Machine Learning

What is Machine Learning?

Machine learning (ML) is the part of artificial intelligence that learns from examples. Instead of a developer writing every rule, the team gives the system past records with known outcomes, such as orders labeled as fraudulent or not, and an algorithm finds the patterns that separate them. The result is a model: a function that takes a new record and returns a prediction, a score or a category.

Training happens once or on a schedule; prediction happens every time new data arrives. A model trained on past customer data can score each new signup for churn risk the moment it lands in the CRM.

ML shows up across most business functions: lead scoring in sales, demand forecasting in operations, product recommendations in ecommerce, anomaly detection in finance and ticket routing in support. Most of these work on structured data from tables and databases, which is where ML has the longest track record in business.

What are the main types of machine learning?

• Supervised learning: the model learns from labeled examples, such as past deals marked won or lost, and predicts the label for new cases. Most business ML is supervised.
• Unsupervised learning: the model finds structure in unlabeled data, for example grouping customers into segments with similar behavior.
• Reinforcement learning: the model learns by trial and feedback, which suits pricing, bidding or recommendation systems that can test options over time.

Deep learning is a family of techniques based on neural networks that is used across all three. It powers image recognition, speech recognition and the large language models behind generative AI. For tabular business data, simpler methods such as decision trees and regression are often just as accurate and easier to explain.

How is machine learning different from generative AI?

Classic machine learning predicts or classifies. Given this customer, how likely are they to cancel? Given this transaction, is it fraud? The output is a number or a label, and it can be checked against what actually happened.

Generative AI produces new content: an email draft, a summary, an image. It is built with machine learning techniques but trained on very large amounts of text or images and used for a different job.

For many business problems, classic ML is the better fit. It is cheaper to run, easier to test and easier to explain to a finance or compliance team. Generative models shine when the input or output is language. Many real systems use both, for example a language model that reads support tickets and a classic model that predicts which ones will escalate.

What data does machine learning need?

A model is only as good as the history it learns from. It needs enough past examples of the outcome you want to predict, the inputs that plausibly drive that outcome, and a reliable record of what actually happened. For a churn model, that means customer records, usage or purchase history, and a clear definition of who churned and when.

The common blockers are rarely algorithms. The outcome was never recorded consistently, key fields live in a separate system, or definitions differ between teams. Data quality work, joining sources and agreeing on definitions usually take more time than training the model.

The training data must also match what the model will see in production. If a field is only filled in after a deal closes, it cannot be used to predict whether the deal will close.

How do you know if a machine learning model is working?

Before launch, test the model on historical data it has not seen and compare its predictions with real outcomes. Pick a measure that matches the business decision: for fraud, how many real cases it catches and how many good customers it wrongly flags; for forecasting, how far off its estimates are on average.

After launch, keep measuring. Models drift when customer behavior, pricing, channels or data pipelines change, and accuracy can drop without any error message. Compare predictions with outcomes on a schedule, watch for shifts in the input data, and retrain when performance slips.

Above all, measure the business effect, not only accuracy: whether the team acts on the scores, and whether those actions change results compared with the old way of working.

How does Vizio AI build machine learning models?

In our DX Studio we build forecasting and scoring models for questions a business already asks: which leads to call first, which customers are at risk, how much demand to plan for. We start with the data platform, because a model needs clean, joined history from CRM, finance and product tools, and we set up the integrations that keep it fed. Scores are pushed back into the tools people use or shown in executive and BI dashboards, so they inform real decisions. We support the models after launch as the data changes.

The starting point is usually a short review of the decision you want to improve and the data that exists to support it.