Core concepts
What is Machine learning?
Machine learning (ML) is the field of building software that learns patterns from data instead of following hand-written rules, covering everything from demand forecasting and fraud detection to the neural networks behind modern language models.
Classic machine learning predates the current AI wave by decades and still quietly runs much of the business world: credit scoring, churn prediction, demand forecasting, recommendation engines, fraud flags. You give the algorithm historical examples (these transactions were fraud, these were not) and it learns the pattern well enough to score new cases. LLMs are one branch of machine learning, but far from the only commercially useful one.
The distinction matters when buying. Problems with structured, numeric history (how many units will this branch sell in December?) are usually better served by classic ML or even well-built statistics than by a language model. Problems involving language, documents, and conversation are LLM territory. A team that reaches for the fashionable tool on every problem is optimising for the pitch, not the outcome. Sometimes the answer is not ML at all: a linear-programming solver builds the nutritionally-valid base plans in our Compumeal product, and the LLM only refines and explains them. Right tool, right layer.
What ML genuinely needs is data: enough history, in usable shape, with the outcome you care about recorded. A surprising amount of "we want AI" engagements begin with three months of making the data exist. If a vendor never asks about your data, they are selling something generic.
Frequently asked
- What is the difference between machine learning and AI?
- AI is the broad goal of software that behaves intelligently; machine learning is the main way we get there today, by learning patterns from data instead of hand-written rules. LLMs are one branch of machine learning, but classic ML like fraud detection and demand forecasting quietly runs much of the business world too.
- Do I need machine learning or just an LLM?
- It depends on the problem. Structured, numeric questions like "how many units will this branch sell in December" are usually better served by classic ML or solid statistics. Language, documents, and conversation are LLM territory. A team that reaches for the fashionable tool on every problem is optimising for the pitch, not the outcome.
- What does a machine learning project actually need to start?
- Data: enough history, in usable shape, with the outcome you care about recorded. A surprising number of "we want AI" engagements begin with a few months of making that data exist. If a vendor never asks about your data, they are selling something generic.
Wondering where this fits in your business?
The free AI audit maps your operations and shows the three systems we would build first, in plain language like this.
