Purchase Probability

Analytics

Also: Purchase Likelihood · Propensity to Purchase

Purchase Probability = model score 0-100 assigned by platform ML to each user based on behavioural signals

What it isML score predicting a user will buy

SourceGA4 predictive audiences

Watch forScore needs enough conversion history

Use it forAudience segmentation and bidding

Quick definition

Purchase probability is a machine-learning score that predicts how likely a specific user is to complete a purchase within the next seven days. Google Analytics 4 (GA4) generates this score automatically for properties with enough conversion history, and it powers predictive audiences you can use in Google Ads.

How it varies across Australia

Purchase probability scores are only generated once a GA4 property crosses minimum conversion thresholds. Many Australian mid-market sites never reach those thresholds in a single property, which means the feature is available to fewer businesses than Google implies.

See data and tracking maturity across Australian industries →

What it actually means

Purchase probability is GA4's way of turning past behaviour into a forward-looking score. For every user in your GA4 property, the model looks at signals like pages viewed, products browsed, cart additions, time on site and historical purchase patterns, then assigns a score representing the likelihood they will buy within the next seven days.

The score itself is not shown directly in the GA4 interface as a raw number. What you see is a derived audience: users in the top segment of that score, labelled something like 'likely seven-day purchasers.' You feed that audience into Google Ads and bid differently for people the model thinks are close to converting.

The mechanism is similar to how churn probability works in reverse. Churn models predict who is about to leave. Purchase probability predicts who is about to arrive. Both rely on the same underlying principle: past behaviour in a cohort predicts near-term behaviour for individuals who look like that cohort.

The catch is data volume. GA4 requires a minimum number of purchase events over a rolling window before predictive metrics activate. Properties below that threshold show a greyed-out card and no audience. This is the feature most often promised in agency pitches and least often actually available to the client.

A purchase probability score is only as good as the conversion data feeding it. Garbage in, confident garbage out.

How to calculate it

Purchase Probability = model score 0-100 assigned by GA4 ML using behavioural signals across recent sessions

Worked example. A user visits your ecommerce site three times in five days, views the same product twice, adds it to cart and abandons. GA4's model compares this pattern against all users who did purchase and assigns a score. If their pattern closely matches historical converters, their score sits near the top of the range. That user appears in your 'likely purchasers' audience.

The Australian context

Australian ecommerce sites often run lower absolute conversion volumes than equivalent US properties because the market is smaller. This makes crossing GA4's minimum threshold harder, and predictive audiences less reliable when they do activate. Australian businesses running separate GA4 properties per brand, state or domain compound the problem by splitting the conversion data across properties. Consolidating to a single property with cross-domain tracking usually produces better predictive audience quality than any optimisation done within a fragmented setup.

Where people get this wrong

Assuming predictive audiences are available just because GA4 is installed.GA4 requires a minimum number of purchase events over a rolling 28-day window to activate predictive metrics. Properties below that threshold cannot generate the score, and most mid-market Australian sites are below it.

Treating a high purchase probability score as a confirmed intent signal.The score is a statistical estimate, not a declaration. Bidding heavily on 'likely purchasers' still requires testing whether the audience actually converts at a lower cost per acquisition than your standard targeting.

Ignoring attribution overlap between purchase probability audiences and remarketing lists.Users who score high on purchase probability have usually already seen your ads or visited multiple times, so they may convert regardless. Without an incrementality test, you cannot tell whether the audience is causing conversions or just predicting ones that would have happened anyway.

Common questions

How do I activate purchase probability in GA4?

GA4 activates predictive metrics automatically once your property records a minimum number of purchase events from a minimum number of distinct users over a rolling 28-day window. You cannot force it. The best path is ensuring purchase events are firing correctly and consistently, then waiting for volume to accumulate.

Can I use purchase probability audiences in Meta Ads?

Not directly. Purchase probability audiences are generated inside GA4 and exported to Google Ads via the linked account connection. For Meta, you would build a comparable audience using your CRM or pixel data to find users who match high-intent behaviour patterns, then upload as a custom audience.

Is purchase probability the same as remarketing?

No. Remarketing targets users who visited your site. Purchase probability targets users the model believes are close to buying, which can overlap with remarketing but also includes users who are further along the path than a simple visit would indicate. The two can and often should be run together.

What happens to the score if I change my conversion setup in GA4?

Changing the events that feed purchase probability resets the model's training data. Renaming key events, removing purchase tags or switching to a different conversion event will cause the model to rebuild from scratch, which takes time and can temporarily degrade audience quality.

Keep exploring

About New Rebellion

New Rebellion is a marketing intelligence consultancy. We build tools, score Australian businesses on how their marketing actually performs, and publish Debrief every day. This dictionary is part of how we work in the open.

How we think →