AI for small businesses: where it saves time, and where it doesn’t (yet)

AI is worth considering for jobs that come up again and again, follow a pattern, and depend on information you already have in writing.

• Answering customer questionsOpening hours, delivery, what a product does, which option suits whom. An assistant can answer from your catalogue, prices and guides at any hour, in Serbian or English.
• Product advice and ordersIt can recommend from your product range and turn the conversation into an order, so the customer doesn’t have to start again in a separate form.
• Sorting incoming requestsIt can read emails and form messages, sort them, and draft replies for your staff to check and send.
• PaperworkIt can pull the numbers out of invoices, delivery notes and forms, and enter them into your system.

What these jobs have in common: a person can check the result, and the information the AI needs already exists.

• When a form, a rule or a search box would doA booking with fixed time slots, a price list, a fixed set of options. A plain form is faster for the customer and can’t make up an answer.
• When every answer must be exactly right without anyone checkingAI can still get things wrong. Anything with legal, medical or financial consequences needs a rule or a person in the loop.
• When the knowledge isn’t written downAn assistant answers from your sources. If the answers live only in one person’s head, write them down first. That’s useful even if you never add AI.
• When the volume is smallIf only a handful of questions come in each week, the setup won’t pay for itself.

• It answers from your sourcesBefore it replies, it looks up your catalogue and documents, so its answers come from what you sell.
• It has one job and stays on itIt declines questions that have nothing to do with your business, and can limit how many questions one visitor asks in an hour.
• Rules come first for anything that costs moneyAn order is only created once the customer confirms it.
• It’s tested on real questionsFrom your business before launch, and its sources and rules are adjusted when it misses.
• Your data stays yoursAI providers’ business services don’t train their models on your data. When data can’t leave your premises at all, an open model can run on your own server.

For a specialist retailer in Serbia, the shop’s website opens with a Serbian-language assistant. It recommends products from the shop’s range and answers questions from the shop’s own buying guides.

When the customer is ready, it confirms the products, asks for an email address and shows the total with delivery. It creates the order only once the customer confirms, and the customer downloads a pro-forma invoice with the payment details. Staff edit products in the admin panel and add new guides by uploading a PDF. The assistant stays on topic and limits how many questions one visitor can ask in an hour.

Read the AI assistant story

There are three parts:

• Building itConnecting the assistant to your products, documents and systems, setting its rules, and testing it.
• Running itAn AI provider charges a usage fee that depends on how many conversations you have. A model on your own server costs a server instead.
• Keeping it currentWhen your products, prices or policies change, its sources have to change too. Staff can do much of this themselves through an admin panel.

1. Pick one repeated jobSuch as the questions your staff answer every day.
2. Collect real examplesThe questions customers actually ask, in their words.
3. Check the answers exist in writingIn your catalogue, guides or policies.
4. Decide where a person stays involvedEspecially for anything that costs money.
5. Test before launchOn those real questions.

Our Digital Audit looks at your day-to-day work and tells you where AI would help, where it wouldn’t, and roughly what each step would cost.

Book a Digital AuditBook a 30-minute call