Should I compare GA4 only to the ecommerce platform total?

No. Use order data, checkout flow behavior, and event payload evidence together. Platform totals alone do not tell you whether the issue is loss, duplication, or attribution drift.

How do I keep this from breaking after the next release?

Build a checkout QA routine that runs after changes to cart, consent, payment, shipping, discounts, or order confirmation logic.

GA4 Predicted Metrics: Churn Probability, Purchase Probability, and Revenue Predictions (2026)

What are GA4 predicted metrics?

GA4 predicted metrics use Google's machine learning to score individual users on three dimensions: purchase probability (likelihood of purchasing within 7 days), churn probability (likelihood of becoming inactive within 7 days), and predicted revenue (expected revenue within 28 days).

These scores power GA4's predictive audiences — pre-built audiences based on these scores that can be published directly to Google Ads. The critical prerequisite: minimum data thresholds. Without sufficient conversion and non-conversion event volumes, the models cannot train and predictive metrics don't activate.

Most SMB properties never reach the threshold; mid-market e-commerce with consistent purchase volume typically qualifies within 3–6 months of implementation.

The three predicted metrics

Purchase probability

Definition: The probability that a specific user will trigger a purchase event within the next 7 days.

Score range: 0–1 (expressed as a percentage in GA4 audiences: "top 10% likely purchasers")

How it works: The model trains on the behavioural characteristics of users who purchased in the past vs those who didn't. It scores current users based on their similarity to historical purchasers — pages viewed, events triggered, recency of visit, session frequency, device type, and other signals.

Threshold to activate:

At least 1,000 users who triggered purchase (or your configured purchase event) in the last 28 days
At least 1,000 users who did NOT trigger purchase in the last 28 days
The model requires both groups to identify differentiating signals

Churn probability

Definition: The probability that a user who was active within the last 7 days will NOT be active in the following 7 days.

Use case: Identify users at risk of churning before they churn — trigger win-back campaigns or incentive offers before they go inactive.

Threshold to activate:

Same structure: 1,000 users who returned after a period of inactivity vs 1,000 who didn't
For apps: based on app session activity
For web: based on session_start event frequency

Predicted revenue

Definition: The expected total revenue a user will generate within the next 28 days, based on their predicted purchase probability and historical average order values.

Use case: Identify high-revenue-potential users for premium bidding or personalisation. Find users with high predicted revenue who haven't yet made a first purchase.

Threshold: Same as purchase probability — 1,000 purchasers and 1,000 non-purchasers in 28 days.

Checking if your property qualifies

GA4 Admin → Predictive capabilities

Want to see whether purchase, revenue, or item-level tracking is drifting in your property?

Audit revenue tracking

This section shows:

Which predictive metrics are active ("Eligible")
Which are not yet active ("More data needed")
How close you are to the threshold (if shown)

If all three show "More data needed," your property hasn't reached the purchase volume required. The most common bottleneck: not enough purchasers (1,000 per 28-day period = ~36 purchases/day minimum). For seasonal businesses, you may qualify during peak periods but lose model eligibility in off-peak periods.

The predictive audiences

GA4 generates six predictive audiences based on these metrics (Admin → Audiences — look for the "Suggested" tab):

Likely 7-day purchasers — top 10% by purchase probability
Predicted 28-day top spenders — top 10% by predicted revenue
Likely 7-day churning users (purchasers) — users who purchased previously and are likely to churn
Likely 7-day churning users (all active) — all active users likely to become inactive

Publishing to Google Ads:

Admin → Audiences → select predictive audience → Google Ads link → enable publishing

Once published (minimum 1,000 users to activate), use for:

Bid increases in Search for "Likely 7-day purchasers"
Win-back Display campaigns for "Likely churning users"
Premium content offers for "Top spenders"

The Smart Bidding integration

The most commercially valuable use of predictive audiences: bid adjustments in Google Ads campaigns.

Search RLSA with purchase probability:

In Google Ads → Audiences → add "Likely 7-day purchasers" as a targeting observation. Set a bid adjustment of +30–50%. Users who are likely to purchase anyway receive higher bids — capturing them before competitors do.

Performance Max with predicted revenue seed:

Use "Predicted 28-day top spenders" as a seed audience for PMax customer acquisition goals. PMax will find users similar to your highest-predicted-revenue users.

Limitations of predictive metrics

Model accuracy decays seasonally: A model trained on summer purchase patterns may underperform in winter for seasonal retail businesses. Monitor conversion rates for predictive audiences seasonally.

Not available in Explorations: Predictive scores are not available as a dimension in GA4 Explorations. You can only use them through the predictive audience interface (published audiences in Google Ads or GA4 audience builder).

Privacy threshold masking: Google may not show scores for individual users in low-traffic segments to protect privacy. The audience minimum size (1,000 users) applies before the audience can be used in Google Ads.