How to Build Your First AI Marketing Agent — Step-by-Step (No Coding)

The short answer: You can build a working AI marketing agent in 90 minutes today, no coding required, using n8n + Claude API + a free WhatsApp Cloud API account. This guide walks through the exact 7-step build for a Lead-Qualifier Agent — the most useful starter agent for any Indian SMB or agency.

What You’ll Build

A WhatsApp lead-qualifier agent that:

1. Receives inbound WhatsApp messages on your business number
2. Asks 3 qualifying questions (budget, timeline, requirement)
3. Scores the lead based on answers
4. Pushes hot leads to Zoho CRM with tags
5. Auto-replies to cold leads politely
6. Notifies you on Slack/email for hot leads

Cost to run: ₹500–₹1,200/month at moderate volume (200–500 inbound messages/day). Saves a typical Nashik agency ~1.5 hours of admin per day.

Prerequisites (15 minutes setup)

1. n8n instance — spin up free on Railway (railway.app) using their n8n template. 5 min, no credit card.
2. Claude API key — sign up at console.anthropic.com, add $5 in credits.
3. WhatsApp Cloud API — free tier from Meta Business Manager. 10 min setup, requires Facebook page.
4. Zoho CRM account — free tier sufficient for testing.

If any of these stalls, the OnePlace AI Agent Building Course bundles all four with paid licenses included. See ai-agent-building-course-nashik.

Step 1 — Create the n8n Workflow Skeleton

In n8n: New Workflow → add a Webhook trigger node. Save the URL n8n gives you — you’ll paste it into Meta’s WhatsApp webhook field in Step 2.

Add a Switch node right after, routing on the inbound message type (text, image, audio). For this build, route “text” messages to the next step. (Other types can be added later.)

Step 2 — Connect WhatsApp Cloud API

In Meta Business Manager → WhatsApp → Webhooks: paste the n8n webhook URL. Subscribe to messages events. Test: send a WhatsApp message to your business number — n8n should fire instantly.

If n8n doesn’t fire, check (a) the webhook is verified in Meta, (b) the n8n workflow is set to Active. 90% of beginner bugs are one of these two.

Step 3 — Add the Claude Reasoning Node

Add an n8n HTTP Request node calling POST https://api.anthropic.com/v1/messages with your Claude API key. The body:

{
  "model": "claude-3-5-sonnet-20241022",
  "max_tokens": 500,
  "system": "You are a lead qualifier for [BUSINESS NAME] in Nashik. Ask the user 3 questions one at a time: (1) what service they need, (2) budget range, (3) timeline. After all 3 answers, output JSON {score: 1-10, summary, recommended_action}.",
  "messages": [{"role": "user", "content": "{{ $json.message }}"}]
}

n8n’s native Claude / Anthropic node also works — use whichever is convenient. The system prompt is where you embed your business rules.

Step 4 — Multi-Turn Memory (the Agentic Part)

For the agent to ask 3 questions in sequence, you need conversation memory. Two options:

1. Simple: Store the conversation in a Google Sheet (one row per WhatsApp number). Read it back into Claude on each inbound message.
2. Advanced: Use Supabase Vector or Pinecone for proper conversational state.

For a first build, use Google Sheets. Add a node that reads the WhatsApp number’s history, appends the new message, and feeds it back to Claude.

Step 5 — Score and Route

Once Claude detects all 3 answers have been provided, it outputs JSON with the lead score. Add an n8n IF node: if score ≥ 7, route to “hot lead”; otherwise “warm/cold”.

Step 6 — Hot Leads to Zoho + Slack

For hot leads:

1. Add a Zoho CRM node — Create Lead with the qualified data and a tag like “hot-whatsapp-agent”.
2. Add a Slack node — send a message to the sales channel with the lead summary.
3. Reply to the WhatsApp user with: “Thanks — we’ll call you within 30 minutes.”

Step 7 — Cold/Warm Leads to Polite Reply

For warm or cold leads: reply with a polite acknowledgement and tag them in Zoho with “warm-whatsapp-agent” for later nurture. Add a follow-up campaign with a 72-hour delay if no human action.

Test, Deploy, Iterate

Test with 10 manual WhatsApp messages spanning the qualifying range. Check Zoho records, Slack alerts, and reply quality. Once stable, set the workflow to Active and let it run.

Watch for the first 48 hours: token cost, false positives, edge cases (typos, mixed Hindi/English). Adjust the system prompt as needed. Most agents reach production stability within 2 weeks of iteration.

Cost Math (Real)

For an agency closing one additional ₹50K project per month thanks to faster lead response, ROI is roughly 40× on infrastructure.

What to Build Next

1. SEO content agent — weekly blog publisher
2. Reporting agent — daily KPI summary email
3. Customer-support agent — RAG over your knowledge base
4. Sales-outreach agent — personalised cold-email sender

Each builds on the same scaffold. Once you have 1, you have 5.

FAQ

1. Do I really need no coding to build an AI marketing agent?

For 80% of marketing use cases, yes. n8n + Claude + simple JSON outputs are sufficient. The 20% that needs JavaScript can be written by Cursor / Claude for you.

2. How long does my first agent take to build?

90 minutes for the basic scaffold; 1 week of iteration to production stability. Most students in the OnePlace cohort have their first agent live by Day 5.

3. What is the minimum monthly budget?

~₹500–₹1,500/month per agent at moderate volume. Most agents pay back in week 1 via saved admin time or one accelerated deal close.

4. Can the agent handle Hindi or Marathi?

Yes. Add this to your system prompt: “The user may write in Hindi, Marathi, English or Hinglish. Detect the language and respond in the same.” Claude handles this natively with high accuracy.

5. What happens when the agent makes a mistake?

Build a human-in-the-loop escalation path from day one. Anything the agent isn’t 70%+ confident about — route to a human with the context. This pattern catches most errors before they reach the customer.