Meerkat LLC mascotMeerkat Book a call
— Case Study · CRM Lead-Gen Pipeline

Every lead lands. None slip through.

This is the end-to-end inbound engine Meerkat runs in production on its own lead flow — the same architecture offered to clients. Webhook capture, enrichment, Claude-based intent classification, deduplicated CRM routing, and alerting replace a fully manual triage process, so a human only touches a lead once the system has already classified, scored, and enriched it.

The design guarantee: no lead is ever silently dropped. Every failure path ends in an alert, not a void.

How it works

  1. Capture everything. Web forms, inbound email, and voice-agent outcomes all arrive by webhook into one intake pipeline — one front door, every channel.
  2. Claude classifies intent. Each inquiry is classified (hot / nurture / spam / support) using versioned prompt templates with few-shot examples, and a personalized first-touch reply is drafted for human approval.
  3. Enrichment before routing. Company and role context is appended to the record, so the CRM entry a rep opens already tells the story.
  4. Deduplicated CRM upsert. Leads are upserted with dedup keys and assigned a pipeline stage — re-submissions and double-fired webhooks can't create duplicates.
  5. Routing + speed-to-lead alerting. Routing rules notify the right owner in Slack with a one-paragraph AI summary, and follow-up goes out within minutes of capture.
  6. Guardrails on every AI call. Confidence thresholds route ambiguous leads to manual review; every LLM call is wrapped with timeout, retry, and token-cost logging; malformed payloads land in a dead-letter queue with an alert.

What it replaces

Before: manual triage — someone reads each inquiry, guesses at priority, copies it into the CRM, and hopes to reply before the lead goes cold. After: classification, scoring, enrichment, CRM entry, and owner notification happen in the seconds after the lead arrives; reps only touch leads worth their time.

Honesty note, Meerkat-style: this pipeline runs in production on Meerkat's own inbound lead flow (lead capture, speed-to-lead, signal listener, round-robin router, and CRM hygiene workflows — all part of the 137 production workflows Meerkat operates). Per-client before/after numbers get published here once pilot clients complete their first full month — no invented metrics on this site, ever. Want proof instead? The quoting-agent live test shows how Meerkat validates systems before claiming anything.

n8nClaude (Anthropic) APIWebhook captureEnrichmentGoHighLevel / Close / HubSpotSlack alertingDead-letter queueToken-cost logging

More case studies

AI quoting agent → Voice AI qualification agent → Full portfolio →