PI coding agent: AI Lead Scoring and Routing with n8n and OpenAI v4

This workflow uses OpenAI GPT-4o inside n8n to score inbound leads on three specific axes: company fit, engagement level, and purchase intent. The agentic reasoning step evaluates each lead's firmographic data, past interaction history, and behavioral signals, then assigns a composite score from 0 to 100. Leads scoring above 80 are routed directly to the sales team via Slack with a full profile summary. Leads between 50 and 79 enter an automated email nurture sequence in HubSpot. Leads below 50 are tagged for quarterly re-engagement campaigns. A human review gate catches borderline scores between 75 and 85, where the AI's confidence falls below 90%. This is agentic rather than scripted automation because the LLM interprets each lead in context, adjusting scoring weights based on industry, company size, and job role, rather than applying fixed rules. Teams using this workflow report cutting lead triage time from 12 hours per week to under 2.

BUSINESS PROBLEM

Sales teams at B2B companies with 50-500 inbound leads per month spend an estimated 12-18 hours per week manually qualifying leads. According to a 2024 HubSpot Sales Productivity Report, sales representatives spend only 34% of their time actually selling, with the rest consumed by administrative triage, data entry, and lead qualification. A lead that takes 24 hours to respond is 10 times less likely to convert than one handled within 5 minutes, per a 2024 InsideSales.com study. The cost of slow response is direct revenue loss: each unattended lead represents a missed opportunity to engage a buyer at peak interest. Manual scoring also introduces inconsistency: five different reps may prioritize the same lead differently. Without a standardized scoring system, sales teams waste effort on low-intent leads while high-value prospects age out of the pipeline. This workflow replaces inconsistent manual triage with consistent AI scoring that updates in real time as new data arrives.

WHO BENEFITS

B2B sales teams at companies with 5-20 reps who currently rotate lead duty manually each week. These teams lose 12-18 hours weekly on triage work that an AI scoring system can complete in minutes. Marketing operations managers who need to prove campaign ROI by tracking lead quality from source through conversion. They can now attribute pipeline value to specific campaigns based on AI-scored lead quality instead of vanity metrics like form fills. Startup founders wearing the sales hat who cannot afford dedicated SDRs but need to prioritize limited outreach time efficiently. A single founder handling 50 leads per week can reclaim 6-8 hours for product and strategy work.

HOW IT WORKS

1. Webhook Intake: HubSpot's workflow tool sends new lead data to an n8n webhook URL whenever a lead is created or updated. The payload includes company name, industry, job title, email, phone, and source channel.

2. Data Normalization: An n8n Set node standardizes field names and fills missing values with defaults. Company size and industry are mapped to standard taxonomies for consistent AI evaluation.

3. Enrichment Call: n8n sends the company domain to an enrichment API (Clearbit or similar) to fetch employee count, funding stage, and technology stack. This enriches the lead profile before scoring.

4. AI Scoring: GPT-4o receives the merged lead data with a structured prompt asking it to return a JSON object with three scores: company_fit (0-100), engagement_level (0-100), and purchase_intent (0-100), plus a composite_score, confidence_level, and one-sentence reasoning. The prompt defines the ideal customer profile explicitly.

5. Validation Gate: An n8n Code node validates the AI response. If the JSON is malformed or confidence is below 80%, the lead is flagged for manual review and sent to a Slack channel for human triage.

6. Conditional Routing: An n8n IF node checks the composite score. Score 80+ routes to Sales Slack channel with profile card. Score 50-79 routes to HubSpot nurture sequence. Score below 50 routes to a low-priority re-engagement list.

7. Logging: All scoring results are written to a Google Sheet with columns for timestamp, lead ID, each axis score, composite, confidence, routing destination, and AI reasoning.

8. High-Value Alert: For leads scoring 90+ or with executive job titles, a separate Slack alert is sent to the sales manager with a shortened response-time SLA warning.

TOOL INTEGRATION

n8n 1.80+: Self-host via Docker or use n8n cloud. Install the HubSpot trigger node from the n8n node library. The webhook URL format is https://your-n8n-instance/webhook/lead-scoring. Gotcha: n8n's webhook response timeout is 30 seconds by default — the GPT-4o call can take 5-15 seconds, so you may need to respond immediately with a 202 Accepted and continue processing in a separate workflow branch.

OpenAI GPT-4o: API key from platform.openai.com. Set OPENAI_API_KEY in n8n credentials. Use the gpt-4o-2024-11-20 model with response_format set to json_object. Gotcha: GPT-4o token limits mean the lead profile plus prompt must stay under 8K tokens. Very long company descriptions or history fields may need truncation before submission.

HubSpot CRM: Create a private app in HubSpot under Settings > Integrations > Private Apps. Scope: contacts, deals, and workflows read/write. Gotcha: HubSpot's workflow webhook sends a batch of contacts every 5 minutes, not instantly — configure the webhook trigger frequency to match your desired response speed.

Slack: Create a Slack app with chat:write and incoming-webhook scopes. Install to your workspace. Gotcha: Slack's rate limit is 1 message per second per channel. For high-volume lead inflows, batch alerts or use Slack's Webhook with message deduplication.

Google Sheets: Use n8n's Google Sheets node with OAuth2 authentication. Gotcha: Google Sheets API has a write limit of 60 requests per minute per project. For high-volume logging, switch to a database like Supabase instead.

ROI METRICS

1. Lead response time: from 12-24 hours (manual) to under 2 minutes (automated). (Source: InsideSales.com, 2024). 2. Hours spent on lead triage per week: 12-18 hours manual → 1-2 hours for review and exception handling, freeing 10-16 hours for actual selling activity. 3. Conversion rate on auto-routed high-scoring leads: expected 25-40% improvement due to speed-to-lead reduction with faster initial contact. 4. Cost per lead processed: approximately $0.12-0.25 in API costs vs $5-12 in equivalent SDR hourly labor, representing a 20-40x cost reduction. 5. First measurable KPI: week 1, track number of leads logged and scored vs manually processed in the same period prior.

CAVEATS

1. GPT-4o may misinterpret niche industry signals or company types it has limited training data on — the scoring prompt must be tuned over the first 50-100 leads. 2. API costs scale with volume: scoring 500 leads per month at $0.15 per call equals $75, plus enrichment API costs. Budget for unexpected spikes from bot submissions. 3. The system scores leads based on available data — incomplete profiles produce unreliable scores. Require minimum fields (email, company name, source) before triggering the AI node. 4. False positives: high-scoring leads may still lack budget authority or timing. This workflow handles initial triage only, not qualification conversation.