AI for small businesses: what actually works

What can AI do reliably for a small business today?

Four tasks work well in practice. First, drafting: writing a reply to a customer email based on the file. Second, summarizing: reducing a long email thread or phone note to its core. Third, sorting: recognizing incoming messages as a quote request, a malfunction or an invoice question, and routing them to the right place. Fourth, answering questions from your own documents: searching price lists, manuals and terms and pulling an answer from them.

What these tasks have in common: the information already exists. AI does not have to invent anything, only find, organize or rephrase. As long as that is the job, the error rate stays low and the result is easy to check.

What can AI not do well yet?

AI sometimes invents facts that are not in your data. Ask something with no source behind it and the answer can sound convincing yet be wrong. Judgment calls that depend on context nobody wrote down are also hard. You know whether a customer deserves an exception or a job should get priority; the system does not.

Letting AI take irreversible actions on its own is therefore unwise: sending emails without review, moving appointments without consultation, promising prices. And AI carries no responsibility. When something goes wrong, a person must be able to explain what happened and why. So count on AI to prepare work, not to finish it without oversight.

Why is "AI proposes, a person decides" the safe setup?

In this setup, AI does the groundwork and a person presses the button. The AI prepares a draft reply, suggests a task or recommends a category. An employee reads it, adjusts it or approves it.

This has three advantages. Mistakes are caught before the customer sees anything. The employee still knows the customer, because they still read every file. And the business stays accountable: every outgoing message has been confirmed by a person.

It costs a few seconds per message. That is the price of control, and for most small businesses that trade is worth it. Fully automatic replies only fit simple, frequent questions with a clear source.

What works in small business customer service?

Start with the incoming stream. AI can sort every message by topic and urgency, find the matching customer and prepare a draft in your company's tone. The employee reviews and sends.

For recurring questions, such as opening hours, delivery times or the status of an appointment, AI can answer directly from your own documents, citing the source. Outside office hours, an AI assistant catches questions and hands difficult cases to a colleague for the next morning.

What does not work: letting AI loose on angry customers, complaints or sensitive files. There you want a person who can read the situation. A good setup recognizes those cases and passes them on.

What can AI do for administration and planning?

A large part of administrative work is retyping and transferring. AI is good at that. Think of copying details from an email into a file, turning a phone note into a task or summarizing a work order for the invoice.

In planning, AI can propose options: which slots fit the calendar, the travel time and the type of job. The planner chooses. The same rule applies here: AI prepares, a person confirms.

What you should not expect: AI running your bookkeeping on its own or making payments. For numbers with financial consequences you want fixed rules and human checks, not a language model making an estimate.

Why do your own data and a knowledge base determine quality?

An AI model knows nothing about your business. It does not know your prices, your way of working or your agreements. Anything it says about those without a source is a guess.

A knowledge base solves that. You collect documents, price lists, manuals and frequently asked questions in one place, and the AI searches there before it answers. The answer then comes from your own material, with a reference to the source. That makes checking simple: if the source is right, the answer usually is too.

The quality of your knowledge base therefore determines the quality of the AI. Outdated documents give outdated answers. Set aside time to maintain the knowledge base, just like your website.

What should you check when choosing AI software?

Four points. Data location: ask where your data is stored and where the AI model runs. For customer data you want storage within the EU.

GDPR: there must be a data processing agreement, and it must be clear whether your data is used to train models. By default, it should not be.

Logging: you want to see afterwards what the AI did, when and based on what. Without a log you cannot trace mistakes or explain them to a customer.

Costs: much of AI is priced per use. Ask how that is capped, so a busy month or a mistake does not produce an unexpected bill. Check every vendor against these four points before you sign.

How does dot apply this in stipt?

dot is the AI colleague in stipt and follows the setup in this guide. dot proposes, you confirm: draft replies, tasks and changes to a file are prepared and only carried out after your approval. No email ever goes out automatically.

Answers from the knowledge base come with clickable source references, so you can check where something comes from. Tools that do work on their own, such as sorting incoming email or the WhatsApp bot, have a per-company on/off switch and are off by default. You decide what dot is allowed to do.

Every AI run is recorded in the dot log: which tool ran, who started it and on which file. Usage is counted in credits with a monthly allowance, so costs stay capped. And your data is stored in the EU. That way you get the groundwork from AI while the decisions stay with you.

Frequently asked questions