AI assistant for a sales team — case
How we built a GPT assistant that cuts incoming-lead handling time from 40 to 6 minutes and plugs into Bitrix24.
Client — a trading company with around 200 leads a day. Managers spent up to 40 minutes on each lead: clarifying parameters, finding similar items in the catalog, calculating price, replying.
The task
Automate the initial reply so the manager gets a lead already with a preliminary estimate and options — and spends time only on the final decision.
Solution
We set up n8n with a chain: lead → request classification → catalog search via vector index → draft answer via Claude Sonnet → push into Bitrix24 as a task checklist.
Technical details
Vector search on pgvector with catalog embeddings. Claude uses tools (function calling): look up an item, calculate delivery, check stock. Replies land in the lead card in Bitrix24 as a comment plus a checklist.
Outcome
Lead-handling time dropped from 40 to 6 minutes. Lead-to-payment conversion grew from 9% to 14%. Integration paid off in the first month. Next step — automated outbound emails with personalization.