GA4 Predictive Audiences need at least 1,000 returning users who have triggered the predictive condition AND 1,000 returning users who have not — both within a rolling 28-day window (Google Analytics Help, 2026). At a typical 2-3% ecommerce conversion rate, that translates to roughly 33,000-50,000 returning sessions every 28 days. Most WooCommerce stores under $2-3M in annual revenue never reach the floor. The audience grays out in the GA4 builder with a generic “not eligible” tooltip and no explanation. You haven’t misconfigured anything. You’re below a threshold the marketing copy never mentions.

Why Most WooCommerce Stores Are Locked Out of Google’s Audience Tool#

Through Q1 and Q2 2026, Google has pushed Predictive Audiences hard inside GA4 marketing materials, the Insights panel, and the integrated Google Ads exporter. Three predictive metrics ship in GA4: purchase probability, churn probability, and predicted revenue (Google Analytics Help, 2026). Each one feeds Suggested Audiences like “Likely 7-day purchasers” and “Predicted 28-day top spenders.” None of it works until your property meets a specific eligibility floor — and the floor is hard.

The Eligibility Math, Translated Into Store Size#

The Google Analytics Help page for predictive metrics states the threshold cleanly: at least 1,000 returning users who triggered the relevant predictive condition (such as a purchase) AND 1,000 returning users who did not, both within a rolling 28-day window. New users in their first 7 days do not count toward the 1,000.

Now translate that into a WooCommerce store. Average ecommerce conversion rates run around 2-3% (eMarketer benchmarks). To generate 1,000 returning purchasers in 28 days, you need roughly 33,000-50,000 returning sessions in that window — call it 1,200-1,800 returning sessions every single day, every day, for four weeks. Repeat-purchase rate compresses this further: only ~30-40% of any given buyer base is a returning user in a typical 28-day cycle, which means the underlying total customer volume is materially higher than the headline 1,000.

For stores under roughly $2-3M in annual revenue, the model literally never trains. The audience builder shows the suggested predictive audience grayed out with a “not eligible” tooltip. The tooltip does not name the threshold. Founders read the marketing, click into the builder, and walk away thinking they’ve misconfigured tagging. They haven’t.

Why the Threshold Exists#

The 1,000-and-1,000 floor is not arbitrary. Predictive metrics are supervised machine learning, and supervised models need both positive and negative training samples at scale. A model trained on 1,000 purchasers but only 50 non-purchasers will overfit on noise. A model trained on 50 purchasers and 1,000 non-purchasers will predict “no purchase” for everyone and call itself accurate. Google’s threshold is a defensive minimum for statistical sanity.

What is fair to question is the marketing-to-eligibility gap. The threshold is buried in a help-page paragraph; the audience builder doesn’t surface it; the SMB founder reading “predictive audiences” in the GA4 splash never sees the math. The product is real. The eligibility just isn’t.

The Seasonal Store Trap#

One detail in the help docs deserves more airtime than it gets. If a property meets the eligibility threshold and then drops below it, the predictive metric stops being available for that property until the threshold is met again. The audience effectively turns off.

For seasonal ecommerce — gift, holiday, fashion, garden — this is a structural problem. The model might train during Q4. By February, the property drops back below 1,000-and-1,000, the audience disappears, and the campaign you wired into Google Ads loses its targeting layer at the worst possible moment. Predictive audiences are most useful when they are stable. GA4’s threshold guarantees they are not.

The BigQuery ML Alternative for Stores Below the Floor#

If GA4’s threshold locks your store out, the architectural answer is to build the same predictive layer on your own data. Real-time WordPress event streaming into BigQuery is the prerequisite — once your WooCommerce events are in BigQuery, BigQuery ML opens up.

BigQuery ML supports the same model families GA4 uses internally — logistic regression, boosted trees, AutoML Tables — at no additional ML licensing cost beyond standard BigQuery storage and query (Google Cloud BigQuery ML documentation). A purchase-probability model is a few lines of SQL:

CREATE OR REPLACE MODEL `your_dataset.purchase_probability`
OPTIONS(model_type='LOGISTIC_REG', input_label_cols=['will_purchase_7d'])
AS
SELECT
  user_id,
  sessions_28d,
  pageviews_28d,
  add_to_cart_count_28d,
  last_purchase_days_ago,
  will_purchase_7d
FROM `your_dataset.user_features`;

Schedule the model to retrain nightly via a BigQuery scheduled query. The output is a column on a purchase_probability table that any reverse-ETL tool can sync to Google Ads as a Customer Match list, to Klaviyo as a segment, or to Meta as a Custom Audience. The same architecture handles churn probability and predicted revenue with different label columns.

The key difference: your model is bound by your store’s statistical reality, not Google’s 1,000/1,000 floor. You won’t get a probability as accurate as one trained on 100,000 purchasers — that’s just statistics — but you will get a working predictive layer that the GA4 audience builder is never going to give you. And the model is trained on your data alone, not on Google’s aggregate dataset with yours mixed in. The same approach extends naturally to custom attribution models for WooCommerce, which run on the identical first-party data foundation.

Transmute Engine™ is a dedicated Node.js server that runs first-party on your subdomain. The inPIPE WordPress plugin captures events from WooCommerce hooks and sends them via API to the Transmute Engine server, which streams to BigQuery in real time alongside its other destinations. Once events are in BigQuery, the predictive layer is a SQL exercise, not a Google eligibility lottery.

Key Takeaways#

• 1,000 + 1,000 in 28 days: Both positive and negative samples within a rolling 28-day window. New users in their first 7 days do not count.
• ~33,000-50,000 returning sessions: The threshold translated into typical WooCommerce traffic at a 2-3% conversion rate.
• $2-3M revenue floor: Most stores below this never see predictive audiences turn on.
• “Not eligible” is a data-volume issue: Not a configuration bug. The tooltip never names the threshold.
• Seasonal stores get whiplash: Cross the threshold in Q4, drop below in February, lose the audience exactly when you want stability.
• BigQuery ML is the alternative: Same model families, trained on your own data, bound only by your statistical reality.

Frequently Asked Questions#

If GA4 Predictive Audiences are grayed out for your WooCommerce store, the architectural answer is to stream first-party events into BigQuery and build the predictive layer on your own data. Seresa builds the server-side infrastructure that makes that practical.