Machine Learning

Machine learning is a branch of artificial intelligence in which systems learn to recognize patterns and make predictions from data, improving their accuracy over time without being explicitly programmed for each outcome. Rather than a developer writing rules like “if the email contains these words, mark it as spam,” a machine learning system trains on thousands of examples of spam and non-spam, learning the patterns that distinguish them on its own.

How does machine learning work?

Machine learning works by exposing a system to large quantities of labelled examples, allowing it to extract statistical patterns that it then applies to new, unseen inputs. The training process adjusts the system’s internal parameters until its predictions match the known correct answers closely enough to be useful.

The three main types of machine learning:

1. Supervised learning: the system trains on labelled data (examples with correct answers). Used for classification (is this email spam?) and prediction (what is this customer’s likelihood of churning?).
2. Unsupervised learning: the system finds structure in unlabelled data. Used for clustering customers by behaviour or detecting anomalies in financial transactions.
3. Reinforcement learning: the system learns by trial and error, receiving rewards for correct actions. Used in robotics, game-playing AI, and increasingly in AI agent training.

According to McKinsey’s 2025 State of AI report, 72% of organizations that reported successful AI adoption were using machine learning for at least one core business process, most commonly customer service, marketing personalization, and supply chain forecasting.

How does machine learning relate to the AI tools businesses use?

Most AI tools that Canadian businesses use daily are powered by machine learning, even when that is not apparent from the interface. Email spam filters apply machine learning to classify messages. CRM lead scoring uses machine learning to rank prospects by conversion likelihood. Document classification in tools like HubSpot and n8n uses machine learning to route records to the right team.

Large language models (LLMs) like Claude and GPT-4 — the technology behind AI assistants — are themselves machine learning systems, trained on billions of examples of human text. The difference between an LLM and a traditional machine learning model is scale and architecture, not the fundamental training approach.

What is the difference between machine learning and rule-based automation?

Rule-based automation follows logic a developer wrote explicitly. Machine learning automation derives its logic from patterns in data, allowing it to handle variation that rules cannot anticipate.

A rule-based system that routes support tickets by keyword will fail when customers describe the same problem in unexpected ways. A machine learning system trained on historical ticket classifications handles novel phrasings by applying the patterns it has learned, without a developer needing to write a new rule for each variation. For business processes with high input variability, machine learning outperforms rule-based approaches significantly.