How to Use n8n with ._Template 174 Ia Triagem Formulario Leads Inbound

Inbound leads rot fast. A form submission that sits untouched for an hour is dramatically less likely to convert than one answered in five minutes — yet most teams still route every submission into the same undifferentiated queue, where a $50k-intent buyer waits behind a student asking for a free trial. The cost isn't just slow response. It's your best rep spending their morning reading spam, your CRM filling with junk, and your highest-value prospects going cold while nobody notices. This is exactly the problem Template 174 — IA Triagem Formulário Leads Inbound solves: it uses n8n plus an AI model to score, classify, and route every inbound lead the instant it arrives.

The problem: undifferentiated inbound queues kill conversion

The typical inbound flow looks like this: a lead fills out a form (Typeform, a website contact form, HubSpot, a landing page), the submission lands in an inbox or a flat CRM list, and a human eventually triages it by hand. Three things break here.

First, latency. Manual triage introduces minutes-to-hours of delay before anyone even decides who should handle the lead. Second, inconsistency. Two reps score the same lead differently, and nobody applies the qualification criteria uniformly. Third, noise. Real buyers are buried under job seekers, competitors doing recon, students, and outright spam. When everything looks the same in the queue, your team defaults to first-in-first-out — the worst possible ordering for revenue.

The fix isn't more discipline or a longer SLA doc. It's removing the human from the classification step entirely and letting them spend their time only on leads worth their time.

The solution: an AI triage layer in n8n

Template 174 inserts an automated decision layer between the form and your CRM. The moment a lead submits, n8n receives the payload, sends the relevant fields to an AI model, and gets back a structured verdict: an intent score, a segment label, a priority tier, and a routing decision. That verdict then drives everything downstream — which channel gets notified, whether the lead auto-enters a nurture sequence, and how fast a human needs to respond.

The reason n8n is the right tool here is that it sits natively between all these systems. It has trigger nodes for the common form sources, first-class HTTP and AI nodes for the classification call, and a branching model (the Switch and If nodes) that turns the AI's structured output into deterministic routing. You get the flexibility of AI reasoning with the reliability of hard-coded business rules on top.

Step-by-step: building the triage workflow in n8n

1. Capture the submission. Start with the trigger that matches your form source. For a generic website form or an external tool, use a Webhook node set to POST and copy its production URL into your form's submit action. If you use Typeform or a native integration, use the dedicated Typeform Trigger or HubSpot Trigger node instead — it handles auth and payload parsing for you. Pin a sample submission so you can build against real data.

2. Normalize the fields. Add a Set (Edit Fields) node right after the trigger to map the raw form payload into a clean, predictable shape: name, email, company, message, company_size, budget. This decouples the rest of the workflow from any one form's field naming, so when you add a second form source you only touch this node.

3. Enrich (optional but high-leverage). Before scoring, an HTTP Request node can pull the email domain and check whether it's a free provider (gmail, outlook) or a corporate domain — a strong buying-intent signal. You can also hit a lightweight enrichment API to attach company size or industry. Even a simple regex in a Code node that flags free-email addresses adds real signal at zero cost.

4. Score with AI. This is the core. Add the AI Agent node (or a basic Message a Model node) connected to an Anthropic Chat Model sub-node — use claude-haiku-4-5 for triage, since it's fast and cheap and classification doesn't need a frontier model. Your system prompt should define the exact criteria and, critically, demand structured JSON output. For example:

"You are a lead qualification engine for a B2B automation product. Given the lead fields, return ONLY a JSON object with keys: intent_score (0-100), segment (one of: enterprise, smb, individual, spam), priority (hot, warm, cold), and reason (one sentence). Score enterprise leads with clear budget and a business email highest. Score free-email addresses with vague messages as spam or cold."

To guarantee parseable output, attach a Structured Output Parser to the AI node with a schema matching those four keys. This turns the model's response into typed fields the rest of the workflow can branch on reliably, instead of you regex-scraping prose.

5. Route on the verdict. Feed the parsed output into a Switch node keyed on priority. Build three outputs:

— Hot: immediate Slack node to your #sales-hot channel tagging the on-call rep, plus a HubSpot/Pipedrive node that creates the deal and sets owner. Optionally a Gmail node fires an instant personalized acknowledgment.
— Warm: create the CRM contact and enroll it in a nurture sequence; notify the channel without an @-mention.
— Cold / spam: log to a Google Sheets or Airtable row for audit and stop — no human touch, no CRM pollution.

6. Always persist the decision. On every branch, append a row to a Sheet or database with the original lead, the AI score, and the reason. This log is what lets you audit and tune the prompt later. Finish with a Respond to Webhook node returning 200 so the form doesn't show an error to the user.

The benefits: speed, consistency, and focus

Once this is live, response time on hot leads drops from hours to seconds — the Slack ping fires before the prospect has closed the tab. Qualification becomes consistent: the same criteria are applied to every single lead, 24/7, with no Monday-morning backlog and no weekend blind spot. Your reps stop triaging and start selling, because the only leads that reach them are pre-scored and pre-segmented. And because every decision is logged with a reason, you get a feedback loop: review misclassified leads weekly, tighten the prompt, and the system gets sharper over time. The cost is trivial — a Haiku-class classification call is a fraction of a cent per lead, far cheaper than a minute of a rep's attention.

Common pitfalls and how to avoid them

Unstructured AI output. The single most common failure is parsing free-form model text. Always use the Structured Output Parser with an explicit schema and instruct the model to return JSON only — never build downstream logic on prose.

No fallback path. If the AI node errors or the model returns something unparseable, don't let the lead vanish. Set the node's "On Error" behavior to continue, and add a default Switch branch that routes anything unscored to a human queue. A lead you can't classify is still a lead — fail safe, not silent.

Over-trusting the score. Keep hard business rules outside the model. If a known-strategic domain submits, force it to hot with an If node regardless of what the AI says. AI is for the fuzzy middle; deterministic rules should guard the edges.

Webhook security. An open webhook can be spammed. Add a secret token check in a Code or If node at the top of the flow, or validate the form source, so you're not paying for AI calls on garbage traffic.

Prompt drift. As your product and ICP evolve, a prompt written six months ago will silently mis-score. Schedule a monthly review of the decision log, look at what got mislabeled, and update the criteria. Treat the prompt as a living config, not a set-and-forget string.

Set up this way, Template 174 turns your inbound form from a flat, lossy queue into a real-time revenue router — one that never sleeps, applies your qualification standard perfectly every time, and puts your team's attention exactly where the money is.