Predictive Analytics

Predictive analytics uses historical data and statistical or machine learning models to estimate what is likely to happen next: which customers will buy, which will stop buying, how much revenue a group will bring in, or how much traffic a page will get. In marketing it turns reports about the past into scores and forecasts you can act on before the outcome happens.

How it works

• Define the outcome you want to predict, such as “buys again within 90 days” or “monthly revenue”.
• Collect historical examples where you already know the outcome, with features that might explain it: recency, frequency and value of purchases, sessions, pages viewed, traffic source.
• Pick the task type. Yes or no outcomes are a classification problem; MLforSEO’s classification lesson describes logistic regression as great for predicting churn, email open rates or ad engagement. Numeric outcomes such as revenue are a regression problem.
• Train on the past, test on data the model has not seen, then score current customers or pages. This is supervised learning: the model learns from labelled history.

Clustering often comes first. As the Introduction to ML for SEO course puts it, most marketing data benefits from exploratory clustering before predictive modelling: customer segmentation describes who your customers are, and predictive models then estimate what each group will do.

How it’s used in SEO and marketing

• Predictive lead scoring: ranking leads by their likelihood to convert so sales follows up on the best ones first.
• Churn prediction: flagging customers likely to leave so retention emails or offers reach them in time.
• Google Analytics 4 predictive metrics: as of October 2026, GA4 offers purchase probability (a user active in the last 28 days logging a key event in the next 7 days), churn probability (a user active in the last 7 days not being active in the next 7) and predicted revenue over the next 28 days. You can build predictive audiences from them for ads and remarketing.
• Forecasting demand and traffic: estimating seasonal search demand or organic sessions to plan content and budgets.

Example

The customer segmentation lesson in MLforSEO’s Introduction to ML for SEO course uses RFM (recency, frequency, monetary) data from a public UK online retail dataset to group customers with clustering. A natural next step is predictive: label which customers bought again in the following quarter, train a classifier on their RFM values, and score everyone in the “at risk” segment. The highest scores get a win-back campaign, and the segment’s common questions feed new landing page and content ideas.

Limitations and things to watch

• The past is not a guarantee. Seasonality, algorithm updates, pricing changes or a new competitor can make yesterday’s patterns wrong.
• You need enough data. GA4, for example, only trains its predictive models when, over a seven-day period in the last 28 days, at least 1,000 returning users triggered the condition (purchase or churn) and at least 1,000 did not.
• Predictions are probabilities. Even a high purchase probability means some of those users will not buy, so use scores to prioritise, and measure the result.
• Correlation is not cause. A model can tell you who is likely to churn without telling you why.
• Models decay. Retrain and check accuracy regularly, and respect consent and privacy rules for the data you feed in.

Learn it on MLforSEO

• Introduction to Machine Learning for SEO (course), including the customer segmentation practical
• Customer Segmentation with Machine Learning Using Three Clustering Approaches
• What Machine Learning Classification Is, and Where It Fits in SEO
• What Machine Learning Clustering Is, and Where It Fits in SEO
• Complete Guide: Understanding ML Classification for Marketers