Your first AI automation, with no code

The first automation should replace a repeated sequence of clicks, not run your business. The safest start is a sequence you already do by hand: read an email, classify it, write a draft, note the line somewhere. Below is that flow in Make, so that a draft lands in your mailbox and nothing is sent.

What you decide first

• One inbox and one label, for example “Enquiry”. Only mail with that label starts the scenario. Everything else stays untouched.
• Three categories you actually use. For example: new offer, existing client, other. More categories make the first run unreadable.
• A sheet with five columns: date, sender, subject, category, draft.
• The rule for the first month: a draft is created, never a sent email.

The scenario in Make

Build a scenario with four modules. The names inside Make change from time to time. The order does not.

• Gmail: Watch emails, filtered to the label. An interval of 15 minutes is enough.
• An AI module (OpenAI, Claude, or Make’s own): the input is subject plus body. The output is JSON only, see the prompt below.
• Google Sheets: Add a row, mapping the JSON fields into the five columns.
• Gmail: Create a draft, not Send an email. The recipient is the sender, the body is the draft field.

Save it and run a single test email through before you switch the scenario on. If the JSON breaks, you see it in the sheet row, not at the client.

Classify the email and write a reply draft.

Categories, only these: new offer | existing client | other.
Reply only as JSON with the keys category and draft.
The draft is a short reply, 90 words at most, with one question you need in order to answer.
Do not invent prices or dates. If the price is missing, ask for it.
No sign-off with an invented name. Close with the placeholder [Name].

Subject: {{subject}}
Body: {{body}}

What goes wrong in the first week

Newsletters and notifications slip into the label if the filter is only “not from me”. Tighten the label by hand until the scenario sees only real enquiries for a week. Second trap: the model writes a price you never gave. That is why the ban is in the prompt, and why you read the draft before it leaves the mailbox. Third trap: duplicate rows when Make retries a mail after an error. A “mail id” column and a filter “only if this id is not already there” stops it.

When you take the next step

Only after you have sent ten drafts in a row with a one-sentence correction at most is automatic sending even a conversation. Before that, the draft is the whole gain: you open the mail and press send, instead of starting from zero. That is often ten minutes per enquiry.

The calm start, without a jungle of tools, is in AI Automation. When one scenario becomes several, continue with Automation Pro, and when a task should finish without you, with AI Agents. If you want to sell this, read AI services that sell today.