The challenge
Inbound leads arrived faster than the sales team could qualify them. Enrichment was manual, scoring was a spreadsheet formula nobody trusted, and routing depended on whoever noticed the notification first.
The cost was not just slow response — good leads were being handed to the wrong rep, and follow-up sequences fired against stale CRM state, so prospects received messages that contradicted conversations they had already had.
Approach
01
Model the existing process honestly
Shadowed the manual qualification flow for a week and wrote down what the team actually checked, including the judgement calls that were not in any documented process. That list became the agent's task decomposition.
02
Agents per responsibility, not per step
CrewAI agents split along enrichment, scoring and routing rather than along workflow steps, so each has one clear contract and can be evaluated independently.
03
CRM as the source of truth, always
Every agent reads live CRM state immediately before acting. No cached snapshots, because a stale snapshot was the root cause of the contradictory-message problem.
04
n8n for the mechanical parts
Triggers, retries, rate limiting and the follow-up sequences run in self-hosted n8n. Agents make decisions; n8n moves data. Keeping that boundary sharp made both halves debuggable.
Architecture
Trigger
n8n webhooks fire on new inbound lead from forms, chat and partner feeds.
Enrich
Enrichment agent pulls firmographic and technographic signals, deduplicates against existing CRM records.
Score
Scoring agent applies the qualification rubric captured during discovery and writes a score plus a written rationale to the record.
Route
Routing agent assigns owner by territory, segment and current load, then n8n triggers the matching sequence.
Observe
Every decision is logged with its rationale so sales leadership can audit why a lead was scored the way it was.
Results
$340KAttributed pipeline
1.8xQualified leads
48%Less manual work
900+Daily automated triggers
<3 minLead response time
What I took from it
Writing the rationale to the CRM record alongside the score did more for adoption than the score itself. Reps trusted the system once they could see why, and started correcting it where it was wrong — which gave us labelled data for tuning.
n8n automation cut lead response times from 4 hours to under 3 minutes. Their n8n expertise is on another level - workflows our internal team would never have conceived.
Stack
Agents
Automation
Models
Integration