You have read the articles. You have seen the demos. You know AI automation is the future.

But you still have not built one.

No judgment. Most people are stuck in the same place. The gap between "AI is amazing" and "I built something with AI" feels enormous. It is not. You can cross it today, in about the time it takes to watch an episode of television.

What We Are Building

A workflow that takes a raw client inquiry (email, form submission, Slack message) and automatically:

  1. Extracts the key details (name, what they need, budget signals, urgency)
  2. Drafts a personalized response using Claude
  3. Sends the draft to you for one-click approval

This is not a toy. This is the exact pattern behind most high-performing client response systems. Once you understand it, you can apply it to proposals, support tickets, content briefs, anything that follows an input-process-output pattern.

What You Need

  • A free Make.com account (the free tier gives you 1,000 operations per month, plenty)
  • A Claude API key (or use the free tier of any LLM, the concepts transfer)
  • 30 minutes of focus

That is it. No code. No servers. No DevOps.

Step 1: Set Up the Trigger (5 minutes)

Every workflow starts with a trigger, the event that kicks everything off.

In Make.com, create a new scenario and add a Webhook module. This gives you a URL. Anything sent to that URL becomes the input to your workflow.

For testing, you will use the built-in test feature. In production, you would connect this to your contact form, email inbox, or Slack channel.

Pro tip: Name your webhook something real, like Client Inquiry Intake. Future you will thank present you when you have 20 scenarios running.

Step 2: Extract the Signal from the Noise (10 minutes)

Raw client messages are messy. They ramble. They include signatures and disclaimers. Your workflow needs the signal.

Add an HTTP module that calls the Claude API. The prompt is the key:

Extract the following from this client inquiry:
- Client name
- What they need (1 sentence)
- Estimated budget (if mentioned, otherwise not stated)
- Urgency level (low/medium/high based on language)
- Key detail that makes this inquiry unique

Respond in JSON only.

This is the moment most people overcomplicate things. You do not need a fine-tuned model. You do not need embeddings. You need a clear prompt and structured output.

Step 3: Generate the Response (10 minutes)

Add another HTTP module calling Claude. This time, you are drafting the reply:

You are a professional but warm business development assistant.
Draft a reply to this client inquiry.

Rules:
- Acknowledge their specific need in the first sentence
- Reference the unique detail so it does not feel templated
- If urgency is high, offer a call this week
- Keep it under 150 words

Notice what is happening: you extracted structured data in Step 2, then used that structure to generate a personalized response. This is the core pattern of useful AI automation: structure first, generation second.

Step 4: Deliver for Review (5 minutes)

Add a final module to send the drafted response somewhere you will see it. Options:

  • Email to yourself with the original message and draft reply
  • Slack message in a dedicated channel
  • Google Sheet row (great for tracking response patterns over time)

Include both the extracted data and the draft, so you can review before sending.

Turn It On

Activate the scenario. Send a test message to your webhook URL. Watch the entire chain fire in seconds.

You just built an AI workflow. It is not hypothetical anymore.

Why This Pattern Matters

The specific workflow above is useful, but the pattern is what matters:

Trigger > Extract/Structure > Generate > Deliver

Once you see it, you see it everywhere:

  • Content pipeline: RSS trigger, extract key points, generate social posts, send to approval queue
  • Support system: ticket created, classify urgency and topic, draft response, route to right team member
  • Sales intel: new lead in CRM, research company, generate personalized outreach, draft email

Every one of these follows the same four-step pattern. The modules change. The logic does not.

Common Mistakes to Avoid

Overbuilding on day one. Your first workflow should be simple enough to build in 30 minutes. You can add complexity later. Most people who fail at automation fail because they try to build the final version first.

Skipping the extraction step. It is tempting to throw raw input directly at the generation prompt. Do not. Structured extraction makes everything downstream better.

Not reviewing outputs. Keep a human in the loop for the first 50 runs. AI is good, not perfect. You will catch edge cases that make your prompts better.

What Comes Next

Once you have one workflow running, the second one takes 15 minutes. The third takes 10. You start seeing automation opportunities everywhere, not because the technology changed, but because you changed.

That shift is worth more than any course or certification.

But if you want to go deeper: multi-step chains, error handling, conditional logic, the full production playbook, the AI Automation Toolkit walks through everything.

Your first workflow is done. Your second one is waiting.

Nova writes for Like One. She believes the best automation is the one you actually build.

Keep Reading