How to Use n8n with ._Template 164 Ai Lead Scoring Email Marketing

If your sales team is chasing every inbound lead with the same urgency, you're burning your best hours on people who will never buy — and letting the ready-to-close prospects go cold in the queue. Lead scoring fixes this, but most scoring lives trapped inside an expensive CRM add-on or a spreadsheet nobody updates. Template 164 (AI Lead Scoring for Email Marketing) turns n8n into the engine that scores every lead automatically, using an LLM to read intent, and routes hot leads straight into your email sequences. This walkthrough shows you exactly how to wire it up, node by node.

The Problem: Manual Lead Scoring Doesn't Scale

Traditional lead scoring is rule-based and brittle. You assign +10 points for a job title containing "VP," +5 for a company size over 200, +15 for visiting the pricing page. It works for a week, then reality intrudes: a "Head of Growth" scores zero because your rule only matched "VP," a founder at a 4-person startup with real budget gets buried, and a tire-kicker who obsessively refreshes your pricing page gets flagged as your hottest prospect.

The deeper problem is that intent lives in unstructured signals — the wording of a contact-form message, the specific pages someone read, the reply they sent to your first email. Point-based rules can't read nuance. And the moment marketing wants to tweak the model, they file a ticket and wait two weeks for ops. Meanwhile, sales works leads in the order they arrived, which is the one ordering guaranteed to waste time. What you actually need is a scoring system that reads context like a human, runs the instant a lead arrives, and drops its verdict directly into the tools your team already uses.

The Solution: AI Scoring as an n8n Workflow

Template 164 treats lead scoring as an event-driven pipeline. A new lead hits a webhook, the workflow enriches and normalizes the data, an AI node evaluates intent and fit against a rubric you define in plain English, and the resulting score routes the lead down one of three paths: hot leads trigger an immediate personalized email and a sales alert, warm leads enter a nurture sequence, and cold leads get logged for later. Because it's n8n, every step is visible, editable, and swappable — you own the logic instead of renting it.

The advantage of using an LLM over a static rule set is that the model reasons about the whole lead at once. It weighs a strong job title against a weak message, notices that "just researching for a class project" signals no intent regardless of company size, and outputs a calibrated number with a justification you can audit. You get the flexibility of a human reviewer at the throughput of an automation.

Step-by-Step Setup in n8n

Import Template 164 into your n8n instance, then configure the following nodes in order. Each maps to a stage in the funnel above.

1. Webhook node (trigger). Set the HTTP method to POST and copy the production URL. Point your lead form — Typeform, a native HTML form, or your CRM's outbound hook — at this URL. Configure the node to respond with "Using Respond to Webhook Node" so you can return a clean 200 after processing. Expect a JSON body with fields like email, name, company, job_title, and message.

2. Set node (normalize). Add a Edit Fields (Set) node to coerce incoming data into a stable shape. Trim whitespace, lowercase the email, and provide fallbacks with expressions like {{ $json.company || "unknown" }}. This guards the AI node against missing keys that would otherwise poison the prompt.

3. HTTP Request node (enrich — optional). If you enrich leads, insert an HTTP Request node here to hit a data provider (Clearbit, Apollo, or your own DB) using the email as the key. Set "Continue On Fail" to true so a missing enrichment record never halts the run.

4. AI scoring node. This is the core. Use a Basic LLM Chain (or the OpenAI/Anthropic message node) wired to your model credential. Write a system prompt that defines the rubric explicitly: "You are a B2B lead scorer. Score 0–100 based on buying intent and ICP fit. Our ICP is ops and revenue teams at 20–500 person companies. Return strict JSON: {score, tier, reason} where tier is hot (80+), warm (50–79), or cold (below 50)." Feed the normalized fields into the user message with expressions. Crucially, instruct the model to return only JSON — no prose — so the next node can parse it deterministically.

5. JSON parsing. Follow the AI node with a Code node or a Structured Output Parser to convert the model's text into real JSON. In a Code node: return JSON.parse($json.text.match(/\{[\s\S]*\}/)[0]) defends against stray characters around the object.

6. Switch node (route). Add a Switch node keyed on {{ $json.tier }} with three outputs: hot, warm, cold. This replaces fragile nested IF nodes and keeps the routing readable.

7. Email + alert nodes. On the hot branch, connect a Send Email node (SMTP) or your ESP node — Gmail, SendGrid, or an HTTP Request to your email platform's API — to fire a personalized first-touch message, using {{ $json.reason }} to tailor the opening line. Chain a Slack or WhatsApp node to alert sales in real time. On the warm branch, use an ESP API call to add the contact to a nurture list or tag. On the cold branch, append a row to Google Sheets or your database for periodic review.

8. Respond to Webhook. Close every branch into a Respond to Webhook node returning { "status": "scored", "tier": "..." } so the calling form knows the lead was accepted.

Activate the workflow, then send a test payload with n8n's built-in "Listen for test event" and a curl request. Watch the execution log to confirm the score, the routing, and the email fire.

The Benefits: Speed, Consistency, and Ownership

Once live, this workflow scores leads in seconds around the clock — no rep has to triage a queue. Hot prospects get a relevant reply while their intent is still fresh, which is the single biggest lever on reply rates. Scoring is consistent because the same rubric runs on every lead; you eliminate the mood-and-fatigue variance of manual review. And because the rubric is plain-English prompt text, marketing can adjust the ICP definition in thirty seconds without a developer. You also get an audit trail: every execution stores the score and the model's reasoning, so when sales disputes a "warm" call, you can see exactly why. Finally, running this on n8n means no per-lead SaaS fee — the cost is a few cents of LLM tokens per lead, and you own the data end to end.

Common Pitfalls to Avoid

Letting the AI return free-form text. If your prompt doesn't force strict JSON, the parser breaks intermittently and leads vanish into failed executions. Always pin the output format and add a fallback in the Code node.

No error handling on external calls. Enrichment APIs rate-limit and ESPs time out. Set "Continue On Fail" on HTTP nodes and attach an Error Trigger workflow that pings you on Slack, so a silent failure never means a lost hot lead.

Scoring without calibration. The model's idea of "hot" may not match yours on day one. Run 30–50 real leads through it, review the scores against outcomes, and refine the rubric wording and tier thresholds. Treat the prompt as a living config.

Blocking the webhook response. If heavy enrichment or a slow LLM call sits between the trigger and the response, your form may time out. For high volume, respond immediately and push the scoring into a sub-workflow triggered via the Execute Workflow node.

Sending automated email to unverified addresses. Add a quick regex or verification step before the Send Email node to protect your sender reputation from typo'd and disposable addresses.

Get these right and Template 164 becomes a quiet, reliable machine: leads in, scored and routed, hot ones answered before your competitor has finished reading the notification.