How ICP Refinement Agent Automates Lead Qualification

Your sales team has 47 deals in the proposal stage. 12 have not had contact in 5+ days. Three have gone completely dark. Which ones are at risk — and which ones just have a slow procurement process? A rep answering this question manually checks Hubspot, Apollo, Notion, cross-references email history, and makes a judgment call on each deal. At 15 minutes per deal, that is 30–60 minutes per cycle of triage before any follow-up happens.

The cost is not just time — it is revenue leakage. Deals slip because signals were missed. Pipeline reviews rely on data that was accurate two days ago. Scoring criteria drift between team members, and the CRM becomes a lagging indicator rather than an operational tool. ICP Refinement Agent automates the lead qualification and revenue operations workflow from data extraction through analysis to structured output, with zero manual CRM entry.

Four Agents. Six Dimensions. A Living ICP That Evolves With Your Market.

The ICP Refinement Agent pipeline runs 4 agents in sequence. Fetcher pulls data from Hubspot and Apollo and Notion, and Formatter delivers the output. Here is what happens at each stage and why it matters.

• Fetcher (Schedule + Code): Schedule Trigger fires monthly (1st 09:00 UTC) or manual Webhook for on-demand runs.
• Enricher (Code-only): Augments company profiles via Apollo.io API with firmographics, technographics, and growth signals.
• Analyst (Tier 2 Classification): the analysis model receives ONE aggregate call with all enriched deal data.
• Formatter (Tier 2 Classification): the analysis model generates a Notion ICP report page with dimension sections, confidence badges, and shift indicators.

When the pipeline completes, you get structured output that is ready to act on. The blueprint bundle includes everything needed to deploy, configure, and customize the workflow:

• ITP-tested 29-node n8n workflow — import and deploy
• Monthly Schedule Trigger (1st 09:00 UTC) or manual Webhook for on-demand runs
• HubSpot API pagination for closed-won deals with company and contact data
• 90-day lookback + 90-day prior comparison for trend detection
• 6-dimension ICP taxonomy: company_size, industry_vertical, tech_stack, growth_stage, buyer_persona, deal_dynamics
• Pattern confidence ratings: HIGH (>70% deal match), MEDIUM (40–70%), LOW (<40%)
• Period-over-period comparison showing how your ICP is shifting
• Apollo.io firmographic enrichment (conditional — toggle on/off)
• Notion ICP report with dimension sections and confidence badges
• HubSpot task with scoring adjustment recommendations
• AGGREGATE architecture: single Analyst + Formatter calls — $0.28/run regardless of deal count
• Dual the analysis model: no the primary reasoning modelrequired
• ITP 8 variations, 14/14 milestones, $0.28/run measured

Scoring thresholds, output destinations, and CRM field mappings are configurable in the system prompts — no workflow JSON edits required. This means ICP Refinement Agent adapts to your specific process, terminology, and integration requirements without forking the entire workflow.

Understanding how the pipeline works helps you customize it for your environment and troubleshoot issues when they arise. Here is a step-by-step walkthrough of the ICP Refinement Agent execution flow.

Step 1: Fetcher

Tier: Schedule + Code

The pipeline starts here. Schedule Trigger fires monthly (1st 09:00 UTC) or manual Webhook for on-demand runs. Fetcher paginates HubSpot API for closed-won deals over a configurable 90-day lookback window plus a prior 90-day comparison period. Collects associated company and contact data for each deal.

This stage ensures all downstream agents receive clean, validated input. If this step returns incomplete data, every downstream agent works with a degraded picture.

Step 2: Enricher

Tier: Code-only

Augments company profiles via Apollo.io API with firmographics, technographics, and growth signals. Conditional — toggle INCLUDE_APOLLO_ENRICHMENT to skip when Apollo credits are limited. Rate-limited with SplitInBatches and retry backoff.

Why this step matters: The result is a prioritized action queue, not just a data dump.

Step 3: Analyst

Tier: Tier 2 Classification

the analysis model receives ONE aggregate call with all enriched deal data. Produces 6-dimension ICP analysis: company_size, industry_vertical, tech_stack, growth_stage, buyer_persona, and deal_dynamics. Each dimension includes pattern confidence ratings (HIGH/MEDIUM/LOW) and period-over-period comparison showing how your ICP is shifting.

Every field in the output is structured for the next agent to consume without parsing.

Step 4: Formatter

Tier: Tier 2 Classification

This is the final deliverable — what lands in your inbox or dashboard. the analysis model generates a Notion ICP report page with dimension sections, confidence badges, and shift indicators. Also creates a HubSpot task with scoring adjustment recommendations so your lead qualification criteria stay current.

The entire pipeline executes without manual intervention. From trigger to output, every decision point follows a documented path. Every execution produces a traceable audit trail.

All nodes have been validated during Independent Test Protocol (ITP) testing on n8n v2.7.5. The error handling matrix in the bundle documents the recovery path for each failure mode.

Ghost contacts, rebranded companies, missing fields — that is what ITP fixtures contain. A 524-day inactive contact is now a standard test case. You do not find out if error handling works by testing happy paths. You find out by throwing data that should not exist and verifying the pipeline does not crash.

— ForgeWorkflows Engineering

Every metric is ITP-measured. The ICP Refinement Agent turns HubSpot closed-won deal patterns into a living ICP — analyzing 6 dimensions with pattern confidence ratings, period-over-period comparison, and scoring recommendations at $0.28/run.

The primary operating cost for ICP Refinement Agent is the per-execution LLM inference cost. Based on Independent Test Protocol (ITP) testing, the measured cost is: Cost per Run: $0.28/run (ITP-measured average). This figure includes all API calls across all agents in the pipeline — not just the primary reasoning step, but every classification, scoring, and output generation call.

To put this in context, consider the manual alternative. A skilled team member performing the same work manually costs $50–75/hour for a sales ops analyst at a fully loaded rate (salary, benefits, tools, overhead). If the manual version of this workflow takes 30–60 minutes per cycle, the per-execution cost in human labor is significant. The blueprint executes the same pipeline for a fraction of that cost, with consistent quality and zero fatigue degradation.

Infrastructure costs are separate from per-execution LLM costs. You will need an n8n instance (self-hosted or cloud) and active accounts for the integrated services. The estimated monthly infrastructure cost is $0.28/month (monthly runs) + HubSpot/Apollo.io/Notion included tiers, depending on your usage volume and plan tiers.

Quality assurance: Blueprint Quality Standard (BQS) audit result is 12/12 PASS. ITP result is 8 variations, 14/14 milestones PASS. These are not marketing claims — they are test results from structured inspection protocols that you can review in the product documentation.

All cost and performance figures are ITP-measured — tested against real data fixtures on n8n v2.7.5 in March 2026. See the product page for full test methodology.

7 files — workflow JSON, system prompts, TDD, and complete documentation.

When you purchase ICP Refinement Agent, you receive a complete deployment bundle. This is not a SaaS subscription or a hosted service — it is a set of files that you own and run on your own infrastructure. Here is what is included:

• CHANGELOG.md — Version history
• README.md — Setup and configuration guide
• docs/TDD.md — Technical Design Document
• icp_refinement_agent_v1.0.0.json — n8n workflow (main pipeline)
• system_prompts/analyst_system_prompt.md — Analyst system prompt
• system_prompts/formatter_system_prompt.md — Formatter system prompt

Start with the README.md. It walks through the deployment process step by step, from importing the workflow JSON into n8n to configuring credentials and running your first test execution. The dependency matrix lists every required service, API key, and estimated cost so you know exactly what you need before you start.

Every file in the bundle is designed to be read, understood, and modified. There is no obfuscated code, no compiled binaries, and no phone-home telemetry. You get the source, you own the source, and you control the execution environment.

ICP Refinement Agent is built for Sales, Revops teams that need to automate a specific workflow without building from scratch. If your team matches the following profile, this blueprint is designed for you:

• You operate in a sales or revops function and handle the workflow this blueprint automates on a recurring basis
• You have (or are willing to set up) an n8n instance — self-hosted or cloud
• You have active accounts for the required integrations: HubSpot account (OAuth2 with deals, companies, contacts scopes), Apollo.io API key, Notion workspace (integration token with Bearer prefix), Anthropic API key
• You have API credentials available: Anthropic API, HubSpot (OAuth2), Apollo.io (httpHeaderAuth), Notion (httpHeaderAuth, Bearer prefix)
• You are comfortable importing a workflow JSON and configuring API keys (the README guides you, but basic technical comfort is expected)

This is NOT for you if:

• Does not score individual leads — that is what Inbound Lead Qualifier does
• Does not prospect against ICP criteria — that is what Outbound Prospecting Agent does
• Does not forecast pipeline revenue — that is what RevOps Forecast Intelligence Agent does
• Does not modify HubSpot deal data or contact records — read-only analysis with task output
• Does not scrape external websites — all data from HubSpot API and Apollo.io API
• Does not analyze lost deals — focuses exclusively on closed-won patterns

Review the dependency matrix and prerequisites before purchasing. If you are unsure whether your environment meets the requirements, contact support@forgeworkflows.com before buying.

Every pipeline has boundaries. These are intentional design decisions, not oversights — understanding them helps you deploy with the right expectations and plan for edge cases in your environment.

Does not score individual leads — that is what Inbound Lead Qualifier does

This is intentional. We default to human-in-the-loop for actions that carry reputational or financial risk. Once your team has validated output accuracy over 20+ cycles, you can adjust the pipeline to auto-execute — the workflow JSON supports it, but the default is conservative.

Does not prospect against ICP criteria — that is what Outbound Prospecting Agent does

We scoped this boundary after ITP testing revealed inconsistent results when the pipeline attempted this. The agents handle what they handle well — extending beyond this scope requires custom prompt engineering specific to your data shape.

Does not forecast pipeline revenue — that is what RevOps Forecast Intelligence Agent does

This keeps the pipeline focused on a single workflow. Adding this capability would introduce branching logic that varies by organization, and the tradeoff between complexity and reliability was not worth it for a reusable blueprint. Fork the workflow JSON if your use case demands it.

Deployment follows a structured sequence. The ICP Refinement Agent bundle is designed for the following tools: n8n, Anthropic API, HubSpot, Apollo.io, Notion. Here is the recommended deployment path:

1. Step 1: Import workflow and configure credentials. Import icp_refinement_agent_v1_0_0.json into n8n. Configure HubSpot OAuth2 credential (deals, companies, contacts scopes), Apollo.io httpHeaderAuth credential (API key), Notion httpHeaderAuth credential (Bearer token), and Anthropic API key following the README.
2. Step 2: Configure lookback window and enrichment settings. The Schedule Trigger defaults to monthly (1st 09:00 UTC). Configure LOOKBACK_DAYS (default 90), MIN_DEALS threshold (default 10), CONFIDENCE_THRESHOLD, INCLUDE_APOLLO_ENRICHMENT (true/false), and NOTION_DATABASE_ID for your ICP report destination.
3. Step 3: Activate and verify. Enable the workflow in n8n. Trigger a manual run via the Webhook URL. Verify the Notion ICP report page appears with 6 dimension sections, confidence badges, and period comparison. Check HubSpot for the scoring task with adjustment recommendations.

Before running the pipeline on live data, execute a manual test run with sample input. This validates that all credentials are configured correctly, all API endpoints are reachable, and the output format matches your expectations. The README includes test data examples for this purpose.

Once the test run passes, you can configure the trigger for production use (scheduled, webhook, or event-driven — depending on the blueprint design). Monitor the first few production runs to confirm the pipeline handles real-world data as expected, then let it run.

For technical background on how ForgeWorkflows blueprints are built and tested, see the Blueprint Quality Standard (BQS) methodology and the Inspection and Test Plan (ITP) framework. These documents describe the quality gates every blueprint passes before listing.

Ready to deploy? View the ICP Refinement Agent product page for full specifications, pricing, and purchase.