Consultants drowned at peak.
A large online store: consultants overloaded with routine product selection and customer education — peaks reach 100–500 requests an hour.
A Telegram-widget recommendation agent with RAG over the catalog, integrated with the client's ERP and deployed on their own servers.
What was in the way
- ✗Peak loads of 100–500 requests per hour
- ✗Consultants spent their time on routine product selection
- ✗Product knowledge lived in the catalog and in heads — not in the sales channel
What the system does
Product selection
The agent asks clarifying questions and recommends items from the catalog.
Catalog-grounded answers
RAG search over live data — answers grounded in the catalog, no inventions.
Unloading the team
The agent closes routine consultations and hands the complex ones to people.
How we roll it out.
Requests and access analysis
We study historical requests, verify CRM and API access, and score the automation potential.
A slice of real traffic
The agent runs on part of the live requests with weekly syncs — real dialogues, not a demo.
Metrics and scale-up
We evaluate volume, response speed and escalation share — then expand the agent to full traffic.