Machine learning

What it means

Machine learning is a field of artificial intelligence in which a system learns from data to find patterns and make predictions without someone writing all the rules. Instead of hand-coding conditions, the model receives many examples and works out by itself how to respond to them.

There is supervised learning, where the model gets examples with correct answers, unsupervised learning, where it finds groups and anomalies on its own, and reinforcement learning, where it learns by trial and error. Large language models are also built on machine learning.

The quality of results depends above all on the data. A model learns whatever it finds in the data, including errors and biases. Data quality and representativeness therefore matter more in machine learning than the choice of algorithm itself.

How it works in practice

In accounting and finance, machine learning is used for example to recognise and categorise documents automatically together with OCR, to match payments with invoices, to forecast cash flow or to flag unusual transactions that may signal an error or fraud.

A company does not need to build its own models. Most accounting, banking and CRM tools already have machine learning built in. When choosing one, it is worth asking what data the model learns from, whether it improves from user corrections and where the data of the company and its clients is processed.

In larger projects, such as forecasting sales from big data, preparation is key: combining sources, cleaning the data and clearly defining what the model should predict. Results are always validated on data the model did not see in training to reveal its real accuracy.

What to watch out for

A model can produce convincing but wrong results, and it is not always clear why it decided as it did. For important decisions, such as on loans or employees, a person who reviews the result should have the final say.

The use of AI systems is also governed by the AI Act, which imposes strict duties on certain high-risk uses. If a model works with personal data, GDPR must be respected as well, especially for automated decision-making about people.