Context
- Industry: B2B Services / Digital Media & Business Process Automation.
- Current Process: Prospective clients submit inquiries via website chat or messengers. Sales managers manually conduct initial discovery, clarify budgets, timelines, and decision-makers, and enter lead records into Zoho CRM. Leadership (CEO / Head of Sales) lacks real-time visibility into conversation quality and funnel bottlenecks without performing manual chat audits.
Problem
- Lead Loss Due to Delayed Responses: Sales reps respond with delays outside business hours or during peak workloads, causing warm prospects to drop off.
- Incomplete Qualification & Manual CRM Entry: Leads often enter Zoho CRM missing key qualification data (budget, timeline, core pain point), lowering conversion rates and consuming manual admin time.
- Zero Systematic QA & Slow Reporting: There is no automated audit of whether qualification scripts are followed properly, and compiling weekly executive funnel reports takes hours of manual data aggregation.
Solution
Built an end-to-end, three-stage multi-agent AI system in n8n that automates the entire top-of-funnel lifecycle without human intervention during initial lead processing:
- WF1 — AI Lead Qualification & CRM Capture Agent: Conducts natural, structured discovery conversations, persists session context in PostgreSQL, gathers key qualification criteria (name, company, niche, monthly budget, decision-maker status, primary pain point, timeline, phone number), and automatically creates a structured lead record in Zoho CRM linked via the session UUID.
- WF2 — AI Conversation QA & Insights Analyzer: Runs on a scheduled loop to fetch qualified leads from Zoho CRM, retrieves full conversation transcripts from PostgreSQL (
Chat Memory Manager), audits each dialogue for script compliance and grammar/logic errors, extracts core customer pain points and winning conversion patterns, and logs the findings into Google Sheets.
- WF3 — AI Funnel Analyst & CEO Telegram Reporter: Aggregates weekly CRM metrics (total qualified leads, total pipeline value, stage distribution, and manager workload) alongside QA insights from Google Sheets, generates a validated JSON report via Structured Output Parser, and delivers an HTML-formatted executive digest directly to the CEO in Telegram.
Process
- Prompt Design:
- Engineered distinct system prompts for three specialized agents: a conversational qualification agent with strict guardrails against premature CRM creation before phone capture (WF1), a strict QA auditor constrained to a single database write per session (WF2), and a sales operations analyst outputting Telegram-safe HTML without unsupported tags (WF3).
- API & Database Integration:
- Connected PostgreSQL (
n8n_chat_histories) for persistent conversation memory and cross-workflow session mapping (session_id ↔ Zoho Skype_ID).
- Integrated Zoho CRM API for automated lead creation and pipeline querying, Google Sheets API for structured QA logging, and Telegram Bot API for executive alert delivery.
- n8n Workflow Engineering:
- Implemented batch iteration (
Loop Over Items) with conditional filtering (If), deterministic JavaScript metric aggregation (Code node) to eliminate LLM arithmetic hallucinations, strict JSON schema enforcement via Structured Output Parser, and regex sanitization (<br> stripping) prior to Telegram dispatch.
Results
- 100% Instant 24/7 Response: Reduced initial lead response time from 15–40 minutes down to < 3 seconds around the clock.
- 6+ Hours Saved Weekly: Fully eliminated manual Zoho CRM data entry (~5 minutes per lead) and cut weekly executive report preparation from 3–4 hours to 0 minutes.
- 100% Dialogue Audit Coverage: Scaled conversation quality monitoring from ~10–15% manual spot-checks to 100% automated auditing of every qualified lead, building a searchable database of customer pain points.