Conversion Value Outliers
Why It Matters
Smart Bidding learns from observed conversion values. A normal $129 e-commerce purchase teaches the algorithm what the average customer is worth; a single $250,000 enterprise contract that fired the same conversion action tells the algorithm an entirely different story. The algorithm cannot distinguish "this is a real but rare outcome" from "this is a data pipeline bug," so it weighs the outlier into its learned distribution. The result: bidding pushes spend toward audiences and contexts that resemble the outlier, even when the outlier is not reproducible. The check identifies extreme value distributions; the team must decide whether to investigate, cap, or filter.
How To Fix It
AdLint's details list each conversion action and the magnitude of the outliers.
For each, investigate the source: are the outliers genuine large transactions, or data pipeline bugs (currency mismatch, decimal-point error, sum of multiple transactions)?
If bugs: fix the pipeline so future values are correct.
If genuine but rare: consider a max-value cap on the conversion action (Google Ads supports this via offline import scripts) or move enterprise-scale transactions to a separate conversion action that does not feed automated bidding.
Re-run Smart Bidding with cleaner signal.
Example
Action: Purchase
Median value: $129
Max value (30d): $48,720 (likely enterprise contract, not typical e-commerce)
Recommended action: investigate; cap or segregate enterprise-scale conversions.Google Ads conversion actions with extreme value outliers (individual conversions reporting values orders of magnitude above the median). Google's value-based bidding documentation, Smart Bidding weighs outliers into its learned distribution, which can push bidding toward irreproducible scenarios. Fix: investigate the source, fix data-pipeline bugs, and consider capping or segregating extreme-value conversions. Source: support.google.com/google-ads/answer/7335652.
Drop this paragraph into your client deliverable. Sources back to the canonical platform documentation linked below.