AI glossary for businesses
Machine learning
In short
Machine learning is a branch of artificial intelligence in which computers learn patterns from past data and use them to make predictions or decisions, without being programmed with explicit rules for each case. It powers sales forecasts, spam filters, product recommendations and fraud detection, and it is also the foundation on which today's generative AI is built.
How machine learning works
Think of an experienced shop manager. After years on the floor, they know that sunscreen sells out in the first heatwave of June and that umbrellas move on the first rainy Monday of autumn. Nobody wrote those rules down. They learned them from seeing what happened again and again. Machine learning does the same with data.
The most common type is supervised learning. You give the system many past examples together with the right answer, for example two years of daily sales along with weather, holidays and promotions. The algorithm finds the relationships and builds a model. You then give it new inputs, such as next week's forecast weather and planned promotions, and it predicts the likely sales.
Other types group similar customers without labels (clustering) or learn by trial and error. Generative AI is a newer branch of machine learning, trained on huge amounts of text and images.
In practice for a small business
A retailer with three stores and an e-shop often runs out of best-sellers in summer and is left with excess stock in autumn. A machine learning forecast, built on past sales, seasonality and promotion dates, can suggest weekly order quantities per product. The buyer still makes the final call. See AI for demand and stock forecasting for the steps.
Many SMEs already use machine learning without noticing: spam filters in email, product recommendations on Shopify apps, fraud checks in card payments and smart bidding in ad platforms.
Why it matters and what to watch out for
Data comes first: a useful forecast usually needs a reasonable history of clean, consistent data. Gaps, duplicate products and changing codes hurt accuracy more than the choice of algorithm.
The past is not the future: models trained on normal years can fail when conditions change sharply. Review forecasts against reality every month.
Bias: a model learns any unfairness in its data. This matters most when decisions affect people.
Regulation: under the EU AI Act, some uses, such as screening job candidates, are high-risk, with obligations phasing in through 2026–2027. If personal data is processed at scale, a DPIA under GDPR may be needed. Check with your DPO or legal advisor.