Machine Learning (ML)

Machine learning is what you use when you can recognise the right answer but can't explain the rule for finding it.

Consider spotting a fraudulent transaction. You could try to write the rules — over £500, unusual country, odd hour. You'd catch some fraud and infuriate a lot of customers buying dinner on holiday. Instead, you show the software a hundred thousand transactions labelled "fraud" and "not fraud", and it works out the pattern itself — including subtle combinations no one would have thought to write down.

That's the whole idea. Traditional software: a human writes the rules, the computer applies them. Machine learning: a human supplies the examples, the computer infers the rules.

Machine learning vs AI — the bit that confuses everyone

Artificial intelligence is the broad field. Machine learning is the technique that made it work. Nearly everything marketed as "AI" today is machine learning underneath, including large language models — an LLM is a machine learning model trained on text.

So when a vendor says "powered by AI", the useful question is: what was it trained on, and how do you know it works? Those two questions cut through most sales decks.

Why it matters for your business

ML earns its keep on tasks with three properties: a lot of examples, a pattern too subtle to write down, and a tolerance for being occasionally wrong.

• Prediction — which leads will convert, which customers are about to churn, how much stock you'll need.
• Classification — sorting enquiries by intent, flagging documents for review, spotting defects.
• Personalisation — recommendations that reflect what someone actually does rather than what they told you once.

That third property is the one businesses underestimate. Machine learning is statistical — it gives you an answer with a confidence, not a guarantee. If your process cannot tolerate being wrong 5% of the time, you need a human in the loop, not a better model.

What it costs you

The expensive part is rarely the model. It's the data. ML needs examples that are plentiful, labelled, and representative of what you'll actually see in production — and most businesses discover their data is none of those three. Budget for the data work, or the model will learn your data's bad habits and reproduce them at scale.

How we use machine learning

For Arnold, computer vision models identify surgical plates from images — a task where an expert knows the answer instantly but cannot articulate the rule. Biosense applies ML to root cause analysis in biotech, surfacing patterns across scientific workflows that would take a researcher weeks to find by hand.

Both are the same shape: a specialist could do it, and couldn't explain how.