Value per Conversion Variance

Coefficient of variation is the standard deviation divided by the mean. A CV above 150% on a single category means the spread of values is larger than the average value. Practically, that looks like a Purchase category where some conversions come in at $12, others at $4,200, and the mix is not explained by product mix variability. The usual root causes: a single Purchase action receives both retail orders ($30-150) and B2B wholesale orders ($1,000-10,000) without splitting them into separate conversion actions. Or the value parameter is reading from two different data layer locations depending on the page type, and the values come through with different magnitudes. Or refund logic is firing through the same path with negative-adjacent zero-dollar values. For Smart Bidding, this is a learning problem. Target ROAS bidding assumes the value distribution per action is roughly predictable. When the variance is this wide, the bidder either hedges (bids low across the board, missing the high-value orders) or overcorrects after a single high-value conversion (bids up, then ROAS collapses on the next 50 small orders).

1. Pull the conversion details for the flagged category in Google Ads Reports. Sort by value descending.
2. Look at the top and bottom of the list. If they represent legitimately different business outcomes (retail vs wholesale, free trial vs paid, etc), split them into separate conversion actions.
3. If they should be one category, investigate the value source for inconsistent readings (different DOM elements, different data layer keys, currency mixing).
4. Once split or repaired, expect Smart Bidding to relearn over the next 2 to 4 weeks.