Stages of Web Analytics: Descriptive, Predictive and Prescriptive Analytics

Why the stages matter #

Every analytics project should start with a question. Beasley (2013) frames web analytics as an empirical process: you define a research question, gather web data, transform it, add context, analyse it, and finally answer the question. The stages of web analytics describe how sophisticated that answer can be. Hussein (2022) arranges them on a single “sophistication/value” ladder: descriptive analytics asks what happened, predictive analytics asks what will happen, and prescriptive analytics asks what should be done. Each extracts more information from the data and the methods, and each adds up more decision value.

Before any of these questions can be answered, the underlying data has to be understood and prepared. Web analytics works with three broad data types: structured data (neatly organised, such as a GA4 export in a table form), semi-structured data (such as tagged event streams or emails) and unstructured data (free text, social posts) (AbdulHussein, 2022). Beasley (2013) adds a second distinction that recurs throughout GA4: metrics are the quantitative measurements expressed as numbers (sessions, total users, total purchasers) while dimensions are the qualitative labels (country, device category, channel) used to describe them.

Raw web data is rarely analysis-ready. The data-preparation runs through collection (gathering data from web sources), exploration (spotting patterns, relationships, inconsistencies and missing values), cleansing (correcting errors), structuring (organising the data so it meets the research question) transformation and enrichment (aggregating metrics and augmenting them with further context). This foundation sits beneath every stage that follows.

Descriptive analytics: “What happened?” #

Descriptive analytics reports on past behaviour. It is the most familiar form of web analytics and the entry point for most organisations. Its toolkit is descriptive statistics (counts, averages and distributions), segmentation and temporal reporting. In GA4 this is the world of standard reports, metrics and KPIs: acquisition reports showing where users came from, engagement and conversion reports, and monetisation reports.

Segmentation is the filtering of data by metrics and dimensions so that specific user or content types can be grouped and analysed (Beasley, 2013). Behavioural segmentation divides the user base into groups based on collective behaviour (An et al., 2018), new versus returning users, or users split by geolocation, traffic source or on-site behaviour.

Temporal (time-series) reporting answers questions about how a response variable changes over time i.e. trends, recurring patterns and seasonal variation (Zhang et al., 2009; Sapateiro & Gomes, 2017). A GA4 report tracking total purchasers month by month, for example, makes a seasonal dip or a post-campaign spike visible at a glance.

It is worth noting that GA4 also changed how this descriptive data is captured in the first place. Its measurement model is built on events and parameters rather than the sessions and pageviews of the older Universal Analytics, which makes reporting far more flexible, but also means analysts must decide in advance what they want to track (Akers, 2025; Heap, 2026).

The bridge: “Why did it happen?” #

A descriptive report can tell you that conversions occured; it cannot tell you why. Closing that gap is the job of causal modelling, and it is what turns reporting into insight. The crucial distinction is correlation versus causation: two metrics moving together is not proof that one drives the other (Bounteous, 2018). Establishing why something happened relies on methods such as controlled experiments (A/B testing), attribution modelling, ANOVA and regression analysis (Shmueli & Koppius, 2011). An A/B test comparing two ad versions, or a regression of “key events” on “purchasers,” can isolate the variable actually responsible for a change before any forecast is attempted.

Predictive analytics: “What will happen?” #

Predictive analytics uses past behaviour to estimate future likelihoods. Methodologically it draws on generalised linear models (GLM), linear and logistic regression, time-series and forecast models, and machine- and deep-learning algorithms (insightsoftware, 2024). GA4 makes this stage tangible through machine-learning-driven predictive features (Akers, 2025). Its predictive metrics include purchase probability (the chance an active user makes a purchase within the next seven days), churn probability (the chance an active user does not return within seven days) and revenue prediction (expected revenue over the next 28 days).

On top of these metrics, GA4 builds predictive audiences such as “likely 7-day purchasers,” “likely 7-day churning users” or “predicted 28-day top spenders.” Beyond GA4, the same predictive logic powers smart bidding in Google Ads, upselling and cross-selling, and churn-rate prediction (Widen, 2020). A useful caution: GA4's predictive metrics (purchase probability, churn probability and revenue prediction) are preconfigured in every property and available to all users, not something an analyst sets up. Their quality still depends on the descriptive groundwork beneath them, though: GA4 can only generate a prediction once the property has collected enough of the relevant events, such as purchases, for its model to become eligible (Heap, 2026).

Prescriptive analytics: “What should we do?” #

Prescriptive analytics sits at the top of the ladder. Rather than describing the past or forecasting the future, it recommends actions that drive a positive outcome: reducing risk, increasing revenue or improving the user experience. Its methods include decision trees, simulations and machine- and deep-learning algorithms. Typical applications are recommending the timing of advertising campaigns, suggesting assortment changes, or generating real-time, personalised browsing recommendations for visitors (Sapateiro & Gomes, 2017).

In a GA4-centred stack, the prescriptive move is usually to act on a prediction. A practical strength of GA4 segments and audiences here is their portability: an audience can be sent on to an ads platform such as Google Ads or to a visualisation tool (Warren, 2011). Exporting a predicted-value or likely-purchaser audience into Google Ads, for instance, lets budget and messaging be steered automatically toward the users most worth reaching: the point at which an estimate of what will happen becomes a concrete recommendation about what to do.

Conclusion #

Web analytics is not a single activity but a progression, and the three stages mark how far up it an organisation chooses to climb. Descriptive analytics establishes what happened, the causal step asks why, predictive analytics estimates what will happen, and prescriptive analytics recommends what to do next. Each question extracts more from the same underlying data than the one before it, and each carries more decision value. In a GA4-centred stack the ladder is unusually concrete: standard reports and segments answer the descriptive question, experiments and attribution close the causal gap, built-in predictive metrics and audiences look ahead, and the export of those audiences into a platform such as Google Ads turns a forecast into an action.

The stages compound rather than compete. A prediction is only as sound as the descriptive groundwork beneath it, and a prescriptive recommendation is only as sound as the prediction it acts on, so climbing the ladder is a matter of building each rung on a solid one below, not of abandoning the lower stages once the higher ones come into view. An organisation that reports well but never asks why, or that forecasts without acting, leaves most of the value on the table.

The real payoff, then, comes from connecting the stages rather than treating them as separate tools. That connection still begins where any analytics project should: with a clear question. The stages simply describe how much a well-prepared dataset, and the methods applied to it, can be made to answer, from a single report of what happened to a concrete recommendation about what to do next.

References #

AbdulHussein, A. (2022). Data analytics and decision making. University of Windsor.

Akers, J. (2025, May). Introduction to Google Analytics 4. Digital Culture Network. https://digitalculturenetwork.org.uk/knowledge/introduction-to-google-analytics-4/ 

An, J., Kwak, H., Jung, S. G., Salminen, J., & Jansen, B. J. (2018). Customer segmentation using online platforms: Isolating behavioral and demographic segments for persona creation via aggregated user data. Social Network Analysis and Mining, 8(1), 1–19.

Beasley, M. (2013). Practical web analytics for user experience: How analytics can help you understand your users. Newnes.

Bounteous. (2018, August 7). Seeing causality in Google Analytics data. https://www.bounteous.com/insights/2018/08/07/seeing-causality-google-analytics-data 

Heap. (2026). Google Analytics 4: Everything you need to know. Heap. Retrieved June 29, 2026. https://www.heap.io/topics/google-analytics-4 

insightsoftware. (2024). Top 5 predictive analytics models and algorithms. https://insightsoftware.com/blog/top-5-predictive-analytics-models-and-algorithms/ 

Sapateiro, C., & Gomes, J. (2017). Leverage web analytics for real time website browsing recommendations. In World Conference on Information Systems and Technologies (pp. 538–548). Springer.

Shmueli, G., & Koppius, O. R. (2011). Predictive analytics in information systems research. MIS Quarterly, 35(3), 553–572.

Warren, J. (2011). Segmentation in web analytics: A fundamental approach. https://www.kwantyx.com/wp-content/uploads/AT_WP_Segmentation_EN.pdf 

Widen, S. (2020, February 10). Predictive web analytics in marketing. Forbes. https://www.forbes.com/sites/forbesagencycouncil/2020/02/10/predictive-web-analytics-in-marketing/ 

Zhang, Y., Jansen, B. J., & Spink, A. (2009). Time series analysis of a Web search engine transaction log. Information Processing & Management, 45(2), 230–245.