A machine learning algorithm is the method a computer uses to learn a pattern from data. If the finished model is the recipe, the algorithm is the cooking method that produced it. There are dozens, but they fall into a handful of families, and most business people never need to touch one directly.
What matters is knowing roughly what each family does, so that when a vendor waves the word "algorithm" at you across a table, you can ask a sensible question instead of nodding politely and hoping.
This is the third piece in a short series. The first explained what machine learning is, the second explained what a model is, and this one takes on the word people find most intimidating and least necessary to fear: the algorithm.
Machine learning: the key facts
- A machine learning algorithm is the method used to find a pattern; the model is the pattern it finds. Algorithm is the recipe, model is the cake.
- The four families you will actually hear about are decision trees, linear and logistic regression, clustering, and neural networks.
- Algorithms learn by predicting, checking how wrong they were, and adjusting — repeated across enormous numbers of examples.
- There is no single best algorithm: the right one depends on the job, the data and what you are predicting.
- You will never have to choose one. It has already been chosen and tuned by whoever built the tool you are using.
- What Is Machine Learning? A Plain-English Guide for Business Owners
- Machine Learning Models Explained: What They Are and How Businesses Use Them
- Machine Learning Algorithms Explained Simply (No Maths Required) — you are here
What is a machine learning algorithm, really?
An algorithm is just a method. A set of steps for getting something done. Your nan's method for a Victoria sponge is an algorithm. Long division is an algorithm. There is nothing sinister about the word, and the tech industry has spent a decade making it sound like witchcraft.
A machine learning algorithm is the particular method a computer uses to learn a pattern from examples. You point it at a pile of data, it follows its steps, and out the other end comes a trained model you can use. Run the algorithm once, use the model for years. That is the relationship, and it is the bit everyone muddles, so it is worth saying plainly.
Algorithm or model: which is which?
The algorithm is the method. The model is what the method produced. Cooking method versus finished dish. You will hear the two words used as if they mean the same thing, and in the pub that is fine, but if you want to sound like you know what you are talking about, that is the distinction that does it.
The main families of algorithm
You could spend a career on this. You do not need to. Here are the families you are actually likely to hear named, in language a human can use.
| Family | How it works, roughly | Good for |
|---|---|---|
| Decision trees | A flowchart of yes/no questions that narrows down to an answer. | Decisions you need to be able to explain and follow, like whether to approve something. |
| Linear & logistic regression | Draws the best line through the data to predict a number or a yes/no. | Forecasting sales, estimating risk, the everyday workhorses. |
| Clustering (e.g. k-means) | Groups similar things together without being told the groups first. | Finding customer segments or spotting the odd one out. |
| Neural networks | Layers of simple connected units, loosely inspired by the brain, that learn very complex patterns. | Language, images, speech. The family behind ChatGPT. |
That last row is where the headlines live. The large language models everyone is talking about are enormous neural networks, trained on a colossal amount of text. Powerful, and famously hard to explain: they can give you a brilliant answer and be quite unable to show their working, which is exactly why they suit drafting and summarising better than they suit unsupervised final decisions.
How does an algorithm actually learn?
Under all the families sits the same simple loop. It is almost embarrassingly like the way a person learns darts.
- Guess. The algorithm makes a prediction, badly at first.
- Check. It compares its guess to the right answer and measures how far off it was.
- Adjust. It nudges its internal settings to be a little less wrong next time.
- Repeat. Millions of times, until the guesses are good enough to be useful.
That is genuinely most of it. Not magic. A very fast, very patient loop of guess, check and adjust, run at a scale no human could sit through. Everything clever about modern AI is that loop, done astonishingly well, on astonishingly large amounts of data.
Do you need to choose one yourself?
No. Unless you are building AI systems from the ground up, which you are not, the algorithm has already been chosen and tuned by the people who built the tool. Choosing between a decision tree and a neural network is their problem. It is not on your desk and it never will be.
What is worth carrying into the room is a decent question or two, so you are not at the mercy of the jargon:-
- What is this actually predicting, and how often is it right?
- If it makes an important decision, can you explain why it decided that? (If the honest answer is no, keep a human on the final call.)
- What data was it trained on, and does that data look anything like mine?
There is no single best algorithm, and knowing that alone puts you ahead of most of the room. The right one depends on the job, the data and what you are trying to predict. Anyone selling you one algorithm as the answer to everything has misunderstood the question.
You do not need to know how the engine is built to drive the car well. You need to know what it is for, when to trust it, and when to keep your hands on the wheel. Get that right and the word "algorithm" stops being something that happens to you in meetings and becomes something you can steer.
Want your team using this properly?
We teach the whole picture, from what machine learning is to how to actually use it on your work, without a single equation.
Book A CallLearn this properly: the modules that cover it
Reading about machine learning is a good start. Using it on your own work is what changes your week. These are the training modules that take this from an idea to a habit, run in person at your premises or live online.
- AI Basics Deep DiveHalf-day team workshop on picking the right AI tool for every job, worked on your own real tasks.
- AI for Data AnalysisWhere machine learning meets your spreadsheets: analysing, summarising and forecasting without a data scientist.
- Basic AI TrainingThe beginner pillar: what a good AI course covers, the five-module syllabus and what you leave able to do.
- Introduction to Claude CodeWhere the algorithms stop being theory: building real things with an AI coding agent.
- AI for BusinessHow teams and SMEs learn AI: use cases, delivery options, outcomes and data safety.
- All Programmes and PricesEvery module, format and current price for individuals, teams and businesses.
Frequently asked questions
What is a machine learning algorithm in simple terms?
A machine learning algorithm is the method a computer uses to learn a pattern from data. If the finished model is the recipe, the algorithm is the cooking method that produced it. There are dozens of them, grouped into a few families, and they all share the same basic loop: make a guess, check how wrong it was, adjust, and repeat until the guesses are good.
What is the difference between an algorithm and a model?
The algorithm is the method. The model is the result. You run the algorithm during training and it produces a model, which is the trained pattern you then use over and over. People often use the words interchangeably, but the tidy version is method on one side, finished thing on the other.
What are the main types of machine learning algorithm?
The common families are decision trees (a flowchart of yes/no questions), linear and logistic regression (drawing the best line through data), clustering algorithms like k-means (grouping similar things), and neural networks (loosely inspired by the brain, and the family behind ChatGPT). Each suits different jobs, but the tools you use have already chosen for you.
Do I need to choose a machine learning algorithm myself?
Almost certainly not. Unless you are building AI systems from scratch, the algorithm has been chosen and tuned by whoever built the tool. What matters for a business user is knowing roughly what the families do, so you can ask a sensible question when a vendor starts throwing the word around.
Which machine learning algorithm is best?
There is no single best one, which is a genuinely useful thing to know. The right algorithm depends on the job, the data and what you are trying to predict. A decision tree is great when you need to see the reasoning; a neural network wins on messy things like language and images but is far harder to explain. Anyone claiming one algorithm is best for everything does not understand the question.
What is a neural network?
A neural network is a machine learning algorithm built from layers of simple connected units, loosely inspired by the brain. Each layer passes a slightly more abstract version of the data to the next, which is how the network learns patterns far too complicated to write down as rules. Language, image and speech tools all run on neural networks, and so does every large language model you have heard of.
Which machine learning algorithm should a beginner learn first?
None of them, honestly, unless you intend to build AI systems for a living. The skill that pays for a business user is knowing what the tools can do, how to prompt them properly and when to keep a human on the final call. If you are genuinely heading for a technical career, decision trees and linear regression are the traditional starting point because you can follow their reasoning by hand.
What is the difference between machine learning and deep learning algorithms?
Deep learning is the subset of machine learning that uses neural networks with many layers. Classic machine learning algorithms such as decision trees and regression work well on tidy, structured data like a spreadsheet. Deep learning is what you need for messy, unstructured things: language, images, audio. It needs far more data and far more computing power, which is why it arrived properly only in the last decade or so.
How do I know if an algorithm making decisions about my business is fair?
Ask three questions: what is it actually predicting and how often is it right, can anyone explain why it reached a particular decision, and what data was it trained on. If nobody can answer the second one, keep a human on the final call. That is the practical core of AI governance, and it is the reasoning we work through with teams in the AI Policy Workshop.
Where we teach machine learning and AI basics
The Oxford AI School is based at East End, Witney, Oxfordshire OX29 6PX, and delivers practical AI and machine learning training in person across Oxford, Oxfordshire, the Thames Valley, London and every major UK city, or live online for teams anywhere in the world. Sessions are hands-on, jargon-free and worked on your team’s own real tasks.
Harry runs The Oxford AI School, teaching teams and businesses the basics of AI in a fun, easy and practical way. He can be reached at Harry@TheOxfordAISchool.com.