From 100 Catalogues to One Conversation — AI-Powered Sales Offer Automation

At Patiris, a 7-person sales team creates approximately 80 offers per month for customers in the premium building materials, tiles, and bathroom fittings market. Each offer requires cross-referencing product data scattered across ~100 supplier PDF catalogues, Excel price lists, web-based catalogues, and 18,000 SoftOne ERP help documents — a process that can take up to 20 minutes just to find a single product or procedure.

The challenge was compounded by the fragmented nature of the data: PDFs from suppliers like FANTINI, KRONOS, NIC, NICDESIGN, LIVING, and MADE arrived in different formats, languages, and layouts. Seven salespeople repeated the same lookup work independently — with no shared product intelligence, pricing memory, or customer history across the team. Each new supplier added to the portfolio multiplied the manual effort without adding capacity.

The growing trend of web-only supplier catalogues added yet another source to check. Without a unified search capability, scaling the business meant scaling headcount — a bottleneck that the team was determined to break.

LCM Go Cloud designed and deployed a GenAI-powered Sales Assistant on AWS — built on Amazon Bedrock with Claude Sonnet 4.6 for conversational intelligence, Qwen 3 VL 235B for vision-based catalogue ingestion, and the AgentCore Runtime with Strands SDK for the agent framework. The platform ingests all supplier catalogues, price lists, and ERP documents into a Bedrock Knowledge Base backed by OpenSearch, enabling salespeople to search products across all suppliers by SKU, description, or natural-language query — and receive automated pricing with customer-specific discounts.

The architecture is fully serverless — no EC2 instances, no VPC. CloudFront serves the SPA, Amazon Cognito handles authentication with role-based access, API Gateway routes requests to BFF Lambdas for CRUD operations and a streaming endpoint for agent conversations, while DynamoDB stores products, customers, offers, and session data. A cross-region design places document ingestion in eu-west-1 (for Qwen model availability) and the chat interface in eu-central-1 (Frankfurt, for low latency to Greek users).

Following a successful proof of concept that validated all six core capabilities — smart product search, automatic pricing, one-click offer building, customer history awareness, kit completion, and bilingual Greek/English support — LCM Go Cloud delivered the production platform with 23 reusable Terraform modules and a zero-secrets GitLab CI/CD pipeline using OIDC. The platform now generates ready-to-send offer slips with automated pricing, VAT, shipping, and XLSX export — compressing what used to take hours into a single conversation.