What is Machine Learning (ML)?

Machine learning is a branch of artificial intelligence in which a system derives its rules from data rather than being given them explicitly. Instead of a developer writing the conditions that identify a fraudulent transaction, a model is fitted to many examples of transactions already labelled fraudulent or not, and infers the pattern.

The practical consequence is that machine learning is only as good as the examples it is shown. A model reproduces the regularities in its training data, including the ones nobody intended it to learn.

Key Takeaways

Rules are inferred from examples rather than written by hand.
Model quality is bounded by the quality and representativeness of the training data.
A model that performs well in testing can still fail in production when the world shifts under it.
Most commercial value comes from unglamorous prediction: ranking, scoring, forecasting and classification.

Understanding Machine learning

A typical workflow separates data into what the model learns from and what it is judged on. The model is fitted to the first and evaluated on the second, because a model scored on the data it memorised will always look better than it is. This split is the discipline that separates a working model from a demonstration.

Performance degrades over time as behaviour changes, a drift that is invisible unless it is monitored. A churn model trained on one year’s customers will quietly lose accuracy as pricing, competitors and the customer base change. Retraining is therefore an operational commitment, not a one-off project cost.

For most businesses the constraint is not algorithms, which are largely commodity, but data: whether the outcome you want to predict has been recorded consistently, and enough times, to learn from.

Real-World Example

A subscription business wants to intervene before customers cancel. Rather than defining at-risk by hand, it fits a model to two years of accounts labelled by whether they churned, using behaviour in the preceding ninety days. The model scores current accounts weekly, and the retention team works the highest scores. The value is not the model’s accuracy in isolation but whether the intervention changes the outcome, which needs a holdout group to establish.

Importance in Business or Economics

Machine learning makes it economic to make a decision many times over where a human judgement would be too slow or too expensive: which result to rank first, which transaction to review, which customer to call. Its business significance is usually in the volume of small decisions rather than the sophistication of any one of them.

Types or Variations

Supervised learning: Learns from labelled examples to predict a known outcome, such as churn or fraud.
Unsupervised learning: Finds structure in unlabelled data, such as clustering customers into segments.
Reinforcement learning: Learns a policy by acting and receiving reward, used in control and sequential decisions.
Deep learning: Uses multi-layered neural networks, dominant for language, image and audio tasks.

Quick Reference

Field: Subfield of artificial intelligence
Learns from: Examples, not hand-written rules
Key discipline: Evaluating on data the model has not seen
Ongoing cost: Monitoring and retraining as behaviour drifts

Frequently Asked Questions

What is the difference between machine learning and artificial intelligence?

Artificial intelligence is the broad goal of systems performing tasks that would require intelligence. Machine learning is one approach to it, in which the system learns from data. All current commercial machine learning is AI; not all AI has historically been machine learning.

How much data does machine learning need?

It depends far more on the difficulty of the pattern than on a fixed threshold. A clean, strong signal can be learned from thousands of examples; a subtle one may not be learnable from millions. The more useful question is whether the outcome has been recorded consistently.

Why does a model get worse over time?

Because the relationship it learned no longer holds. Prices change, competitors enter, customer mix shifts. This is called drift, and it is only detectable if model performance is monitored against actual outcomes after deployment.

Let’s build the brand your work already earned.

Machine Learning (ML).