Email Intent Classifier

AI reads inbound emails, scores buyer intent across 7 categories, and routes to Pipedrive — deals, activities, and notes created automatically.

AI reads inbound emails, scores buyer intent across 7 categories, and routes to the correct Pipedrive pipeline stage. 30-node swarm. $0.116/email blended cost. This came from a support-turned-sales handoff problem — replies containing buying signals were sitting in shared inboxes for hours because nobody could tell intent from subject lines alone.

Last updated March 8, 2026

Sales teams send and receive thousands of emails per week, but fewer than 10% receive systematic follow-up analysis. Response time, sentiment shifts, and intent signals buried in email threads go undetected. Automated communication intelligence catches patterns that human review at scale cannot.

4-Agent PipelineinputGmail Trigger01ResearcherSender web_search02Classifier7-class intent03Routerconf >= 0.7?high confPipedrivelow confReviewspamLog & Skip$0.116/email blendedITP-measured across 20 emails

Four Agents. Seven Intents. Emails Routed to Pipedrive Automatically.

The Researcher

Step 1The Researcher

Tier 1 Reasoning

Receives inbound email data from Gmail and conducts web-based sender research — company intelligence, role seniority, and deal context from Pipedrive. Outputs a structured research payload that gives the Classifier the context it needs for accurate intent scoring.

The Classifier

Step 2The Classifier

Tier 1 Reasoning

What does The Classifier actually decide? Scores each email across 7 intent categories: DEMO_REQUEST, PRICING_INQUIRY, OBJECTION, RENEWAL_INTEREST, CHURN_SIGNAL, GENERAL_INQUIRY, and SPAM_OR_NOISE. Outputs intent classification with a confidence score (0–1.0) and chain-of-thought reasoning.

The Router

Step 3The Router

Conditional

This step exists because raw data alone is not enough. Three-way confidence gate: high confidence (≥ 0.7) routes to Pipedrive CRM — deals, activities, and notes created based on intent category. Low confidence routes to human review with full classification context. SPAM/noise logs and skips with zero additional LLM calls.

The Syncer

Step 4The Syncer

Integration

Without this step, upstream analysis sits idle. Writes to Pipedrive based on intent: DEMO_REQUEST, PRICING_INQUIRY, and RENEWAL_INTEREST create Deals + Activities. OBJECTION and CHURN_SIGNAL create Notes on existing deals. GENERAL_INQUIRY creates an Activity. Non-blocking writes — pipeline never stalls on CRM errors.The web search API fee is $0.03. The token cost from 30K-40K injected context tokens adds $0.06. The search fee is one-third of actual cost.

That's the full pipeline. Here's what it intentionally does NOT do — and why those boundaries exist.

What It Does NOT Do

×

Does not send reply emails — output is Pipedrive records and classification data

×

Does not process email attachments — classification is based on subject and body text only

×

Does not support CRMs other than Pipedrive in v1 — Pipedrive-specific deal, activity, and note creation

×

Does not handle email threading — each inbound email is classified independently

With those boundaries clear, here's everything that ships when you purchase.

The Complete Customer Success Bundle

8 files.

CHANGELOG.mdVersion history
README.mdSetup and configuration guide
blueprint_dependency_matrix.mdThird-party service dependencies
confidence_threshold_guide.mdConfidence threshold guide
email_intent_classifier_v1_0_0.jsonn8n workflow (main pipeline)
intent_taxonomy.mdIntent taxonomy
system_prompt_classifier.txtClassifier system prompt
system_prompt_researcher.txtResearcher system prompt

The technical specifications below are ITP-measured, not estimated.

Tested. Measured. Documented.

All values below are from Independent Test Protocol (ITP) testing — not estimates, not projections. Measured across 20 test emails spanning all 7 intent categories, borderline confidence cases, and edge cases.

Workflow Nodes

30 specialized n8n nodes

Blueprint Quality Standard

BQS v2 — 12/12 PASS

Agent Roles

4 (Researcher, Classifier, Router, Syncer)

Credentials Required

Anthropic API + Pipedrive API + Gmail OAuth2

Bundle Files

9 files

Cost per Email (ITP-Measured)

$0.116 (ITP-measured blended average)

Inspection & Test Plan

EIC-01 through EIC-08 + U-01 through U-06 — 14/14 PASS

Classification Consistency (3 Runs)

Max variance 0.02 across 3 runs (threshold <= 1.0)

n8n Versions Tested

Self-hosted

Tested on n8n v2.7.5, March 2026

Email Intent Classifier v1.0.0 — Technical Reference━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━Pipeline:     Gmail Trigger → Researcher (web_search) → Classifier (7-class intent) → 3-Way Router → Pipedrive / Human Review / LogTopology:     Modular Agentic Swarm (4 agents, linear-then-branching pipeline)Routing:      Confidence ≥ 0.7 → CRM action | < 0.7 → human review | SPAM/NOISE → logIntents:      DEMO_REQUEST | PRICING_INQUIRY | OBJECTION | RENEWAL_INTEREST | CHURN_SIGNAL | GENERAL_INQUIRY | SPAM_OR_NOISENode Count:   30LLM:          Claude Opus 4.6 (Researcher + Classifier — Tier 1 Reasoning)Cost:         $0.116/email (ITP-measured blended avg)ITP:          14/14 PASS, 20/20 intent accuracy (100%)BQS:          12/12 PASSTool A:       Gmail (email trigger)Tool B:       Pipedrive (CRM routing)

What You'll Need

Platform

n8n 2.11.2+

Est. Monthly API Cost

$10–60/month depending on email volume

Credentials Required

  • Anthropic API
  • Pipedrive API
  • Gmail OAuth2

Services

  • Pipedrive CRM
  • Gmail account

Setup Track

Quick Start

~15 min

All credentials live, n8n running

Full Setup

1–2 hrs

Needs API config + tables

From Scratch

2–4 hrs

No n8n, no credentials

Email Intent Classifier v1.0.0

$199

one-time purchase

What you get:

  • 30-node n8n workflow (.json) — you own it
  • 2 tested agent system prompts (ITP-validated)
  • 8 SDC inter-agent schemas
  • Intent taxonomy with 7 categories and examples
  • Confidence threshold tuning guide
  • Error handling matrix (26 failure modes documented)
  • Dependency matrix with ITP-measured costs
  • README setup guide (15 minutes)
  • All sales final after download

Frequently Asked Questions

What are the 7 intent categories?+

DEMO_REQUEST (meeting/demo requests), PRICING_INQUIRY (pricing/proposal discussions), OBJECTION (sales objections or pushback), RENEWAL_INTEREST (renewal or expansion signals), CHURN_SIGNAL (cancellation or downgrade threats), GENERAL_INQUIRY (other business emails), and SPAM_OR_NOISE (newsletters, recruitment, automated notifications).

What happens when confidence is below 0.7?+

Low-confidence classifications route to the human review path. The pipeline returns the full classification context — intent category, confidence score, reasoning, and sender research — so a human can verify and take action. No CRM records are created automatically for low-confidence emails.

How does the SPAM path save cost?+

SPAM_OR_NOISE emails are detected early in the pipeline. Once classified as spam, the pipeline logs the result and skips all downstream processing — no Pipedrive writes, no additional LLM calls. SPAM path executes in under 100ms with zero additional cost beyond the initial classification.

What Pipedrive records does this create?+

Depends on intent: DEMO_REQUEST, PRICING_INQUIRY, and RENEWAL_INTEREST create a new Deal + Activity + Note. OBJECTION and CHURN_SIGNAL add a Note to the existing deal. GENERAL_INQUIRY creates an Activity only.

What credentials are required?+

Three credentials: Anthropic API key (used by Researcher and Classifier agents), Pipedrive API token (for deal, activity, and note creation), and Gmail OAuth2 (for email trigger). The README includes step-by-step setup for all three.

How does the Researcher gather sender intelligence?+

The Researcher uses web_search with the sender email domain and name as anchors. It investigates the sender's company, role, recent news, and cross-references against existing Pipedrive contacts. Output includes a research confidence score — unknown senders get lower confidence but are still classified.

Can I change the confidence threshold?+

Yes. The threshold is a single value in the Router node (default: 0.7). Lower it to route more emails to CRM automatically, or raise it for stricter human review gating.

What does this blueprint NOT do?+

It does not send reply emails — output is Pipedrive records and classification data. It does not process attachments — classification is based on email subject and body text. It does not support CRMs other than Pipedrive in v1.

How much does each email classification cost?+

Approximately $0.116 per email (blended average). Actionable emails cost $0.134, SPAM emails cost $0.000 beyond initial classification. At 500 emails per month, total API cost is approximately $58.

Is there a refund policy?+

All sales are final after download. Review the Blueprint Dependency Matrix and prerequisites before purchase. Questions?

Read the full guide →

Related Blueprints