The Lead Qualification Autopilot
AI reads your inbox, enriches each sender from their website, scores the lead 0 to 100, and drafts a personalized reply. You spend five minutes a day reviewing and hitting send. Built on n8n, Gmail, Firecrawl, Claude, and Airtable. Nothing sends without your approval.
Section One
What This System Does
Every minute, the workflow checks the connected Gmail inbox for new unread mail. Real inquiries get identified, enriched with a scrape of the sender's company website, scored 0–100 by a hybrid of hard rules and AI judgment, and answered with a personalized 120–180 word draft that references specifics from their email and their business. The draft lands in Gmail Drafts, the lead lands in your Airtable CRM, and everything that came through the door gets logged, including the newsletters and vendor pitches that were never leads at all.
Fig. 1 — system flow
Detect
Gmail
New unread email checked every minute
Enrich
Firecrawl
Scrapes the sender's company site
Score
Rules + AI
0–100, hybrid of hard rules and judgment
Draft
Claude → Gmail Draft
Personalized reply — never auto-sent
The rule that makes this work
The draft is the design, not a training wheel. Speed-to-lead is the whole game on inbound, and this makes the default outcome a ready-to-send reply instead of an email sitting unanswered for two days. The Gmail draft keeps your judgment in the loop at a cost of five minutes a day, and the AI explains every score in one sentence so you can trust the queue instead of re-reading every inquiry.
Section Two
Your Five Minutes a Day
Open the queue
Airtable → Lead Qualification CRM → Leads, sorted by Score. Anything with Status Draft Ready is waiting on you.
Read the two sentences that matter
The Score Reason tells you why the AI rated the lead the way it did. The Enrichment Summary tells you what it learned from their website. Ten seconds each.
Send the draft
Gmail → Drafts, find the reply addressed to that lead. Edit if needed, add your name after "Best,", and hit Send. Then flip the Airtable Status to Contacted.
Close the loop
As conversations progress, update Status to Qualified or Disqualified. That keeps the CRM honest and gives you real conversion data on what the scorer gets right.
Hot leads first. Aim to answer Hot (70+) leads within a couple of hours — speed is the single biggest lever on inbound conversion, and it's the one thing the automation cannot do for you. Warm (40–69) can wait for the daily pass. Cold (<40) is a judgment call: skim, reply if there's a spark, disqualify if not.
Section Three
The Airtable CRM
Two tables. Leads is your working queue, one row per lead, upserted on email address so a repeat inquiry updates the record instead of duplicating it. Activity Log is your paper trail, one row per processed email, leads or not.
| Field | What It Tells You |
|---|---|
| Score / Tier / Score Reason | The 0–100 score, Hot/Warm/Cold tier, and the AI's one-line explanation for the number. |
| Status | New → Draft Ready → Contacted → Qualified / Disqualified. Draft Ready is your action signal. |
| Enrichment Summary | What the AI learned about the company from its website. |
| Original Email | The inbound message, preserved verbatim. Never rely on a summary when the source is one click away. |
| Draft Subject / Draft Body | A copy of the outreach draft. The live version sits in Gmail Drafts. |
Activity Log: one row per email with the outcome (Processed, Skipped - Not a Lead, or Error), the score, details, and the n8n execution ID. That ID is your tracer: any row in the log maps straight to a run in n8n → Executions, so you can see exactly what happened at every node.
Why non-leads get logged
Newsletters, spam, and vendors selling to you never generate drafts, but they still get a row. Nothing disappears silently. When a real lead seems to have vanished, the log tells you in seconds whether it was filtered, errored, or never arrived, instead of you wondering whether the system ate it.
Section Four
How Scoring Works
▸ Rules first — 0 to 50 points
- Business email domain — +20
- Buying-intent keywords (budget, timeline, pricing, hire…) — +5 each
- Substantive message length — +5
▸ AI judgment — 0 to 50 points
- Clear project need — +15
- Decision-maker signals — +10
- Stated budget or timeline — +15
- Fit, judged from their website — +10
Why you can trust it. Every score ships with a one-sentence reason, written by the model that assigned it. When a lead scores oddly, read the reason before you assume the system is broken; most of the time it saw something you skimmed past. When the rubric is genuinely wrong for your business, it's plain English in one node, and you tune it there.
Tiers: Hot 70+, Warm 40–69, Cold <40. The AI also filters non-leads out entirely; those are logged as Skipped and no draft is created.
Section Five
Troubleshooting
Every failure mode has a first place to look. Start here before opening a single node.
| Symptom | Check |
|---|---|
| No new leads appearing | Is the workflow Active in n8n? Is the email actually unread? Only unread mail triggers a run, and the Promotions and Social tabs are excluded by design. |
| A lead scored oddly | Read the Score Reason in Airtable; the AI explains itself. If the rubric itself is off, tune it in the "Score Lead & Write Draft" node's system message. |
| No website summary | The sender used a personal email (gmail, yahoo, etc.), or the site blocked scraping. Either way the workflow continues; enrichment failure is non-fatal. |
| Something failed mid-run | n8n → Executions shows every run with the failing step highlighted. Cross-reference the Execution ID column in the Activity Log. |
| Airtable steps fail with "Forbidden" | The Airtable personal access token doesn't include this base. At airtable.com/create/tokens, add the "Lead Qualification CRM" base with data.records:read and data.records:write scopes, or grant the whole workspace. |
The debugging shortcut
The Execution ID in the Activity Log is the bridge between "a lead looks wrong in Airtable" and "here is the exact node that produced it." One lookup replaces guessing. If you run the n8n connector in Claude, you can also just paste the ID and ask Claude to read the execution and fix whatever broke.
Section Six
Tuning & Handoff
Everything an operator would ever want to change lives in five handles, all in the n8n workflow editor. Handing this system to a client means pointing at this page.
| Handle | Where |
|---|---|
| Which inbox | "New Inbound Email" node → Gmail credential. Swap to the client's account here. |
| What counts as a lead, rubric, draft tone | "Score Lead & Write Draft" node → System Message. Plain English; edit freely. |
| Keyword signals | "Normalize Lead" code node → KEYWORDS list. |
| CRM destination | The three Airtable nodes (base and table IDs). |
| Polling frequency | Trigger node. Currently every minute. |
Safety rails built in
Email never auto-sends; output is always a Gmail draft. No duplicate CRM records; upsert matches on email address. Nothing dropped silently; skipped and errored emails are logged. Enrichment failure is non-fatal; if a website can't be scraped, scoring proceeds without it.
Running costs: n8n cloud on your existing plan. Airtable free tier. Claude API at roughly $0.01–0.03 per lead on Sonnet. Firecrawl at one credit per business-domain lead, with the free tier covering about 500 a month. Gmail is free. The marginal cost of never losing a lead again rounds to a coffee a month.
“The draft is the design, not a training wheel.”
— The Lead Qualification Autopilot, p. 2
§ Fin.
Rather have it built for you?
This guide is one system from the Soulwire Studio workflow library. We build AI automations for businesses that run on process: intake, scheduling, proposals, compliance-heavy operations. If you'd rather spend your time closing deals than debugging lead gen, book an audit and we'll find the automation hiding in your business.
