AI automation: how to automate your business with AI

You've probably heard that AI can "automate everything." That's not true. But AI can automate quite a lot, and what it can actually automate can save you hours every week.

This guide is about how to identify what's worth automating, which tools you need, and how to actually do it. With concrete examples, not theoretical fluff.

What is AI automation?

Let's clear up the terminology. There are three levels of automation:

Traditional automation

Rule-based. "If X happens, do Y." No AI models involved. Example: an email that sends automatically when an invoice is paid.

RPA (Robotic Process Automation)

Software that mimics human clicks. Logs into systems, copies data, fills in forms. Still rule-based, but more advanced.

AI automation

Automation where an AI model makes decisions. Instead of fixed rules, the AI analyzes the situation and acts based on understanding. Example: an AI that reads incoming emails, assesses whether it's a lead inquiry, extracts relevant information, and creates a ticket in the CRM.

It's the third level we're focusing on. AI as the brain, automation tools as the body.

Where does AI automation make the biggest impact?

Not everywhere. AI automation works best where you have:

1. Repetitive tasks that require some judgment
2. Unstructured data (text, emails, documents) that needs interpretation
3. Decisions that follow patterns but aren't simple yes/no rules
4. Volume that makes manual handling time-consuming

Here are the most common use cases:

Customer communication

• Categorize incoming emails by topic and priority
• Generate suggested replies that customer service can edit and send
• Summarize long email threads for quick overview

Lead handling

• Qualify leads based on email content or form data
• Enrich contact information automatically
• Create personalized follow-up emails

I've built a concrete system for this. Read my guide on building an AI agent for leads with Claude and n8n (in Swedish).

Reporting and summarization

• Summarize weekly reports from different departments
• Generate insights from customer reviews or feedback
• Create dashboard texts from raw data

Content production

• Transform blog posts into social media posts
• Generate product descriptions from specifications
• Create newsletters based on the week's updates

Invoicing and finance

• Read and categorize invoices (OCR + AI)
• Match invoices against orders
• Flag discrepancies for manual review

The automation audit: find what to automate

Before you start building, take a day and do an audit. It will save you weeks of automating the wrong things.

Step 1: List your repetitive tasks

Write down everything you do more than once a week. Don't be too ambitious, focus on the obvious:

• Emails I send regularly
• Data I move between systems
• Reports I compile
• Responses I give to customers

Step 2: Score each task

For each task, assess:

Tasks with high total scores (15+) are your best candidates.

Step 3: Prioritize based on ROI

Start with the task that gives the most time savings with the least complexity. Your first automation project should be a "quick win" that demonstrates value, not a six-month project.

The tool stack: what you need

You need two things: an automation platform and an AI model.

Automation platforms

n8n (my recommendation)

• Visual workflow editor
• Self-hosted (free) or cloud (from €20/month on annual billing, €24 monthly)
• Support for hundreds of integrations
• Full control over your data
• My beginner's guide to n8n

Make

• More user-friendly interface
• Great for non-technical users
• Pay per operation

Zapier

• Most pre-built integrations
• Easiest to get started
• Most expensive at high volume

AI models (the brain)

• Claude API: Best for text analysis, assessment, and generation. My favorite.
• OpenAI API (GPT-5 family): Broad capability, large ecosystem
• Gemini API: Good for Google-integrated workflows

The third option: agent tools

The industry's focus is shifting. You no longer always need an automation platform. Agent tools like Claude Code, Claude Cowork, and OpenAI's Codex can now handle basically any computer task: reading folders, compiling reports, operating other programs, and running scheduled tasks in the background, even in the cloud while your computer is off.

Both companies are also building full work suites around the agents. Beyond Claude Code and Cowork, Anthropic has launched Claude Design for visual material and Claude Science for research work. OpenAI has rebuilt Codex into a general work agent with computer use and 90+ plugins covering Microsoft 365, Notion, and Slack, plus ChatGPT agent, Deep Research, and shared Workspace agents for teams.

Rule of thumb: choose n8n when the flow should be deterministic, run at volume, and connect to many systems. Choose an agent when the task requires judgment, is hard to describe as a fixed flow, or is simply something you would have done at the computer yourself. Several of my own "automations" these days are in practice a scheduled Claude agent with clear instructions.

Workflow integrations

Depending on your stack, you'll need to connect:

• Email: Gmail, Outlook
• CRM: HubSpot, Pipedrive, Salesforce
• Finance: QuickBooks, Xero, FreshBooks
• Communication: Slack, Teams
• Documents: Google Drive, SharePoint

3 concrete automation recipes

Enough theory. Here are three automations you can build today.

Recipe 1: Automatic lead qualification

The problem: You get 20 inquiries per week via email or contact forms. You spend 30 minutes per day reading, assessing, and responding.

The solution:

1. Trigger: New email/form submission comes in
2. AI step: Claude reads the content and assesses:
   • Is this a potential customer? (yes/no/maybe)
   • What type of service is being requested?
   • How urgent does it seem?
   • Contact information extracted
3. Routing: Based on the AI's assessment:
   • High priority → Slack notification to you + CRM entry
   • Medium → CRM entry + automatic response with more info
   • Low/spam → Archived

Time saved: ~2.5 hours per week Difficulty: Medium

Recipe 2: Weekly report from customer reviews

The problem: You have customer reviews coming in on Google, Trustpilot, or via email. You never have time to analyze them properly.

The solution:

1. Trigger: Scheduled every Friday at 9:00 AM
2. Data collection: Fetch new reviews from all sources
3. AI step: Claude analyzes and generates:
   • Summary of the week's feedback
   • Most common positive and negative themes
   • Suggested actions
   • Sentiment trend compared to last week
4. Output: Report sent to Slack or email

Time saved: ~1 hour per week Difficulty: Low

Recipe 3: Content repurposing

The problem: You publish a blog post but never have time to create social media posts, newsletter snippets, or LinkedIn posts from it.

The solution:

1. Trigger: New blog post published (webhook)
2. AI step: Claude reads the post and generates:
   • 3 LinkedIn posts from different angles
   • 1 short summary for newsletter
   • 5 Twitter/X threads with key insights
3. Review: Drafts sent to you for approval
4. Publishing: After approval, published automatically

Time saved: ~2 hours per blog post Difficulty: Medium

ROI: is it worth it?

Let's run an example with assumed numbers. Adjust for your own business. Say you or your employees spend 10 hours per week on tasks that can be automated.

That doesn't include the reduced error rate. Or the fact that your staff can focus on more valuable work.

Common mistakes (and what NOT to automate)

Mistake 1: Automating the wrong things

Don't automate tasks that:

• Require empathy and human connection (difficult customer conversations)
• Have legal consequences if wrong (contract documents without review)
• Happen so rarely that the automation costs more than it saves

Mistake 2: Building too complex from the start

Your first automation project should be simple. One trigger, one AI step, one output. Build complexity gradually.

Mistake 3: No error handling

AI makes mistakes. Your automation needs to handle that. Always add:

• A fallback for when the AI is uncertain
• Human review for critical decisions
• Logging so you can track and improve

Mistake 4: Ignoring maintenance

Automations need maintenance. APIs change, business rules update, AI models improve. Schedule a monthly check of your active automations.

Get started today

Here's your plan:

1. Do the automation audit (1 day). List and score tasks.
2. Pick one task to automate. The simplest, not the most important.
3. Set up n8n (my guide) or Make.
4. Build a prototype in an afternoon.
5. Test for a week alongside manual handling.
6. Evaluate and iterate.

Want to understand the technology behind AI tools? Read my guide on what generative AI is. Need to choose the right tools? Check out my comparison of the best AI tools in 2026.

Frequently asked questions

Do I need to know how to code to automate with AI?

Not necessarily. Tools like n8n and Make have visual editors where you drag and drop. But basic technical understanding helps, especially for connecting systems via APIs and troubleshooting.

How much does AI automation cost?

A basic setup with n8n (self-hosted) and Claude API can cost under $50 per month. With cloud solutions and more tools, you'll spend $100 to $300. Compare that to the time you save, and it almost always pays for itself.

Is AI automation safe for sensitive data?

With the right setup, yes. Use business plans with DPA agreements. Self-hosted n8n keeps all data on your own server. Avoid sending personal data through free versions of AI tools.

How long does it take to build an automation?

A simple automation (email to Slack notification with AI analysis) can be built in an afternoon. More complex flows with multiple systems and conditions take one to two weeks. Start simple and build out gradually.

Can AI automation integrate with my existing tools?

Almost certainly. Tools like n8n and Make support hundreds of integrations including CRM systems, email, Slack, accounting software, and more. If a tool has an API, it can usually be connected.

Do I need n8n if I'm already using Claude Code or Codex?

Both, really. The agents shine when a task requires judgment or is hard to describe as a fixed flow, and they can now handle most things you'd do at a computer. n8n is best when the same flow needs to run reliably at volume and connect to many systems. In practice they complement each other: build fixed flows in n8n and let an agent take the messier tasks.

What happens if the AI makes a mistake in an automation?

Always build in error handling. Have a fallback for uncertain decisions, add human review for critical steps, and log everything. AI automation should assist, not replace human judgment in important decisions.

---

Want to automate your business with AI but don't know where to start? Book a call and we'll create an automation plan for your specific business.