AI automation for small businesses: practical examples

The examples below are for businesses of a few to a few dozen people: a repair shop, a clinic, a store, a service company. For each one: what happens, what it's built with and where a person steps in.

What an AI automation is

An automation is a series of steps that starts on its own when something happens: an email arrives, a form is submitted, it's Monday morning. The steps move data from one place to another, send messages and create rows in spreadsheets.

Many automations don't need AI. "When a form comes in, add a row to the spreadsheet and send a confirmation email" is the same every time. AI (a model such as Claude or GPT) comes in at the step where text has to be understood: a free-form email, a scanned invoice, a question in the customer's own words.

Eight practical examples

1. Replies to questions that keep coming up

By email, website chat, text or WhatsApp, the same questions arrive every day: hours, prices, location, whether something is in stock. The automation passes the question to an AI model along with the business's information (price list, hours, delivery rules), and the model writes the reply.

Where a person steps in: questions the information doesn't clearly answer, complaints and quotes go to someone on the team, with the whole conversation attached.

The channel matters. In the US, texts sent from a regular 10-digit number through an application must be registered for A2P 10DLC; unregistered traffic is charged extra carrier fees and filtered more heavily. Outside the US, WhatsApp is more common: through the WhatsApp Business Platform, replies sent within 24 hours of the customer's last message are free, and you pay for messages the business starts, plus the provider's fee if there is one.

2. Orders and requests from email, entered in a spreadsheet

Customers send orders in emails that are each worded differently. The automation reads the email, the AI pulls out the data (product, quantity, address, requested date) and a new row appears in Google Sheets, Excel or your business software. The customer gets a confirmation.

Where a person steps in: when information is missing or the order is unusual, the row is flagged for review and no automatic confirmation goes out.

3. Data pulled from documents

Supplier invoices, contracts, delivery notes, photographed receipts: the AI reads the document and fills in the fields you need (supplier, amount, due date, document number).

Documents that already arrive as structured data, such as e-invoices in XML, can be read by a simple automation without AI. AI helps with scans and with documents that come in different layouts.

Where a person steps in: amounts and due dates are checked before payment.

4. Reports that arrive on their own

Every Monday morning, the automation gathers the week's numbers (sales, bookings, new leads) from the tools you use and emails them. The AI can add three or four lines explaining what changed since the previous week.

Where a person steps in: reading the report. Nobody spends an hour putting it together by hand any more.

5. New leads from a form, sent to the CRM

A customer fills in the form on your website. The details go to the CRM, the sales spreadsheet and the mailing list without anyone copying them. The AI can read the free-text message and label it: what kind of request it is, how urgent, who on the team should take it.

Where a person steps in: picks up the labelled request and replies.

6. A summary of a meeting or a call

The recording is transcribed and the AI writes a summary with the decisions and each person's tasks, emailed to the participants. Recording requires the consent of the people recorded, and some US states require everyone on the call to agree.

Where a person steps in: checks the summary before it goes to a client.

7. An assistant for the company's documents

Procedures, price lists, past quotes and answers given to customers sit in dozens of files. An internal assistant answers the team's questions from those documents and shows which document each answer came from.

Where a person steps in: the documents have to be kept up to date; the assistant only knows what they say.

8. Suggested replies to reviews

For each new review, the AI drafts a reply in the business's tone. Someone reads it, edits it if needed and publishes it.

Where a person steps in: approves every reply. A negative review never gets an automatic response.

What they're built with

Prices change often; check them on the vendors' pages before you budget.

What it costs in total

An automation has three costs:

1. The build, paid once: how many steps the flow has, how many tools it connects, how many special cases it has to handle.
2. The tools, monthly: from nothing to a few tens of dollars for a small business.
3. The AI model, by usage: depends on how many messages or documents go through it and how long they are.

A useful question before building: how many hours a month does the work you want to automate take today? If the answer is one hour, it probably isn't worth it. If it's dozens of hours, it almost certainly is.

Data and the rules

• Privacy. Customer messages and documents contain personal data. They should pass only through the tools that are needed, and those vendors should have data processing terms in place. Healthcare providers covered by HIPAA need a business associate agreement with any vendor that handles patient information.
• Model training. Anthropic (Claude) and OpenAI (GPT) state that data sent through their APIs isn't used to train their models by default. The consumer chatbot apps have different rules, so customer data doesn't belong there.
• Data on your own server. n8n can be installed on a company server, so data doesn't pass through extra services.
• Telling people it's a bot. California law makes it unlawful to use a bot to communicate with someone in California online with the intent to mislead them about its artificial identity in order to sell goods or services; disclosing that it's a bot avoids liability. In the EU, Article 50 of the AI Act has applied since 2 August 2026: people must be told they're interacting with an AI system at the latest at the first interaction. In practice, the bot says so in its first message.

How to start

1. For one week, write down what you do by hand that repeats: what, how often, how long it takes.
2. Pick the work that takes the most time and has clear rules. That's your first automation.
3. Write the steps exactly as a person does them today, exceptions included.
4. Test the automation on real examples, with a person checking every result.
5. Let it run on its own only once the results are right, keeping human review for the uncertain cases.

On our AI automation page you'll find how we work, from choosing the first flow to the adjustments after it goes live.