Quick answer: To build your first AI automation without code, follow four steps: (1) Pick one repetitive task that follows clear rules — email triage, lead capture, or invoice processing are ideal starters. (2) Choose a no-code platform: Make (£7/month, beginner-friendly) or n8n (free, self-hosted, more power). (3) Build a simple workflow: trigger → AI decision → action. (4) Test with 5 real examples, then let it run in supervised mode for one week before going hands-off. Most small business owners complete their first working automation in 90 minutes to 2 hours. The most common first automation — email triage — saves 8-12 hours per week on average.
Why build your first automation now?
The numbers are impossible to ignore. Small business owners spend 15-20 hours per week on tasks that follow the same pattern every time — sorting emails, copying data between apps, generating reports, chasing follow-ups. That is nearly half the working week spent on work that a machine can do faster, more consistently, and for under £10/month.
The SBE Council's 2026 Small Business Tech Use Survey found that 89% of small businesses now leverage AI to automate repetitive tasks. Employees save an average of 5.6 hours per week. For business owners and managers, that number rises to 7.2 hours.
The tools are ready. The cost is trivial. The only question is which task to automate first.
What can you actually automate? (Real examples)
Here are the five automations that small businesses build most often in 2026 — with real time savings:
| Automation | What it does | Setup time | Time saved/week | Monthly cost |
|---|---|---|---|---|
| Email triage | Reads incoming emails, categorises them, drafts replies, flags urgent items | 60-90 min | 8-12 hours | £0-7 |
| Lead capture → CRM | New form submission → creates contact → sends welcome email → notifies you | 30-45 min | 3-5 hours | £0-7 |
| Invoice processing | Forwards invoice → extracts vendor/amount/date → creates accounting record | 45-60 min | 3-5 hours | £0-7 |
| Social media scheduler | Write once in a sheet → AI generates variations → schedules across platforms | 30 min | 2-4 hours | £0-15 |
| Meeting note → CRM | AI note-taker joins call → summary + action items → pushed to CRM | 20 min | 2-3 hours | £8-10 |
Every one of these replaced a manual process that cost real money every week.
Step 1: Pick the right tool for the job
Two platforms dominate no-code automation for small businesses. They solve different problems.
Make (formerly Integromat) — start here if you are new
Make uses a visual drag-and-drop builder. You see your workflow as a diagram — trigger on the left, actions flowing right, branches splitting off. It is designed for people who think visually rather than in code.
| Feature | Make |
|---|---|
| Learning curve | Shallow — first automation in 30 min |
| Visual builder | Drag-and-drop scenario editor |
| App integrations | 1,500+ |
| AI capabilities | Native AI modules (OpenAI, Anthropic, etc.) |
| Free tier | 1,000 operations/month |
| Paid starts at | £7/month |
| Best for | Most small business automations |
n8n — graduate here when you outgrow Make
n8n is open-source and self-hosted. You install it on a £5/month VPS and it runs forever with no monthly platform fee. It has a built-in AI agent framework (LangChain integration) that Make cannot match. The trade-off: the interface is more technical.
| Feature | n8n |
|---|---|
| Learning curve | Moderate — first automation in 60-90 min |
| Visual builder | Node-based workflow editor |
| App integrations | 400+ (fewer than Make, but covers the essentials) |
| AI capabilities | Full LangChain integration for AI agents |
| Cost | Free (MIT license), self-hosted |
| Ongoing cost | Your VPS cost (typically £5/month) |
| Best for | AI agents, complex logic, cost-sensitive users |
Recommendation: Start with Make's free tier. Build your first automation. If you hit Make's limits — or want to build an AI agent rather than a simple automation — migrate to n8n. The thinking is the same; only the interface changes.
We use both platforms daily in our AI Kickstart workshop — you build your first automation during the session, whichever platform you choose.
Step 2: Map your workflow before touching any tool
The #1 mistake beginners make is jumping into a platform and clicking around. You do not know what to click because you have not decided what the automation should do.
Take 10 minutes. Write this down on paper or in a note:
1. Trigger: What starts the automation?
- "A new email arrives in my inbox"
- "Someone submits the contact form on my website"
- "I receive an invoice PDF as an email attachment"
2. Steps: What happens next? List them in order.
- "Read the email content"
- "Ask AI to categorise it as URGENT, QUERY, INVOICE, or GENERAL"
- "If URGENT: send me a Slack message"
- "If QUERY: draft a reply and save to drafts"
- "If INVOICE: forward to accounts@ and create a task in my project tool"
- "If GENERAL: archive"
3. Output: What should exist when the automation finishes?
- "A draft reply sitting in my Gmail drafts folder"
- "A new contact in my CRM"
- "A Slack notification on my phone"
This is your blueprint. Now you can build it in any platform in under an hour.
Example blueprint — Email triage:
Trigger: New email in Gmail
Step 1: Extract sender, subject, body Step 2: Send to AI with prompt: "Categorise this email. Reply categories: URGENT (client emergency, deadline change), QUERY (question about services, pricing, availability), INVOICE (attached or mentioned invoice), GENERAL (newsletter, notification, spam). Return only the category and a 2-sentence draft reply." Step 3: Route based on category: - URGENT → Slack DM to owner + mark email as starred - QUERY → Save AI draft to Gmail drafts + add label "To Review" - INVOICE → Forward to accounts@ + create Asana task - GENERAL → Archive
This took 90 minutes to build in Make. It now handles 200+ emails per week — automatically.
Step 3: Build and test (the 5-example rule)
Now open your chosen platform and build the workflow. Follow your blueprint step by step. Both Make and n8n let you test each step individually — use this. Do not build the whole thing and then test. Build one step, test it, build the next.
The 5-example rule: Before you let the automation run on its own, run 5 real examples through it. Not test data. Real emails. Real form submissions. Real invoices.
You will find edge cases:
- A lead email mentions "cost" but is actually asking about "cost of living" — the AI will miscategorise it
- A form submission arrives in Portuguese but your AI prompt is in English
- Someone sends an email with a 5-page PDF attachment that exceeds the AI's context window
Each edge case teaches you something. Fix the most common failures, then move to supervised running.
Supervised mode: For the first week, let the automation do everything EXCEPT send messages or take irreversible actions. Make calls this "disable scenario"; n8n calls it "manual mode." Review the output each morning. When you trust it — usually after 20-30 successful runs — turn it on fully.
This is exactly how we structure the AI Kickstart session: build, test with your real data, supervise, then launch.
Step 4: Go live safely
Before turning off supervised mode, do three things:
1. Set up error alerts. Both Make and n8n can notify you when an automation fails. Route alerts to Slack, email, or Telegram. A silent failure is worse than no automation — you think it is working, but it is not.
2. Add a human approval step for anything irreversible. If your automation sends emails to clients, processes payments, or modifies your CRM, add a step where it pauses and waits for you to click "approve." Most platforms have a built-in "wait for approval" node. Use it.
3. Document the automation. A simple note with: what it does, what triggers it, what to do if it breaks, and how to turn it off. Your future self — or the colleague covering for you — will thank you.
Common first-automation mistakes
| Mistake | What happens | How to fix it |
|---|---|---|
| Automating a task you do once a month | Setup time exceeds lifetime savings | Use the 3x/week rule: if you do it less than 3x/week, do not automate it |
| Building the whole thing before testing | 2 hours in, you discover step 1 is broken | Build one step, test it, then build the next |
| Giving the AI vague instructions | AI returns generic, unusable output | Be specific: "Categorise this email as URGENT, QUERY, INVOICE, or GENERAL" not "Sort my emails" |
| No error alerts | Automation fails silently for 3 weeks before you notice | Turn on notifications before going live |
| Skipping supervised mode | Automation sends a badly-worded draft to a real client | Run supervised for one week. Review output daily. |
| Choosing a task with fuzzy rules | "Follow up with interesting leads" — the AI cannot guess what "interesting" means | Pick tasks with clear, written rules |
When your automation is ready to become an agent
A simple automation follows a fixed path: trigger → step 1 → step 2 → step 3 → done. It works brilliantly for tasks with predictable inputs and fixed outcomes.
An AI agent is the next step. It loops: trigger → AI processes → AI decides → AI acts → checks result → decides again if needed. Agents handle tasks where the right action depends on context.
When to upgrade from automation to agent: - Your automation keeps hitting edge cases that need judgement - You find yourself adding more and more "if this, then that" branches - The task involves a conversation (with a customer, a lead, a team member) - You want it to proactively check for things rather than wait for a trigger
Our Build Your First AI Agent workshop is the natural next step — one day, zero to deployed agent. Businesses ready for advanced systems move to AI Mastery, covering multi-agent orchestration over 8 weeks.
Frequently asked questions
Do I need to know how to code?
No. Both Make and n8n are visual, drag-and-drop platforms. If you can draw a flowchart on a whiteboard, you can build an automation. The AI step — where the tool makes decisions — is driven by plain-English instructions, not code.
How much does it actually cost?
Your first automation costs £0-7/month. Make's free tier (1,000 operations/month) is enough for most single automations. n8n is free forever if self-hosted. You only pay when you scale up or add paid AI API access (typically £5-15/month for a generous allowance).
How long does it take to build?
Your first automation: 60-120 minutes. Your second: 20-30 minutes. The learning is front-loaded. Once you understand triggers, steps, and routing, the pattern repeats across every tool and every use case.
What if the AI gets something wrong?
It will. That is normal and expected. The goal is 80% accuracy on version one. The other 20% of cases get handled manually while you refine the automation. AI accuracy improves over time as you add rules, examples, and constraints to your prompts. Our Prompt Engineering for Teams workshop focuses specifically on writing instructions that minimise errors.
Can I automate something that involves multiple languages?
Yes. Modern AI assistants handle multilingual input natively. If your business operates in English and Portuguese — common for UK companies serving Portuguese markets — include a language instruction in your AI prompt: "If the email is in Portuguese, draft the reply in Portuguese. If it is in English, reply in English."
What if I get stuck?
Build the basic structure first — trigger and one simple action. Get that working. Then add AI. Then add branching logic. Three small wins beat one big failure. If you want guided help, our consulting services range from a 90-minute AI Kickstart to a full Business Optimisation audit that maps your entire operation for automation opportunities.
Start today
The hardest part of automation is not the technology. It is choosing the first task and starting.
Open your email. Look at the last 20 messages. How many followed the same pattern — categorise, draft reply, file away? That is your first automation. It will take 90 minutes to build and it will save you more time than anything else you do this week.
If you want to build it with someone guiding you — rather than alone — join AI Kickstart. 90 minutes. Your first automation, live by the end of the session.
Last updated: 7 July 2026