AdAI
Definition

Algorithm is a step-by-step procedure for solving a problem. A recipe, in mathematical form. Some algorithms compute (the algorithm a spreadsheet uses to add a column). Some sort (the algorithm a search engine uses to rank results). Some learn (the algorithm a machine learning model uses to find patterns in your data). The word covers all of them.

Key Takeaways

  • An algorithm is a procedure. Inputs go in, the algorithm runs, outputs come out. Same idea whether the procedure is calculating tax or training a neural network.
  • In an AI context, 'algorithm' usually means a machine learning algorithm: a procedure for learning patterns from data. Decision trees, random forests, gradient boosting, neural networks, k-means clustering, all algorithms.
  • An algorithm is not the same as a model. The algorithm is the recipe. The model is the result of running the recipe on your data.
  • Most business software runs dozens of algorithms quietly in the background. Lead scoring, fraud detection, search ranking, recommendations. You do not usually need to know which one.
  • The same algorithm gives different results for different businesses because the data it learns from is different. The algorithm is the same. The model is yours.
88%
of organisations now use AI (and the algorithms behind it) in at least one business function, up from 78% the previous year
Source: Aon Global Risk Management Survey, 2025
1968
the year Donald Knuth published Volume 1 of The Art of Computer Programming, the foundational text on algorithms that still shapes computer science teaching nearly six decades later
Source: Knuth, The Art of Computer Programming, Volume 1, Addison-Wesley, 1968

In Simple Terms

An algorithm is what a computer does step by step to get from a question to an answer. Add up these numbers. Sort this list. Find the shortest route. Predict which customer will churn. Each of those is a different algorithm doing a different job, but the structure is the same: defined inputs, defined steps, defined outputs.

When people say "the Facebook algorithm" or "the Google algorithm," they usually mean a system of hundreds of algorithms working together, deciding what you see and in what order. Same word, much bigger thing. For business software, "algorithm" most often refers to a specific machine learning procedure: the recipe a model uses to learn from data.

The practical point for an SMB owner is that you almost never need to know which algorithm is running. What you need to know is what the tool does, what data it uses, and whether its outputs are good for your business. The algorithm is the engineer's concern. The result is yours.

The Algorithms You Already Rely On

A list of algorithms doing real work in business software you probably use.

PageRank and its descendants. The algorithm Google originally used to rank search results, by counting how many other pages linked to a page. Modern Google uses thousands of signals, but PageRank's "links as votes" idea is still in there. Every time you search, this family of algorithms runs.

Gradient boosting (XGBoost, LightGBM, CatBoost). The workhorse algorithm behind most modern lead scoring, churn prediction, fraud detection, and credit scoring. It wins most prediction contests on tabular data. HubSpot, Salesforce, and most CRM scoring tools use some variant.

Neural networks (transformers, CNNs). Power every modern image recognition, voice recognition, and language model. ChatGPT, Claude, Gemini, the speech-to-text in Otter, the image tagging in Google Photos: all neural-network-based algorithms.

Naive Bayes. An old algorithm (1960s) still doing real work in spam filtering. Cheap to run, surprisingly accurate, easy to update. Behind some of Gmail's earliest spam protection and still part of many systems today.

Collaborative filtering and matrix factorisation. The recommendation algorithms behind Netflix, Spotify, and Amazon's "customers who bought this also bought." Tells the system that two users with similar past behaviour will probably like similar things going forward.

Algorithm vs Model vs Tool

Algorithm

The procedure. The recipe. The mathematical steps. Random forest is an algorithm. So is gradient boosting. So is k-means clustering. Pure procedure, no data attached.

Model

The result of running the algorithm on data. Your churn-prediction model is what comes out when you run the gradient boosting algorithm on your historical customer data. Two businesses using the same algorithm produce different models.

Tool or product

The software wrapped around the model. HubSpot is a tool. Inside it is a model. Inside the model is an algorithm. As a buyer, you pay for the tool. The model and algorithm are how it works under the hood.

“Premature optimization is the root of all evil.”

Donald Knuth, Professor Emeritus, Stanford UniversityStructured Programming with go to Statements, ACM Computing Surveys, 1974

Frequently Asked Questions

What is the difference between an algorithm and a model?
An algorithm is the procedure. A model is the result of running that procedure on data. Random forest is an algorithm. The random forest trained on your customer data to predict churn is a model. Same relationship as a recipe (algorithm) and a baked cake (model).
What algorithms do business software products actually use?
A handful do most of the work. Decision trees and random forests power most lead scoring and tabular prediction. Gradient boosting (XGBoost, LightGBM) wins most prediction contests. Neural networks power image, voice, and language tasks. Naive Bayes still runs much of spam filtering. PageRank-style graph algorithms drive search and recommendations. You almost never need to know which one is running under the hood.
Is the 'Facebook algorithm' or 'TikTok algorithm' a single thing?
No. Both are systems of hundreds of algorithms working together: ranking models that decide what to show next, classifiers that detect violations, recommendation models that find similar content, prediction models that estimate watch time. When people say 'the algorithm changed,' they usually mean one of those components was reweighted or retrained.
Do I need to understand algorithms to use AI in my business?
No. The same way you do not need to understand internal combustion to drive a car. Modern AI tools abstract the algorithm behind a button. What matters for an SMB is knowing what each tool does, what data it touches, and how reliable it has been on your inputs, not which gradient is being optimised inside.
Can the same algorithm give different results for different businesses?
Yes, almost always. The algorithm is the same. The data it learns from is different. Two businesses using HubSpot's lead scoring algorithm will get different scores for the same lead, because each model is trained on their own historical conversion data. The output is shaped by the inputs.

Related Glossary Terms & Resources

Join 5,000+ SMB owners getting weekly AI agent insights

Subscribe Free