ROAS Statistical Outliers

AdLint calculates the interquartile range across conversion actions with at least 10 conversions and a positive ROAS, then flags anything outside 1.5 times the IQR. It also flags any ROAS above 100 or below 0.01, which are arithmetically possible but practically diagnostic of broken data. The common causes: cents passed where dollars were expected (or vice versa), inflating ROAS by 100x. Currency mismatch where EUR or GBP revenue is treated as USD. A test conversion with a $99,999 value that was never deleted. A conversion action where the value parameter is reading a string like "12.99 USD" instead of the number 12.99 and getting parsed inconsistently. On the low end, a tag firing on every pageview and counting page views as conversions while the value parameter rounds to near zero. Left alone, outlier ROAS feeds Target ROAS bidding bad data. The bidder either chases the inflated action with too much budget or starves the deflated action of impressions. Either way the account drifts off where actual revenue lives.

1. Open each flagged action and pull recent conversion-level data from the Conversions report.
2. Look for value units that do not match the rest of the account (cents vs dollars, wrong currency, parsed-as-string).
3. Compare a sample of conversions to the actual orders in the commerce backend for the same time range.
4. Repair the value source at the tag or import level, then delete any one-off test conversions polluting the history.
5. Once values are correct, give Smart Bidding 2 to 4 weeks to relearn before judging campaign performance against the corrected baseline.