Most advice about AI automation is written for companies with an IT department. If you run a bakery, a design studio, a consultancy or a three-person online shop, you do not need an “AI transformation”. You need five or six hours of your week back. This guide is a realistic 30-day plan to automate a small business with AI: one focused week at a time, with specific tools, honest time estimates, and a clear list of things you should never hand over to a machine.
A word on expectations before we start. Automation done well is quiet. Customers should not notice it. What they should notice is that you reply faster, invoice on time, and publish consistently. That is the goal — not a chatbot with your logo on it.
Before day one: the 90-minute audit
Do not open a single AI tool yet. Instead, take a notebook and track your working week with three columns:
- Repeated tasks — anything you do more than three times a week (answering the same email, formatting invoices, resizing images, posting to social media).
- Time per task — an honest estimate, including the mental cost of switching to it.
- Judgment level — score each task 1 to 3. A “1” is pure mechanics (renaming files). A “3” requires taste, empathy or legal awareness (pricing a proposal, handling a complaint).
The rule for the entire month: automate the 1s, assist the 2s, keep the 3s human. Most small businesses discover that 40 to 60 percent of their admin time sits in category 1 and 2. That is your target — not your craft, and not your relationships.
The stack: four tools, not fourteen
The fastest way to fail at automation is to sign up for eleven tools in a weekend. Everything in this plan runs on four pieces, and three of them start free:
- One AI assistant (Claude or ChatGPT, around €20/month for the paid tier). This is your drafting engine for email, content and documents. Pick one and learn it properly rather than hopping between three.
- One connector platform (Zapier or Make). This is the plumbing that moves data between your form, your inbox, your task list and your invoicing tool without you touching it.
- One scheduler (Buffer, Publer, or the native scheduling inside the platforms you use). Publishing on time is a software problem, not a willpower problem.
- The tools you already have. Gmail templates, Notion, your accounting software’s built-in AI features. Before adding anything new, check what is already hiding in the tools you pay for — most owners are surprised.
Resist everything else until the 30 days are over. Every additional tool is another login, another monthly fee, and another thing that silently breaks in August.
Week 1 — Communication: your inbox stops running your day
Email and messages are where most owners bleed time, so we start here.
Days 1–3: build your reply library
Pull up your sent folder and find the ten emails you write most often: quote requests, “are you available”, delivery questions, gentle payment reminders, “thanks but no”. Paste each into an AI assistant (Claude and ChatGPT both work well here) with a prompt like:
“Rewrite this as a reusable template in my voice: warm, direct, no corporate filler. Mark the parts I need to personalise in [brackets].”
Store the results wherever you actually work — Gmail templates, a Notion page, or a text-expander such as Espanso or TextBlaze. This alone typically saves two to three hours a week, and it is the least risky automation you will ever ship.
Days 4–7: draft, never send
Next, set up AI-assisted drafting for the non-standard emails. The workflow that works: read the incoming message yourself, ask the AI for a draft (“reply declining politely, offer the March slot instead”), then edit and send. The important discipline is in the second word — draft. Auto-sending replies to real customers is where small businesses damage relationships. You are the sender; the AI is the typist.
By the end of week one you should be answering routine email in a third of the time, with zero change in how your customers experience you.
Week 2 — Content and marketing: one hour in, a week of output
Consistency beats brilliance in small-business marketing, and consistency is exactly what AI is good at.
The batch session
Block ninety minutes. Bring three raw ingredients: a customer question you answered this week, a result you are proud of, and an opinion you hold about your industry. Ask the AI to turn each into platform-shaped drafts — a short LinkedIn post, an Instagram caption, a newsletter paragraph. Then do the part only you can do: cut anything you would not say out loud, and add one concrete detail from your actual work. Generic AI content is easy to spot; specific detail is what rescues it.
The scheduling layer
Pipe the finished posts into a scheduler (Buffer and Publer both have usable free tiers) so publishing happens without you. The pattern to aim for: AI drafts, you edit, software publishes. Each layer does what it is best at.
Product and service descriptions
If you sell online, this is the week to refresh listings. Feed the AI your existing description plus three real customer phrases (“finally an invoice template that doesn’t look like a spreadsheet”) and ask it to rewrite for clarity, not hype. Keep your price and promise decisions human — that is category-3 judgment.
Week 3 — Operations: the invisible machinery
This is the least glamorous week and the one with the highest permanent payoff.
Connect your tools
Pick one automation platform — Zapier if you want the largest catalogue, Make if you want more power per euro — and build exactly three automations:
- New enquiry → your task list. Website form submissions land in Notion, Trello or your CRM with a due date attached. Nothing falls through.
- Invoice paid → thank-you email queued. Draft goes to you for one-click approval, or sends automatically once you trust it.
- New booking → calendar + confirmation. Tools like Cal.com or Calendly handle the back-and-forth of scheduling entirely; this is one of the rare cases where full automation is safe from day one.
Documents and admin
Use AI to draft the boring documents you keep postponing: a proposal skeleton, a client onboarding checklist, a returns policy. Draft with AI, verify every factual and legal claim yourself, and — for anything with legal weight — have a professional read it once. An hour of a lawyer’s time on an AI-drafted contract is dramatically cheaper than a lawyer drafting from zero, and dramatically safer than shipping the draft raw.
Meeting notes
If your week involves calls, an AI notetaker (Fathom, Otter, or the recorder built into your video tool) turns every meeting into a summary with action items. Tell participants you are recording — it is both polite and, in many jurisdictions, legally required.
Week 4 — Numbers, review, and the keep/kill decision
Days 22–25: light-touch finance
Automate the mechanical edges of money, never the judgment. Good candidates: receipt capture and categorisation (Dext, or the AI features in QuickBooks and Xero), payment reminder sequences with a human-approved tone, and a monthly “ask the AI to explain my numbers in plain English” session using your accounting export. Bad candidates: pricing, tax positions, and anything you would need to defend to an auditor. Software categorises; you decide.
Days 26–30: the review
Return to your week-zero audit and measure honestly. For each automation ask three questions: Does it actually save time once maintenance is counted? Has quality dropped anywhere? Has a customer noticed — in a bad way? Kill anything that fails. A small business running four reliable automations is in far better shape than one running fifteen fragile ones.
What you should not automate
This list is the difference between a business that uses AI and a business that got hollowed out by it.
- Complaints and apologies. An upset customer can tell when a machine answers. This is the moment your business is actually judged — show up personally.
- Pricing and negotiation. AI does not know your costs, your capacity this month, or what this client relationship is worth over five years.
- Final legal, tax and contractual wording. Draft with AI, decide with professionals.
- Hiring and people decisions. Screening summaries, fine. Judgments about human beings, never.
- Your voice on important days. A milestone, a public mistake, a loyal client’s tenth order — write those yourself.
A worked example: one consultancy, one month
To make this concrete, here is how the plan looks for a two-person marketing consultancy — a composite of the small service businesses this works best for.
Week zero surfaced the usual suspects: forty minutes a day on email, three hours a week writing proposals from scratch, social media posted “when there’s time” (meaning rarely), and invoices sent late because invoicing is nobody’s favourite job.
Week one, they built twelve reply templates from their sent folder — enquiry responses, scope clarifications, polite declines. Email time dropped to fifteen minutes a day almost immediately.
Week two, one ninety-minute batch session now produces two weeks of LinkedIn posts, each anchored on a real client situation (anonymised). Engagement went up, not down — the posts got more specific, because the editing pass focused entirely on adding real detail.
Week three delivered the big structural win: enquiry form → Notion pipeline with a 24-hour follow-up reminder, and a proposal skeleton the AI pre-fills from their discovery-call notes. Proposal drafting went from three hours to forty-five minutes — with the pricing section still done by hand, every time.
Week four, they automated invoice reminders (day 7 gentle, day 14 direct, both human-approved before sending) and killed one automation from week three that kept misfiling enquiries — which is exactly what the review is for. Net result: about seven hours a week recovered, and the first month in a year with zero overdue invoices.
None of this is spectacular, and that is the point. Spectacular is fragile; boring is compounding.
The realistic scoreboard
Run the plan honestly and a typical solo or small-team business ends the month with: five to eight hours a week recovered, an inbox that is processed instead of feared, marketing that publishes on schedule, enquiries that never slip, and a clean monthly view of the numbers. What it will not do is run your business for you — and after thirty days of this, you will understand why that is the good news. The hours you get back are for the category-3 work: the craft, the relationships, the decisions that made it your business in the first place.
Frequently asked questions
How much does it cost to automate a small business with AI?
A working stack costs roughly €20–60 per month: one AI assistant subscription (around €20), an automation platform’s starter tier, and a scheduler’s free plan. Start on free tiers and pay only where you feel friction — most businesses need far fewer paid seats than they expect.
Do I need technical skills to follow this plan?
No. Everything in this plan is configured in plain English through visual interfaces. If you can set up an email filter, you can build a Zapier automation. The scarce skill is not technical — it is knowing your own processes well enough to describe them, which is exactly what the week-zero audit gives you.
Which single automation should I build first?
The reply library from week one. It costs nothing, carries no risk, saves hours immediately, and teaches you the core habit of working with AI: give it context, demand your own voice, and keep the final read for yourself.
Go further
If you want the full system rather than the summary — the audit worksheets, the exact prompts for every week, and the automation recipes ready to copy — The Automation Playbook packages this 30-day plan into a step-by-step working document. You can also browse our other guides and playbooks, or see everything we make for small teams in the shop.