What the Workflow Actually Does
A lead fills out your form. The workflow catches the submission, reads the company behind the email address, researches it on the open web, and asks Claude for a fit score with a written reason. HubSpot gets the score, the tier, and the reasoning on the contact record, and the lead is routed before anyone opens their inbox.
This diagram mirrors the real n8n canvas below, stage by stage.
The actual workflow, as published on n8n.io. Every step in this playbook exists on this canvas.
What is AI lead qualification?
AI lead qualification is the use of a language model to research, score, and route inbound leads the moment they arrive, instead of leaving triage to whoever checks the inbox next. A typical system reads the form submission, gathers public information about the company, and returns a fit score from 1 to 10 with a written reason, usually within five minutes of the submit.
The point is not replacing human judgment. It is making the first pass instant and consistent. Every lead gets researched the same way, scored against the same ideal customer profile, and lands in the CRM with the reasoning attached, so a person can override any call the model makes. The workflow documented on this page does exactly that with Claude, HubSpot, Slack, and Gmail.
How It Works
Your form posts five fields to the workflow's lead-intake webhook: name, email, company, job title, and message. Before anything else runs, the workflow checks that HubSpot has the three custom properties it writes to, ai_lead_score, ai_lead_tier, and ai_score_reason, and creates them if they are missing.
It extracts the company domain from the email address, fetches the company homepage, and runs a web search for third party coverage of the company, deliberately excluding the company's own domain. The score gets built on outside evidence, not just the company's marketing copy.
Claude receives the lead profile and your ideal customer description, then returns a fit score from 1 to 10, a written reason, and a drafted reply email for the lead. Thresholds you control map the score to hot, warm, or cold.
The contact is upserted with score, tier, and reason, and the lifecycle stage is set by your thresholds. Hot leads get a HubSpot deal, a Slack alert to sales, and an immediate email. Warm and cold leads get tier matched replies. Nothing waits in a queue.
The exact payload the webhook expects. Point your form tool at the webhook URL and map these five fields.
What You Need to Run It
Get the Exact Workflow, Free
The system documented above, as an importable n8n template. Runs on any n8n instance, including the free self hosted version. Nothing is locked, nothing is gated.
Hosted on n8n.io · import in one click
Run It Yourself, or Have It Built
If your stack matches the template, HubSpot, a form, Slack, and Gmail, you can import this and be live the same day. Paste four credentials, describe your ideal customer in a paragraph, set two thresholds, and send a test lead through.
n8n's free self hosted tier runs it fine. No paid n8n plan, no proprietary connectors, and no involvement from us.
The failure points are rarely on the canvas. They live at the edges: forms that send different field names, a CRM full of legacy properties, scoring criteria nobody ever wrote down, and no one watching when an API changes underneath you. That work lands on whoever imported the template.
Builds run $5,000 to $15,000 fixed price, live in 2 to 4 weeks, scoped after the free audit: a 10 to 15 minute video showing where your leads leak. Five per month.
Before You Ask
Accuracy comes from the inputs, not the model. Scoring against a written ideal customer profile with outside research beats inbox triage for consistency, and every score ships with a written reason on the contact record, so a person can override any call in seconds. Treat the first two weeks as calibration: read the reasons, adjust the ICP text and thresholds, and the scores settle.
HubSpot's built in scoring adds and subtracts points from rules you define, mostly activity signals like opens, page visits, and form fills. This workflow judges fit. It researches the company, reads the message, and explains its score in plain language. The two work well together. Rules tell you who is active, this tells you who is worth your time.
n8n's self hosted version is free, and Claude usage for research plus scoring runs cents per lead, not dollars. The real investment is roughly two hours of setup plus the thinking behind your ICP and thresholds. There is no subscription and nothing locked. The workflow is yours after import.