How to get started with AI in your business

Artificial intelligence has gone from something only large companies did to something every business gets pitched weekly. The problem is rarely the technology. The problem is that projects start in the wrong place: with a tool someone heard about, instead of with a task that actually costs you time.

Here is the approach we use when helping Norwegian small businesses, clubs and associations put AI to work.

Write down what you did last week that you also did the week before. Did you answer the same questions by email? Write the same kind of quote? Copy numbers from one system into another?

The gain from AI is almost always in the repetitive work. A task that takes twenty minutes and happens five times a week is worth automating. A task that takes two days but happens once a year usually is not.

Customer enquiries: an AI chat that answers from your own prices, opening hours and routines handles the questions you answer every day, and passes the rest to a human.

Content: product copy, news and social posts are time-consuming to write, and easy to draft with AI, as long as someone reads it before it goes out.

Moving information: orders that need to reach accounting, forms that should become a task, email that should become a follow-up. This is often integration with a thin layer of AI on top, not the other way around.

A pilot is one limited task, put into production, used by real people. It costs considerably less than a full project, and it answers the only question that matters: does this actually save us time?

If the answer is yes, you expand. If it is no, you saved the money that would have gone into phase two.

Which data leaves the building, and who receives it? Can sensitive information be kept out? And what does the agreement say in writing about data processing?

This is not legal trivia. It decides whether the solution can be used on real customer data, or only on what is already public.

A language model guesses when it does not know. So we limit what it answers, let it show the source of its answer, and log what is said. For anything with financial or legal consequences, we keep a human in the loop.

An AI that says «I do not know, here is our phone number» is worth far more than one that confidently answers wrong.

Sometimes the answer is an ordinary form, a better email template or a simple integration with no AI at all. That is cheaper, more predictable and easier to maintain.

A supplier who never recommends that is selling a product, not a solution.

Find the repetitive task. Build a small pilot. Settle the data questions. Plan for wrong answers. Expand only once the pilot is actually being used.

Want a walkthrough of where AI could give time back in your business? That conversation is free and without obligation.