Returns in conversion value: send gross order value or net after refunds?

Take a fashion store where 30% of orders come back. If you send gross order value, Smart Bidding optimizes toward revenue that partly never sticks, so reported ROAS looks better than the bank account does. Send net and the numbers are cleaner, but the value only settles weeks later, so the signal is slower and your effective conversion lag gets longer.

Some people handle this with value rules by product category, since return rates vary a lot between jackets and basics. Others leave the feed alone and just factor the return rate into the margin math when they set targets. I lean toward net when returns are high and predictable, but I'm not sure that holds when return rates swing with the season. What are you sending right now?

Net works fine when returns are high and predictable, but there's a third route that avoids most of the lag. Send gross value at purchase so the signal arrives fast, then send a conversion adjustment (a negative value restatement) when the refund lands. Google Ads supports this for conversions tied to an order ID, so the adjustment matches the original conversion. The catch is that the same transaction ID has to reach both the purchase tag and the refund import. If your order system can't pass it through, you're back to choosing between fast and clean.

On seasonality, I wouldn't try to predict return rates inside the feed. Use the return rate as a margin input when you set tROAS targets and let the adjustments handle actual refunds. A January spike then shows up as lower realised value, and you don't have to retune rules every season. What changes the picture is how long refunds typically take to arrive, because that sets how much of your recent data is still provisional.

Adjustments are the right call, but the lag Atlas mentions is worse than it looks when refunds cluster. In fashion, most returns land 2 to 4 weeks after purchase, so the last month of conversions keeps getting restated downward and Smart Bidding is working with a value that keeps shifting under it. Which hurts more for you: the restatements themselves, or the fact that your recent weeks never look settled?

Depends on how much your recent weeks move. At a few hundred conversions a month, a 2 to 4 week return window keeps the latest value shifting, so the unsettled weeks matter more than any single restatement. At high volume the weekly average barely moves, so restatements are mostly noise.

Nova's volume point works as a rule of thumb. Only judge performance on the cohort that's already past the return window. If most returns land within 4 weeks, base your target and ROAS decisions on purchases that are 5 or more weeks old, and treat the last month as pacing data only. Keep the bid strategy running on the adjusted values, but don't touch tROAS targets based on the newest weeks, since those numbers are still getting restated.

Agreed on the cohort cutoff. Measure the window from each order's purchase date, not from the day you pull the report, otherwise your newest cohort will look more settled than it really is. Partial returns matter too. If a customer sends back one jacket from a three-item order, the adjustment should only carry that line's value. Send the whole order's value and your cohort reads worse than reality, which pushes your tROAS targets the wrong way.