Conversion value for e-commerce: gross order value or net of returns?

Take a store where about 20% of orders come back within 30 days. If you import gross purchase revenue, Smart Bidding is learning from money the store never keeps, and it will chase categories with high return rates, sizing-heavy apparel being the classic example.

People usually split into two camps. One says send gross order value, because that's what the customer actually paid and changing it creates a mismatch with what finance sees. The other says net out refunds and cancellations, either through conversion value rules or a daily offline import, so the bidding model stops overvaluing returnable products.

I lean toward netting returns out, but only when the return rate is high and stable enough to matter. Below that, the reconciliation work probably isn't worth the effort. The harder case is when return rates swing a lot between product lines. How do you handle that?

For the swing case, I'd stop trying to fix it in the value field. Conversion value rules key off location, device and audience, not product category, so you can't net out sizing-heavy apparel and leave homewares alone. What does work is sending net value per order, either through an offline import or as a conversion adjustment once the return window closes. Returns are recorded per order, so the model then sees the real money for each one.

The catch is lag. With a 30-day window, Smart Bidding trains on inflated values for a month, which means a stable return rate matters more than the average. If apparel returns run 35% and homewares 5%, I'd split those lines into separate campaigns with their own targets rather than blend them into one number.

Agreed on the campaign split, but mixed carts are where it gets messy. Someone who buys a sweater and a kettle in one order can't be sorted into a category, so that order carries a blended value, and the apparel part keeps pulling the model toward returnable items. You only get clean signal from campaigns that hold a single product type, so check how much of your apparel revenue comes from mixed carts before you lean on the split.

A campaign split sorts clicks by search intent, not orders by basket. Someone who clicks a sweater ad and then adds a kettle still lands in the apparel campaign, so what you're controlling is which query brought the click, not what the order is worth. That's why mixed-cart share matters less than the query mix. If your apparel campaigns pick up clicks from broad terms that also pull in home goods, the split won't clean up the signal, no matter how tidy the campaign names look.

Agreed. I'd look at the split in the search terms report rather than the campaign names. A quick rule of thumb: if the top 20 queries by spend could all be typed by the same kind of shopper, the split is probably clean enough. If "sweater" sits next to "gift ideas" or "winter gifts," the campaign is still mixing baskets, and return rates will often differ sharply from term to term. Move the broad or gifting terms into their own campaign, or add them as negatives, before you trust the target. The query is the only signal you have before the order happens, so it's the lever that actually shapes what the basket ends up worth.