Turn Emails Into Customers.

Everything starts with a trigger. A webhook sits ready like a catcher with the mitt up — the moment someone submits a lead magnet, demo request, or trial form, it captures the email address and fires the workflow. No polling, no batch jobs, no waiting.

The username half of the email is noise for now. The node strips it down to the root domain — the part that actually points to a company — and confirms the input is a real email address before anything downstream runs.

A regex blacklist omits the throwaways: gmail, outlook, temporary inboxes, and the long tail of personal addresses that will never map to a company. A null/empty check sits right behind it as error handling, so the workflow stays robust on bad input.

Instead of a POST or GET that pulls the entire page of HTML and JavaScript, this uses a HEAD request — it asks only for the status code. A 200 to 399 means the site is live. It is a cheap pre-check that keeps you from wasting Firecrawl or any scraper credits on dead domains.

Now the real scrape. Firecrawl reads the site and hunts the footer for the company LinkedIn profile URL — the same place you usually find X, YouTube, and Instagram, a company's whole digital presence. It is chosen here for cost, infrastructure, and stability. The extracted URL gets a not-empty / not-null guard before moving on.

First a check confirms the profile is a company, not a personal one. Then an OpenAI agent is grounded with the R.T.O. framework — Role ("you are a LinkedIn data analyst"), Task (extract followers, headquarters, location), and Output (return structured JSON, numbers as numbers). Grounding the agent to one narrow job is what stops it going loosey-goosey.

AI output comes back a little messy sometimes. A simple code node sanitizes and structures it, then a normalization step standardizes values — turning every variant of a country into one canonical name so the scoring logic has clean inputs.

The fun part. Three signals get scored independently against your criteria — country (United States lands a 10, others scale down to 2, 1, or 0), staff count, and industry — then combined into one holistic, algorithmic value. This is the code you should refine with your sales team or founder for your exact ICP.

Two outputs land together: a qualitative read ("moderate fit") and a quantitative score out of 100. One email has now become a ranked, defensible lead grade your whole team can act on.

The lead is pushed to your CRM (Salesforce, HubSpot, whatever you run) and a Slack message notifies the team — "you've got a new lead, go check it out." The high-fit ones can buzz your Slack and your phone so a rep gives them white-glove attention immediately.

Scores segment the flow. High-ticket leads get funneled one way for human-led follow-up; the rest route into automated campaigns. Same workflow, different language and urgency depending on how good the fit is.

A dedupe check makes sure the lead is not already in the sequence, the first name is parsed from the email and properly cased, then the workflow waits ~5 minutes — because messaging someone the literal second they sign up is creepy — before enrolling them in the right Instantly campaign.