AI lead processing
LeadFlow AI
A lead from the website is sorted by AI within seconds, lands in the CRM and reaches the manager in Telegram with a ready draft reply — the email to the client goes out in one tap.
- Category
- AI lead processing
- Year
- 2026
- Stack
- n8n, Gemini API, PostgreSQL, pgvector, Telegram Bot API, three.js
The task
A small business gets leads from its website, messengers and email, and a manager reads each one by hand, copies it into a spreadsheet and writes a reply — 5–15 minutes per lead, while the hot ones get lost in spam. We showed how to automate this using a fictional furniture company, Dub i Stal.
What we did
The quiz landing page calculates the price from the price list as the client answers. Then n8n picks the lead up: an LLM detects its type, urgency, budget and language, and filters out spam, duplicates and prompt injection separately. The lead is stored in PostgreSQL and mirrored to Google Sheets, and the draft reply is built from the company's knowledge base (RAG), with every number checked against it. The manager gets a Telegram card with the gist, the budget and Send / Reject buttons — the email goes out from the company's domain only after a person approves it. Alongside there is a mini CRM with a funnel, notes, analytics and Excel export, plus a weekly report for the owner in Telegram.
Result
On test leads, sorting, saving to the CRM and drafting a reply take about 7 seconds and ≈ $0.0015 per lead. On 46 evaluation leads (Ukrainian, Russian, English, mixed language, spam, injection attempts) the type is detected in 100% of cases and spam in 97.8%. The logic is covered by 26 automated tests.
Mobile version