Definition
Machine learning (ML) is a branch of artificial intelligence in which software learns patterns from historical data instead of following rules written by a programmer. You show the system thousands of past examples, such as invoices marked paid or unpaid, and it builds a statistical model that predicts the outcome for new cases. The more relevant data it sees, the better it usually gets. Machine learning is the engine behind almost every practical AI system in use today.
In a company, machine learning typically lives inside products you already buy: credit scoring in your bank, demand forecasting in your ERP, product recommendations in your online shop, churn prediction in your CRM. Building your own model makes sense when you have a lot of proprietary data and a decision that repeats often, for example pricing, lead scoring or predictive maintenance. The work is mostly data preparation, not algorithms.
Today machine learning splits into two worlds. Classic ML on tabular data (spreadsheets, databases) remains cheap, fast and very effective for forecasting and scoring. Deep learning and large language models handle text, images and audio. A common misconception is that ML needs huge data. Many valuable business models train on a few thousand well-labelled rows. Another is that the model is the hard part. Data quality and deployment are.
In practice
An online retailer trains a machine learning model on 2 years of order history to predict which customers will not buy again in the next 90 days, then targets only those with a retention offer instead of discounting everyone.
Why it matters
Machine learning turns the data your company already collects into repeatable decisions. It is the cheapest way to add margin to processes that run thousands of times a month.
Frequently asked questions
- What is the difference between machine learning and deep learning?
- Deep learning is a subset of machine learning that uses neural networks with many layers. Classic machine learning works very well on structured data such as sales tables. Deep learning shines on unstructured data such as text, images and speech, and it powers today's large language models.
- How much data do I need for machine learning?
- Less than most people think for structured business problems: a few thousand clean, labelled examples can be enough for a useful forecasting or scoring model. Quality and relevance matter more than volume. For language or vision tasks, you usually reuse a pre-trained model and add a small amount of your own data.