Why Poor GA4 Data Is Hurting Your Google Ads Performance

Learn why GA4 predictive audiences become ineligible and how ecommerce tracking issues impact machine learning, attribution, and Google Ads optimization.

Skylar van Dalen-Flude

Senior Data Analyst

Bridges the gap between what the data actually says and how it’s interpreted across the business. Passionate about bringing clarity to complex GA4 and analytics environments where the answers are rarely straightforward. Turns stakeholder confusion into clean measurement strategies and, occasionally, uncomfortable truths.

Why GA4 Predictive Audiences Stop Working: How Poor Ecommerce Data Breaks ML and Smart Bidding

Bad ecommerce data doesn't just produce wrong numbers. It can silently disable GA4's most powerful machine learning features, locking brands out of predictive audiences and weakening Google Ads bidding performance.

GA4's built-in machine learning models require clean, structured ecommerce data and strict eligibility thresholds that many properties fail to meet, often because of preventable implementation issues. The compounding cost is significant: brands with strong data foundations unlock forward-looking audience targeting and better bid optimization, while competitors with broken implementations lose access to these capabilities entirely and cannot retroactively repair the missing historical data.

In this guide, we break down the most common GA4 ecommerce tracking issues that prevent predictive audiences and machine learning features from working properly, how to identify them, and what teams should validate before the damage compounds.

How GA4 Predictive Metrics and Audiences Work

GA4 offers three predictive metrics powered by on-property machine learning:

purchase probability, likelihood an active user purchases within 7 days
churn probability, likelihood an active user goes inactive within 7 days
predicted revenue, expected revenue from a user over the next 28 days

Both purchase and in-app purchase events feed into the purchase probability and revenue prediction models.

These predictions are generated once per day for each active user, and they power both Explorations reporting and predictive audiences that automatically sync into Google Ads, Display & Video 360, and Search Ads 360.

Brands that successfully qualify for predictive audiences gain access to significantly more advanced targeting strategies compared to standard remarketing audiences based only on historical behavior.

GA4 Predictive Audience Eligibility Requirements

GA4's predictive engine requires data quality and sufficient volume that many properties struggle to maintain.

Per Google's official documentation, a property must meet all of the following:

Minimum User Threshold

At least 1,000 returning users who triggered the relevant condition (for example, made a purchase) and at least 1,000 returning users who did not within a 7-day window, evaluated over a rolling 28-day period.

Sustained Model Quality

The model must maintain sufficient quality over time. If it drops below the minimum threshold, GA4 stops generating predictions and predictive audiences may become unavailable.

Required Event Parameters

For purchase probability and predicted revenue, the property must send purchase events with both value and currency parameters. If either parameter is missing, the predicted revenue model cannot train properly.

Five pre-built predictive audiences become available once eligibility requirements are met:

likely 7-day purchasers
likely first-time 7-day purchasers
likely 7-day churning users
likely 7-day churning purchasers
predicted 28-day top spenders

Custom predictive audiences can also be built using percentile-based thresholds.

When a property doesn't meet prerequisites, these audiences appear grayed out and marked as "Not eligible to use" in the audience builder. There is no proactive notification, so teams must manually monitor eligibility status.

Six Ecommerce Tracking Issues That Silently Break GA4 Machine Learning

The most damaging failures fall into distinct categories, each with specific downstream consequences for predictive audiences, Smart Bidding, and machine learning quality.

1. Missing or Malformed Purchase Parameters

GA4's ecommerce specification requires a transaction_id, value, currency, and an items array on every purchase event.

Common errors include:

sending currency symbols instead of ISO 4217 codes
embedding currency in the value string
incorrect capitalization of parameter names
incomplete purchase payloads

Impact

If currency is absent, even when value is present, revenue data may not display correctly in reports at all. The predicted revenue model loses critical training data.

The purchase event may still count as a conversion, which makes the issue difficult for teams to detect.

Fix

Validate every purchase event in DebugView before publishing.

Confirm:

value is numeric
currency is a valid ISO 4217 string
the items array is populated with at least one object

2. Duplicate Transactions

Double-fired purchase events commonly result from:

thank-you page reloads
GTM misconfiguration
SPA routing issues
dual tracking implementations
browser/server duplication

Impact

Inflated revenue corrupts model training data and makes ROAS appear artificially high, causing Smart Bidding systems to optimize toward inaccurate signals.

Media teams may unknowingly trust campaign performance based on duplicated conversions rather than actual business outcomes.

Fix

Prevent duplicates at the implementation level using cookie or localStorage-based transaction logging.

Don't rely on GA4's server-side transaction ID deduplication as multiple practitioners consistently report it is unreliable in the GA4 UI.

3. Fragmented User Identity

Without user ID implementation, a single user browsing on mobile, desktop, and tablet appears as separate users.

A particularly destructive error is setting user ID to fallback values such as:

"none"
"null"
empty strings

after logout or failed authentication states.

Impact

Fragmented identity inflates user counts and reduces the returning-user signals GA4 needs to meet predictive audience eligibility thresholds.

Since Google Signals was removed from GA4's reporting identity in February 2024, user ID is now the primary mechanism for cross-device stitching, making implementation gaps more costly than ever.

This becomes especially problematic for brands with longer purchase cycles where users regularly switch devices during the buying journey.

Fix

Implement user ID for authenticated users and ensure it is only set when a valid identifier exists.

Audit user ID population rates in BigQuery regularly to establish a baseline and catch regressions.

4. Missing Ecommerce Funnel Events

GA4's ecommerce funnel includes:

view_item_list
select_item
view_item
add_to_cart
begin_checkout
add_shipping_info
add_payment_info
purchase

Many implementations only fire the purchase event and skip the rest.

Inconsistent item ID formats across events create additional issues by preventing GA4 from constructing unified product journeys.

Impact

Limited behavioral signals reduce machine learning model quality and weaken predictive audience accuracy.

Google explicitly states that collecting a broader variety of recommended ecommerce events improves predictive model performance.

Fix

Implement the full recommended ecommerce event sequence and enforce a consistent item ID schema across all events.

Document the product ID standard clearly across analytics and development teams.

5. UTM Parameter Inconsistencies

GA4 treats UTM values as case-sensitive.

For example:

Facebook
facebook
FACEBOOK

are interpreted as separate traffic sources.

Additional problems often include:

non-standard medium values
inconsistent naming conventions
redirects stripping UTM parameters
SPA frameworks removing query parameters

Impact

Fragmented attribution data weakens the channel-level signals GA4's machine learning models use to score and predict user behavior.

Traffic may also route into the Unassigned channel, reducing reporting reliability.

Fix

Enforce a centralized UTM taxonomy using shared naming conventions and standardized URL builders.

Audit acquisition reports regularly and create custom channel definitions where necessary.

6. Referral Exclusion Misconfiguration

Third-party payment processors such as:

PayPal
Stripe
Klarna

are not excluded from GA4 referrals by default.

When users leave and return during checkout, GA4 may start a new session attributed to the payment processor instead of the original marketing source.

Impact

Revenue attribution becomes distorted and conversion credit shifts away from the campaigns that actually drove the purchase.

This weakens both reporting accuracy and Smart Bidding signal quality.

Fix

Add payment processor domains to GA4's unwanted referrals list.

Verify the implementation by confirming payment domains no longer appear as acquisition sources in reporting.

How to Validate and Repair GA4 Ecommerce Tracking

DebugView and GTM Preview are the first line of defense for ecommerce QA.

Teams should validate the full ecommerce journey, including product views, add-to-cart interactions, checkout progression, and purchase completion. Purchase events should fire exactly once and contain valid transaction_id, value, currency, and item parameters.

A strong benchmark is maintaining GA4 revenue within approximately 5% of the ecommerce platform's native reporting.

BigQuery export provides an even more powerful validation layer by allowing teams to identify duplicate transaction IDs, missing parameters, empty item arrays, inconsistent schemas, and user ID gaps through SQL-based monitoring.

One important limitation is that GA4 data cannot be repaired retroactively. If tracking was broken historically, that data loss remains permanent.

BigQuery as a QA and Monitoring Layer

When predictive audiences stop generating, advertisers lose access to forward-looking targeting capabilities and fall back on traditional remarketing audiences based only on historical behavior.

The impact on Smart Bidding can be substantial.

Google's bidding systems rely heavily on accurate conversion data to optimize bids, prioritize high-value users, and allocate spend effectively. When ecommerce tracking becomes corrupted, bidding algorithms optimize toward inaccurate signals.

Over time, the gap compounds. Brands with clean data continuously improve machine learning quality and bidding performance, while brands with corrupted data steadily weaken their optimization baseline.

Final Thoughts

Poor ecommerce tracking in GA4 does more than create reporting discrepancies.

It directly impacts predictive audiences, Smart Bidding performance, attribution quality, and the reliability of downstream marketing decisions. When machine learning models train on incomplete or corrupted conversion data, optimization performance declines quietly over time.

The challenge is that GA4 data issues are largely irreversible. Missing parameters cannot be backfilled, duplicate purchases cannot truly be removed from historical reporting, and predictive models cannot rebuild lost behavioral history retroactively.

That makes implementation quality critically important.

Teams that continuously validate ecommerce tracking, monitor data quality, and enforce consistent tagging standards are far more likely to unlock GA4's predictive capabilities and maintain stable optimization signals across Google Ads and other marketing platforms.

In practice, clean data is not just an analytics requirement. It is a competitive advantage.

More Insights