Case study · Ecommerce
72% faster listings. 18% lift on search-to-cart.
A specialty retailer ($180M GMV, 230K SKUs) replaced its manual listing pipeline with an AI catalog-enrichment agent. Backlog cleared in 6 months. Returns down. CVR up.
- Industry
- Ecommerce / specialty retail
- Region
- US + Canada
- Size
- 230K SKUs, $180M GMV
- Stack
- Shopify Plus · Postgres · Claude · OpenSearch
- Engagement
- 10 weeks build + 4 weeks rollout
Results
Measured outcomes after 6 months.
72%
Reduction in time per new SKU listing
18%
Lift in search-to-add-to-cart on enriched SKUs
11%
Drop in return rate from spec mismatches
4.5x
ROI inside 6 months
The challenge
Catalog ops couldn't keep up with the supplier pipeline.
- 01
Catalog backlog of 40,000 partially-listed SKUs. Each new SKU took 35-50 minutes of merchandiser time to research, write copy, normalize attributes against the taxonomy, and confirm category fit. Backlog kept growing faster than the team could clear it.
- 02
Supplier data arrived in 11 different formats — PDFs, XLSX, CSV, scanned spec sheets, supplier portal exports. Manual reconciliation against the internal product schema was where most of the merchandiser time went.
- 03
Returns were elevated on 15-20% of SKUs traceable to under-described or mis-described products. Buyers got something that didn't match their expectation built from the product detail page.
- 04
Search relevance suffered because half the catalog had thin or missing structured attributes (material, fit, sizing system, certifications). The internal search couldn't filter on what wasn't in the data.
The solution
Four capabilities of the enrichment agent.
01
Supplier-document extraction agent
A retrieval + extraction agent reads supplier PDFs and images, extracts structured attributes against the internal product schema with per-field confidence scores, and routes low-confidence extractions to a merchandiser review queue. Handles the 11 supplier formats with one pipeline.
02
Merchant-voice description generator
Given structured attributes, the agent generates product descriptions in the retailer's brand voice using examples from existing high-converting listings. Merchandisers edit instead of writing from scratch — typically 5-8 minutes per SKU vs. 35-50.
03
Taxonomy normalization + category fit
The agent classifies each SKU into the internal taxonomy, suggests parent/child relationships, flags potential miscategorizations against similar existing listings, and detects when a new SKU might warrant a new category leaf.
04
Spec-mismatch detection on returns
Cross-references return-reason text from CS records against the listed product attributes. Surfaces patterns like 'returns citing fit issues on SKU family X' to the merchandising team weekly with linked evidence, so the underlying listings get fixed.
"We were going to hire 4 more catalog ops people. Instead we cleared a 40,000 SKU backlog with the team we had — and the listings are better than the ones we wrote ourselves."
VP Merchandising
Specialty Ecommerce Retailer ($180M GMV)
Related services
AI Modernization →Embedding AI agents into existing product and operations surfaces. From $50,000.