How to Build an AI Lead Enrichment Workflow?

A new contact lands in your CRM with a name and a company, and nothing else. No email, no title, no idea whether it fits.

Your reps either fill the gaps by hand or skip the lead. Both cost you pipeline.

This guide explains what AI lead enrichment is, what it should add to your CRM, and shows a workflow we tested on 14 HubSpot contacts.

You can run the workflow from an AI Agent in Claude Code, Codex or Cursor, or from n8n, every morning.

We ran both on real HubSpot contacts. Copy the workflow below, or follow the same method with your own tools.

We will walk you through the process step by step.

Everything is in one public GitHub repo: the skill, a small helper script, the fit rules and the n8n workflow. Paste the prompt into your AI Agent, or pick the n8n tab to run it every morning.

Agent kit Free · Public

GitHub n8n

Let your AI agent set it up

Paste this prompt into Claude Code, Codex or Cursor. Your agent clones the public repo, follows its instructions and asks before every change.

Prompt · GitHub Copy

Clone https://github.com/evaboot-labs/ai-lead-enrichment, read AGENTS.md
and follow the skill. Enrich my new HubSpot contacts, and show me
every change before you write anything.

In the repo

• The ai-lead-enrichment skill: pull, waterfall, judge, review, write back
• enrich.py: every HubSpot and Evaboot call, the checks and the review
• FIT.md: who fits, and the field rules
• The n8n workflow, with its setup guide

Open the repo on GitHub

Set it up in n8n

Paste this prompt into your AI agent and it walks you through the import, one step at a time. Or open the workflow JSON on GitHub and import it yourself.

Prompt · n8n Copy

Set up the AI lead enrichment workflow in my n8n.
Clone https://github.com/evaboot-labs/ai-lead-enrichment and follow docs/n8n.md.
Walk me through the three HubSpot properties, importing
n8n/ai-lead-enrichment.json, the HubSpot, Evaboot and Anthropic
credentials, and the settings, one step at a time.

In the repo

• The workflow JSON: every morning at 7, from HubSpot back to HubSpot
• docs/n8n.md: the properties, the three credentials and the settings
• The same write-back rules as the script

Open the n8n JSON on GitHub Open the repo on GitHub

We used HubSpot, Evaboot and Claude in our example, but with a few tweaks, the same method works with another CRM or model.

What Is AI Lead Enrichment?

AI lead enrichment is completing a lead record with the data it is missing, then letting AI add what that data means.

The data comes from an enrichment tool: an email finder, a profile export or a company database. AI cannot find an email or a job title on its own; it can only guess one.

AI adds the judgment: which department the title belongs to, how senior the person is, and whether the lead fits who you sell to.

What an AI Enrichment Workflow Adds to Your CRM

Each field has a source and a rule. Here is what our workflow writes in HubSpot:

Nothing your team wrote is ever overwritten. The workflow fills empty fields and its own three fields, nothing else.

How to Build an AI Lead Enrichment Workflow

Our workflow turns an incomplete contact into a checked, complete one, in six steps.

The agent finds and judges. You read the changes, and only then does it write.

1. Add Three Fields to HubSpot

The workflow keeps its own work in three contact properties: AI enrichment status, AI enrichment summary and AI enriched at.

The status takes a contact out of the next run, so nobody is enriched or charged twice:

2. Pull Only the Contacts That Need It

enrich.py pull asks HubSpot for contacts created in the last 7 days with no email or no job title that were never enriched.

We tested it on 8 new contacts with a first name, a last name and a company: three with a Sales Navigator id, three with a company website only, and two with a public LinkedIn URL.

3. Run the Evaboot Waterfall

A waterfall tries the richest source first. A contact with a Sales Navigator id gets a profile export: title, months in role, company size, industry, growth and a verified email.

Every other contact goes to the email finder, on name and company. It returns an email and its status, and nothing else:

A public LinkedIn URL is not enough for the profile export: Evaboot needs the Sales Navigator id, which its own exports store. Both public URLs went to the email finder.

4. Let the Agent Judge Each Contact

The agent writes a job function, a seniority, a fit and one summary line per contact, from FIT.md and the facts Evaboot found.

Without a title, it says so instead of guessing. Five of our eight contacts got an email but no title, so their fit stayed at maybe.

5. Review Every Change

enrich.py check rejects a bad judgment: a seniority HubSpot does not know, a summary over 30 words, or a guess at someone's gender. Then enrich.py review shows every field that will change:

One email was held. It was valid, but not on the company's domain, so the workflow wrote it in the summary for a person to check instead of in the email field.

The agent also flagged a contact that was not a person: a company page named after the marketing team, marked "no".

6. Write Only the Empty Fields

On your yes, enrich.py push --apply updates HubSpot in one call. Our test wrote 24 fields on 8 contacts:

Each contact now shows what was found and why it fits, in its own record:

How to Run Your Lead Enrichment in n8n

n8n runs the same waterfall every morning. A HubSpot search finds the contacts, two Evaboot nodes complete them, Claude judges them, and each contact is updated with the same rules.

Here is the whole workflow, after a run on 6 new contacts:

The setup takes four steps:

1. Import the Workflow and Add Three Credentials

Import n8n/ai-lead-enrichment.json from the repo, then add a HubSpot app token, an Evaboot API key and an Anthropic key.

The Evaboot key is a Header Auth credential: the name is Authorization, the value is Bearer followed by your key.

2. Fill In Who Fits

The Settings node holds how many days back to look, how many contacts to take per run, and who your buyers are:

Keep the buyers short: the roles and the company size you sell to, and who is not a fit.

3. Check the Two Evaboot Calls

The profile export sends the contact's Sales Navigator URL to Evaboot and gets the full profile back:

The email finder sends the first name, the last name, the company and the website, and gets the email and its status:

4. Run It, Then Publish

Click Execute workflow and open a contact in HubSpot. In our run, 3 contacts were enriched by the profile export and 3 got an email from the finder.

Publish the workflow, and it runs every morning at 7 on the contacts created since the last run.

Example: 14 Contacts, Two Ways

We ran the agent on 8 contacts and n8n on 6 more, all created the same day in our test HubSpot:

Three things stood out. Every contact with a Sales Navigator id came back with a title; none of the others did.

The email finder is not fully stable. On one n8n run, it found no email for one contact; on the next run, it found a safe one.

Claude judged the same contact differently on two runs: maybe once, no the next time. Keep the summary as a hint for your reps, not a gate.

Common Lead Enrichment Mistakes

Most enrichment problems come from writing too much, not finding too little.

Here are the four mistakes we see most:

1. Overwriting What Your Team Wrote

A rep corrected a title last week. A blind enrichment run puts the old one back.

Fill empty fields only, and keep the AI's work in its own fields.

2. Trusting Every Safe Email

"Safe" means the mailbox exists, not that it belongs to your lead's company.

Compare the email's domain with the company website, and hold the ones that differ.

3. Expecting a Title From a Name

An email finder returns an email, not a job. Five of our eight contacts had no title after enrichment.

Store the Sales Navigator id when leads enter your CRM: Evaboot's exports keep it, and the profile export needs it.

4. Enriching the Whole CRM Every Day

Every lookup costs a credit. Enriching every contact every morning burns credits on people you will never call.

Enrich new contacts once, and let a status field take them out of the next run.

Frequently asked questions