Marketing automation with AI: less manual work, clear rules

What marketing automation is

Marketing automation means an event (trigger) starts a defined flow without anyone clicking. What’s new is that individual steps can now be AI steps: classifying an enquiry, summarising text, drafting a reply.

A fixed flow with individual AI steps is usually better than an AI agent that decides everything itself: predictable, cheaper, easier to check. When an agent is worth it is covered in AI agents for marketing.

Tools compared

Prices exclude tax, per vendor pages. Allowances (tasks, credits, executions) drive cost more than the base price.

“Try” links are affiliate links (advertising): I earn a commission at no extra cost to you.

Rule of thumb for choosing: if you automate rarely and aren’t very technical, start with Zapier or your newsletter tool’s automations. If you automate a lot or want to keep data in-house, look at n8n. HubSpot pays off when CRM and marketing are meant to grow together anyway.

Five flows that almost always pay off

More on the individual building blocks: email sequences in AI in email marketing, chatbots in AI chatbot for your website, lead generation in AI lead generation.

Lead routing and enrichment

Lead routing means every enquiry reaches the right person fast, with the context they need. AI helps most with free text: what does the person actually want, how urgent is it, does it fit the offer?

• Classify: AI reads the message and suggests topic, urgency and owner.
• Summarise: two sentences for sales instead of a long email.
• Enrich: add company size or industry from public sources, only with a legal basis and notice to the person (Art. 14 GDPR).
• Draft a reply: a draft that a person checks and sends.
• Be fast: an enquiry that sits for two days is often lost.

Careful with automatic marketing emails to new contacts: asking a question isn’t consent to marketing. In Germany, marketing email needs prior express consent under § 7 UWG.

Data protection: processor agreements, US tools, self-hosting

• Map data flows: which data goes from which tool to which? Every connection is a data transfer.
• Data minimisation: pass only the fields the next step needs.
• AI steps: check whether the AI vendor uses data for training, and set that in the settings.
• Access: never store API keys in shared documents; keep permissions as small as possible.

Building in AI steps safely

An AI step in an automation is powerful because it understands free text. It’s also the least reliable point, because it doesn’t always answer the same way.

• Fixed output: let the AI choose only from a list (e.g. topic A, B, C or “unclear”) rather than write freely.
• Allow “unclear”: an “unclear” goes to a person instead of the AI guessing.
• No outward action without sign-off: drafts yes, sending only after review.
• Log it: store the AI step’s input and output so you can find errors.
• Spot checks: look at a few runs every week.

My rule of thumb: if you let AI change things, you’re responsible. Set sign-offs before the automation goes live, not after the first mistake.

Build it yourself or have it built?

Many automations can be built in an afternoon. The hard parts are upkeep, errors and data protection.

My rule of thumb: what needs judgment, context and accountability stays in-house, such as which customer data flows where and who signs off. What can be clearly briefed, measured and swapped goes outside, such as the technical build of an integration.

One flow step by step: from form to CRM

A typical first project: every contact form enquiry should land sorted in the CRM, and the right person gets a short message. In Zapier, Make or n8n it looks almost the same.

1. Trigger: the form tool reports a new submission (via webhook or a ready-made integration).
2. Clean up: trim spaces, lower-case the email, check required fields. Incomplete submissions go to a separate list.
3. Check duplicates: is the contact already in the CRM? Update instead of creating a new one.
4. AI step: the message is assigned a topic from a fixed list, with “unclear” as the way out.
5. CRM entry: create the contact with topic, source and date.
6. Notification: a short message with a summary to the right person.
7. Error path: if a step fails, you get an alert and the submission isn’t lost.

Test every step on its own with real but harmless test data, and mark test enquiries so they don’t show up in reports.

Automations inside your newsletter tool

For email flows you often don’t need a separate automation tool at all. Newsletter tools like MailerLite have their own automations: an event such as joining a group starts a series of emails with delays and conditions.

• Upside: no extra service, no further data transfer, unsubscribes are respected automatically.
• Limit: flows outside email (CRM, chat messages, spreadsheets) need a connection to the outside.
• Careful when editing: some tools pause a running automation when you change an email in it. Reactivate it deliberately afterwards and check that new subscribers flow in again.

More on sequences, subject lines and deliverability in AI in email marketing.

Upkeep: automations age

An automation that runs today can silently fail in three months: a tool changes its API, a field gets renamed, an access token expires. Plan upkeep from the start.

• Keep a list of all automations with purpose, tools and owner.
• Send error alerts to an address someone actually reads.
• Check costs and allowances monthly; successful flows use more.
• Switch off automations you no longer need instead of letting them run.
• After changing forms or fields, test every affected flow.

My rule of thumb: if you let AI change things, you’re responsible, including for the automation nobody has looked at in months.

Lead scoring with AI: useful, but keep it explainable

Lead scoring assigns points to decide which contacts get attention first. Traditionally that follows fixed rules: company email adds points, a pricing-page visit adds points. AI can also rate free text, such as how specific an enquiry is.

• Start with a few understandable rules your sales team gets.
• Have the AI give a one-sentence reason, not just a number.
• After a few weeks, check whether high-scored contacts actually buy more often.
• No contact should get no reply at all just because of an AI score.
• Fully automated decisions with significant effects on people are tightly limited by Art. 22 GDPR; in marketing, a priority order that a person decides on is almost always enough.

Checklist before going live

• Flow described on paper, owners named
• Triggers, conditions and actions tested, including with bad input
• AI steps with fixed output and an “unclear” case
• No outward action without sign-off
• Processor agreements with all tools involved, privacy policy updated
• Only necessary data fields passed on
• Consent for marketing emails checked
• Error alerts set up
• Logging and a weekly spot check planned

Metrics

12 prompts for marketing automation

Here’s a list of my recurring marketing tasks with estimated time per week: [list]. Which three are most worth automating, and which should stay manual? Briefly explain.

Describe this flow as steps with trigger, conditions, actions and exceptions: [description in words]. Mark where a person should sign off.

I want to automate [flow]. My tools: [list]. Compare n8n, Make, Zapier and my newsletter tool’s built-in automation by effort, cost at [number] runs per month and maintenance.

Assign this enquiry to exactly one category: [categories + “unclear”]. Reply with the category name only. Enquiry: [text]

Summarise this enquiry in two sentences for sales: need, urgency, next step. Don’t invent anything not in the message: [text]

Write a reply draft to this enquiry, max 120 words, with one concrete next step. I’ll check and send it myself: [text]

Here’s my automation flow with all tools and fields passed on: [description]. Which data flows where? Which fields are unnecessary? Where do I need a processor agreement? A first pass, not legal advice.

Write ten test cases for this flow, including bad input, empty fields, duplicate sign-ups and special characters: [flow].

My automation stopped with this message: [error + step]. What does it mean in plain words, and how do I fix it?

Here are this week’s numbers from [tools]: [data]. Write a summary with three anomalies and one recommendation, max 150 words.

My flow runs [number] times a month with [steps] steps. Estimate the cost on Zapier, Make and n8n Cloud from their current pricing pages, and say which figures I need to verify there myself.

Write short documentation for this automation: purpose, trigger, steps, tools involved, owner, what to do on errors: [flow].

The most common mistakes

1. Automating chaos. Clarify the process first, then build.
2. Letting AI decide freely. Plan fixed outputs and an “unclear” case.
3. Marketing without consent. An enquiry isn’t newsletter consent.
4. No processor agreements. Every tool with personal data needs one.
5. No error alerts. Silent failures only surface when customers complain.
6. Nobody responsible. Every automation needs a person who maintains it.