline.
How We Built It
Our Approach
Audit & Taxonomy Design
We audited 2 years of product data, identified 340+ unique attribute combinations, and built a standardised taxonomy of 85 categories with hierarchical tagging rules. This became the foundation for the AI model's classification targets.
Multi-Modal Classification Pipeline
Built a pipeline combining GPT-4 Vision for image-based classification with structured text extraction from supplier CSV feeds. The system cross-references both signals to assign category, colour, material, gender, and 12 other attributes with confidence scoring.
Human-in-the-Loop Review Interface
Developed a lightweight internal review dashboard where the team only needs to verify low-confidence classifications (typically <15% of items). High-confidence items flow straight into the catalogue — cutting the team's review workload by 80%.
Shopify Integration & Go-Live
Integrated the pipeline directly into their existing Shopify product upload workflow via a custom app. New products are auto-tagged within minutes of upload. Deployed with a 2-week parallel run alongside the manual process to validate accuracy.
The Results
What We Delivered
80%
Reduction in manual tagging time
Team went from 12 hours/week of manual tagging to reviewing only edge cases — about 2.5 hours/week.
94.7%
Classification accuracy
Across 10,000+ SKUs, the system matched or exceeded human accuracy on product categorisation.
3x
Faster product drops
New product lines now go live within hours of supplier delivery, not days.
Delivery Timeline
Shipped in 6 weeks
Data audit and taxonomy design
AI classification pipeline development
Review dashboard and Shopify integration
Parallel validation run and go-live
Client Feedback
“We were spending entire days just tagging products. Now the system handles it automatically and our team focuses on merchandising strategy instead.”
Operations Lead
SIZE
Tech Stack