Case Study
Grupo MCoutinho gave every seller one place to ask, and answered in four weeks.
Cloud202 built a production-grade, AWS-native conversational AI Sales Assistant inside Microsoft Teams, putting live Pulse CRM data and 7,000+ Portuguese sales documents behind a single natural language interface, scoped precisely to each individual seller.
4wk
Idea to Production
20-30min
Saved Per Seller Per Day
7,000+
Documents Indexed
0/8
Security Test Failures
Customer Profile
Company: Grupo MCoutinho
Industry: Automotive - Vehicle Sales, Workshops & Aftersales
Location: Portugal
Business: One of Portugal's leading automotive groups, operating vehicle sales, workshops, and aftersales services across multiple brands.
The Challenge: Multiple systems, one customer question.
Grupo MCoutinho's sales teams navigated multiple disconnected systems to answer a single customer question, slowing down sellers and breaking the sales conversation. A seller mid-conversation had to leave it to go looking, and by the time the answer surfaced, the moment had passed.
The group needed a solution that could surface live CRM data and 7,000+ Portuguese sales documents in a single, natural language interface, scoped precisely to each individual seller. Not a shared search box over a shared corpus, but an assistant that knew whose data it was allowed to return.
What We Built: An assistant that lives where sellers already work.
A production-grade, AWS-native conversational AI Sales Assistant inside Microsoft Teams, delivered in 4 weeks. Sellers ask questions in natural language and get instant answers drawn from two sources.
1. Live Pulse CRM Data - Leads, customers, vehicles, opportunities, complaints, and workshop bookings, retrieved in real time. The assistant is read-only against the CRM, so it can never write, alter, or delete a record.
2. A Semantically Indexed Knowledge Base - 7,000+ Portuguese sales documents, filtered by brand and month, so a question about one brand's current promotion never returns last quarter's pricing for another.
3. Per-Seller Identity Binding - Seller identity is bound at tool construction, not exposed as a model-visible parameter, so nothing a user types can change whose data is returned.
Architecture
Agent Runtime
Amazon Bedrock AgentCore - ARM64, Claude Sonnet 4.6
Inference
EU Cross-Region Inference profile (eu-central-1)
Memory
AgentCore Memory - per-seller, session + long-term recall
Knowledge Base
S3 Vectors, Titan Embeddings v2, Amazon Rerank
Document Ingestion
Container Lambda, SQS queue, Dead-Letter Queue
CRM Integration
Read-only Pulse REST API via Secrets Manager
Channel
Microsoft Teams via Bot Framework adapter Lambda
Safety
Amazon Bedrock Guardrails on every model call
Infrastructure
Terraform - one shared module across dev and production
Brand and month tagging at ingestion ensures answers always reflect current pricing and promotions. Stale document retrieval is eliminated at the ingestion layer, not patched over in the prompt.
Outcomes: What changed for the business.
20-30min
Seller productivity. Estimated 20-30 minutes saved per seller per day, time that returns directly to customer-facing activity.
7,000+
Knowledge always current. As the document library grows, the knowledge base scales automatically with no manual curation.
0/8
Zero data leakage risk. Per-seller scoping enforced at the tool layer, not the prompt layer. Security testing across 8 adversarial scenarios - cross-seller data access, identity override attempts, company-wide data dump requests - returned zero failures.
4wk
Enterprise-ready from day one. Bedrock Guardrails on every model call, hashed seller identifiers in logs, and OWASP Top 10 for LLM controls validated against the live agent.
AWS Partnership
Delivered in partnership with the Amazon Spain account team as part of Cloud202's AWS GenAI Competency programme.
Cloud202 is an AWS Advanced Tier Services Partner holding the AWS Generative AI Competency, EKS Service Delivery, CloudFormation Service Delivery, and APN Customer Engagement Program designations.
Technologies
See your best AI idea in production.
Contact Cloud202 to discuss how Qubitz AI can automate your workflows on AWS.