Automation Fieldbook
Field note

October 6, 2026

Build Your First AI Lead-Triage Automation: A Step-by-Step Guide for Non-Coders

Why lead triage is the perfect first automation

Every small business has the same quiet leak: leads arrive — contact forms, emails, DMs, referrals — and sit. Someone means to follow up, gets busy, and three days later the lead has gone cold or chosen a competitor. Studies on lead response times all point the same direction: the business that responds first usually wins, and "first" increasingly means "within minutes, not days."

Lead triage is the perfect first automation because it's high-value, easy to understand, and hard to get dangerously wrong. You're not replacing judgment; you're doing the boring sorting work instantly so a human can exercise judgment faster. Here's the whole idea in one sentence: when a new lead arrives, AI reads it, scores it, writes a two-line summary, and routes it to the right place — before you've finished your coffee.

I built this exact workflow for a friend who runs a small web design studio. It took about two hours including testing, and it now handles roughly 40 leads a month. Total running cost: under $15/month. Here's how to build yours.

What you'll need (15 minutes of setup)

That's it. No code, no server, no developer.

Step 1: Map your current process on paper (10 minutes)

Before touching any software, grab a piece of paper and answer three questions:

  1. Where do leads come from? (List every source, then circle the top one.)
  2. What makes a lead good? (Budget? Timeline? Location? Company size? Write down your actual gut criteria — "has a real project, budget over $2k, not a student asking for free advice.")
  3. What happens to a good lead vs. a bad lead today? (Good: you call them. Bad: they sit in your inbox forever?)

This paper sketch becomes your automation spec. Most failed automations fail here — not in the software, but because nobody defined what "good" means. The AI can only score leads against criteria you give it.

Step 2: Create the trigger — "when a new lead arrives" (10 minutes)

In Make, create a new scenario and add your lead source as the trigger module:

n8n note: same idea — Form Trigger, Gmail Trigger, or Webhook node. Zapier note: same — pick your form app as the trigger.

Run the trigger once manually and confirm it pulls in a real lead. If it can't see your test data, nothing downstream will work — fix it here, not later.

Step 3: Add the AI scoring step (20 minutes)

This is the heart of it. Add an AI module (in Make: the OpenAI "Create a Completion" module, or Make's built-in AI; in n8n: the AI Agent or OpenAI node; in Zapier: an AI step) and give it a prompt like this:

You are a lead qualifier for [your business type]. Score this lead 1–10 and reply in exactly this format:
SCORE: [number]
SUMMARY: [two sentences max — who they are, what they want]
RED FLAGS: [any, or "none"]

A score of 8+ means: [your criteria from Step 1, e.g. "real project, budget mentioned or implied over $2k, timeline within 3 months"].
A score below 5 means: [e.g. "vague request, no budget signal, likely price-shopping or spam"].

Lead details: [map in the name, email, message, and any form fields here]

Three things make this work instead of producing confident nonsense:

  1. Force a strict output format. "Reply in exactly this format" lets the next step parse the score reliably. Without it, the AI writes lovely paragraphs your automation can't read.
  2. Put YOUR criteria in the prompt. Generic prompts produce generic scores. The ten minutes you spent on paper in Step 1 is what makes this actually useful.
  3. Keep it cheap. Use a small model (GPT-4o-mini class or equivalent). You're classifying short text, not writing novels. The expensive models add nothing here except cost.

Test with 3–5 real past leads — good ones, bad ones, and a spammy one. If the scores don't match your gut, adjust the criteria wording, not the model. Prompt wording is the whole game.

Step 4: Add the routing logic (15 minutes)

Now branch based on the score. In Make, add a Router (free — routers don't consume credits) with three paths:

The polite auto-reply matters more than you think. Most small businesses either ignore cold leads (rude, and occasionally the "cold" lead was actually fine) or waste hours on them. An instant, helpful auto-response with pricing and FAQs converts a surprising number of "cold" leads into warm ones — and it costs you nothing.

Step 5: Log everything to one sheet (10 minutes)

Add a final step on every path: append a row to a Google Sheet with timestamp, name, score, summary, and which path it took. This is your audit trail and your training data. Once a month, spend 15 minutes scanning it: are the scores matching reality? Did any hot lead turn out to be junk, or vice versa? Adjust your Step 3 prompt accordingly.

This feedback loop is what separates a workflow that slowly rots from one that gets smarter. The automation doesn't learn on its own — you learn from its log and improve the prompt. Budget 15 minutes a month. That's the entire maintenance cost.

Step 6: Test like you mean it (20 minutes)

Before going live, run these five tests:

  1. A perfect lead (should score 8+, hit your phone).
  2. A vague "just curious" message (should land in nurture).
  3. Obvious spam (should score 1–2, get the auto-reply, never bother you).
  4. A lead with missing fields (empty budget field, no timeline — make sure nothing errors out).
  5. Ten rapid-fire submissions (make sure the scenario doesn't choke or double-send).

For each test, check: did the right path trigger? Did the CRM entry look right? Did the human get notified correctly? Fix failures now — they're ten times more annoying live.

Step 7: Go live, then watch it for a week

Turn it on. For the first week, glance at the log sheet daily — not because it'll break, but because you'll spot prompt improvements immediately. Common first-week tweaks: the AI scores too generously (tighten criteria), the summary misses the actual ask (tell it what to prioritize), or a lead source you forgot about isn't connected.

After week one, drop to the monthly 15-minute review. That's it. You're done.

Common mistakes (so you can skip them)

What this costs

Build cost: ~2 hours of your time, $0 (all platforms have free tiers for testing).
Running cost, ~40 leads/month:
- Make Core: $9/month (this workflow uses roughly 200–400 credits/month — trivial)
- AI (OpenAI API, small model): $1–3/month
- CRM: $0 (HubSpot free tier) | Notifications: $0 (Slack/email)
- Total: roughly $10–12/month

n8n alternative: self-hosted = $0 + ~$5–10/month server; each lead = 1 execution regardless of steps.
Zapier alternative: Professional from $19.99/month annual; each lead burns 4–6 tasks (AI step counts extra — check current AI metering).

The math that matters: if this workflow saves you from losing even one decent lead per quarter, it has paid for itself many times over. My friend's studio credits it with two extra projects in six months — roughly $8,000 in revenue against ~$70 in costs.

Your homework

Don't build the perfect version. This week: do Step 1 on paper (10 minutes), sign up for Make's free tier, and connect just your main lead source as a trigger. That's a 30-minute task. Next week, add the AI scoring. The week after, routing. Small businesses don't fail at automation from lack of ambition — they fail from trying to boil the ocean on day one. Start with the leak that's costing you money right now: the leads sitting in your inbox.

Enjoyed this? Join the newsletter for one practical automation tip a week. No spam, no hype.

← Back to all field notes