How Kravet built an AI assistant that knows every product and drafts client emails in seconds
Kravet Inc. is a 5th-generation global leader in luxury textiles, furniture, and home décor — with 1,000+ employees and 125,000+ products. Velocity built a RAG-powered internal AI assistant that improved answer accuracy from below 60% to nearly 90%, enabling staff to instantly retrieve product specs, inventory data, and draft client-ready emails.
Challenges
Thousands of internal files contained outdated, conflicting, or unreadable information — causing AI responses to be unreliable and misleading
Product data on the website wasn't fully accessible and didn't support exact-match search by SKU or product code
Unstructured formats — PDF scans, mixed file types, inconsistent naming — blocked reliable knowledge retrieval
AI-generated answers were unpredictable, pulling from inconsistent sources with no way to trace or validate them
Our solution
We built Kravet's internal AI assistant using GPT-4 on Azure OpenAI with Retrieval-Augmented Generation (RAG) — ingesting 1,000+ internal documents, the full Kravet blog, and all 125,000+ product pages. We began with a zero-risk 3-week pilot on Kravet's data as-is, using iterative testing with client-curated questions to identify root causes of inaccuracy and redesign the knowledge pipeline.
Zero-risk RAG pilot
Delivered a 3-week proof-of-concept with built-in RAG — testing on Kravet's existing data to identify failure modes before committing to a full build.
Enterprise knowledge ingestion
Ingested 1,000+ static internal files, the full Kravet blog archive, and all product website pages into a unified, queryable knowledge base.
Iterative accuracy testing
Ran structured testing rounds using client-curated questions to measure accuracy improvements at each iteration and validate each pipeline change.
Data pipeline redesign
Identified problematic data sources — outdated files, PDF scans, inconsistent formats — and collaboratively rebuilt the information pipeline for reliable retrieval.
The results
Accuracy gains achieved
Accuracy improved from <60% to nearly 90% through iterative, data-driven refinement.
Rapid prototype delivery
First functional prototype
delivered in 3 weeks.
Enterprise-scale knowledge
AI assistant now answers questions across 125,000+ products—with details on materials, colors, collections, and specs.
Real-time inventory
Real-time inventory insights delivered through
seamless system integrations.
Automated client communications
AI can now draft client-ready emails, combining product info + inventory checks + tone of voice selection.
Engagements with external consultants can suffer if the external party is very linear in their approach. But the Velocity team was incredibly collaborative and eager to understand our use case. It made our team feel like we were truly partnered on our project. I believe the approach taken by the Velocity team is what allowed us to progress to a successful launch. A less collaborative or more rigid approach would probably have caused us to disengage and assume the pilot as unsuccessful before launch.
Jesse Lazarus, Chief Technology Officer @ KravetVelocity AI Services Used
