AI for small businesses: where to start without overspending or putting your customers at risk

If you run a small business, you have probably been sold AI in two opposite ways: as the answer to everything, or as something for big companies. Neither helps. What is useful is more modest and more concrete: identify three or four tasks you repeat every week, test whether AI saves you time on them and keep control of what reaches your customers.

Where it usually helps most

Not every task gains the same. These are the ones that give back the most time with the least risk:

• Everyday writing. Replies to emails, WhatsApp messages in a professional tone, product descriptions, social media posts, website copy. You provide the content; AI organises and polishes it.
• Customer service with review. Draft replies to frequent questions, complaints or negative reviews, which you check before sending. There is a template for delicate cases in Difficult emails.
• Spreadsheets. Formulas you cannot remember, cleaning up customer or stock lists, sales summaries by month or product.
• Organising information. Turning call notes into a task list, preparing a meeting agenda, summarising a contract with a supplier.
• Ideas and comparing options. Names for a promotion, questions for a survey, pros and cons of two suppliers (with the Decide with criteria template).

Where to be careful

• Prices, deadlines and conditions. If AI drafts a reply that promises a discount or a delivery date, that is a commitment from your business. Always check the figures.
• Legal, tax or health information. Use it to understand a topic and prepare questions, not to decide. Rules change and depend on your country; the final word belongs to a professional or the official source.
• Chatbots that answer on their own. In 2024, a Canadian tribunal ordered Air Canada to answer for the incorrect information its virtual assistant gave a passenger. If you put a bot in front of customers, what it says is yours (we cover this in Why AI makes things up).

What it really costs

• Free version. The main assistants have free plans with usage limits. They are enough to test whether a task works.
• Individual paid plan. Around 20 dollars a month for ChatGPT, Claude or Gemini. It is worth it when you already use the tool daily and run out of messages or need features such as uploading large files.
• Team and business plans. They cost more per person but include privacy commitments (for example, not training on your data by default) and central administration. They are the right option if several people handle customer information.
• AI included in what you already pay for. Many email, office, online shop and invoicing tools are adding AI features. Check what you have before signing up for something new.

The simplest way to know whether it pays off: measure how long a task takes you today, try it with AI for two weeks and compare.

Your customers' data is not yours to share

When you paste a customer's name, phone number or purchase history into a chatbot, you are processing personal data, and in most countries that is regulated. In Colombia, for example, Law 1581 of 2012 sets the rules for processing personal data and requires, among other things, the data subject's authorisation and adequate security measures.

In practice:

• Anonymise before you paste. Replace names with "the customer" and remove phone numbers, emails and ID numbers.
• Use business plans if you will regularly work with real customer data.
• Do not upload complete databases to personal tools to "analyse" them.

The full guide is in What never to paste into an AI chatbot.

A 30-day plan

Week 1: choose a task

List what you repeat each week and how long it takes. Choose a task that meets three conditions: it recurs often, it has no serious consequences if it goes wrong the first time and you can review the result quickly. Answering frequent questions or writing product descriptions are good candidates.

Week 2: create your base prompt

Write an instruction you can reuse, with context about your business, your tone and what it must never say. The structure is in How to write a prompt that works the first time. For example:

You are the person who answers messages for [BUSINESS NAME], a
[TYPE OF BUSINESS] in [CITY]. Our customers are [PROFILE].

Tone: friendly and professional, short sentences.
Never promise discounts, deadlines or refunds: if the customer asks
for them, write [CHECK] so I can complete it.

Information you can use:
- Opening hours: [HOURS]
- Payment methods: [PAYMENT METHODS]
- Shipping: [CONDITIONS]

Draft a reply to this message:
[CUSTOMER MESSAGE, WITHOUT PERSONAL DATA]

Week 3: use it and note the failures

Use the prompt every time the task comes up. Note what you correct: if you always change the same kind of thing, add that instruction to the prompt.

Week 4: decide

Compare how long it takes now with before. If you save time and the quality is good, keep using it and choose the next task. If not, drop it without guilt: not every task improves with AI.

Keep reading

• How to write a prompt that works the first time: the structure for building your templates.
• What never to paste into an AI chatbot: how to protect your customers' data.
• Your first AI automation without coding: the next step once you have mastered a task.
• How to check an AI answer before you use it: what to review before sending something to a customer.
• In the glossary: generative AI, workflow and hallucination.

Sources

• OECD (2024). Generative AI and the SME Workforce. OECD Publishing.
• Congress of the Republic of Colombia. Statutory Law 1581 of 2012, on general provisions for the protection of personal data.
• Moffatt v. Air Canada, 2024 BCCRT 149, Civil Resolution Tribunal of British Columbia. Coverage: Global News (February 2024).