Did you know that Google’s smart bidding transfers “learnings” about the predicted conversion value of clicks on product A to other products that are similar to A?
It does.
And that’s a good thing.
Here’s one of the reasons this is important: imagine a scenario where over 20% of your campaign’s conversions are generated via clicks on best-selling products that recently went out of stock.
This can be a HUGE challenge for Google’s smart bidding:
- It has learned to predict the conversion value of clicks on these best-selling products, sending them more clicks, and generating more conversion data that it can use to update its models. However…
- Now that these products are out of stock, it can’t send clicks to products for which it has high confidence in high conversion values. It first needs to recalibrate by gathering more conversion data for the other products for which it doesn’t have the same level of confidence in its predictions yet.
- This recalibration can potentially inflate predicted conversion values for clicks on the other products that are still in stock.
Result: your ROAS will (temporarily) fluctuate like the stock market after a Trump tweet, and potentially be waaaaaaaay below your target.
Not good.
Google’s smart bidding sort of tries to anticipate scenarios like these by “transferring” learnings about the conversion value of clicks on product A to other products that it thinks are similar to A.
“Similar”?
Yes, Google’s smart bidding uses deep neural networks that “embed” products in a multi-dimensional space where it “groups” similar products together.
This embedding process uses the values of the attributes in your product feed. Think Google Product Category, Brand, Color, Gender, Price Buckets, and… Product Type!
Here’s the thing:
To better enable Google’s smart bidding algorithms to handle the transfer of “learnings” from out_of_stock products that had good performance to other “similar” products, we need to improve the value of the attribute ‘product_type’ in the product feed to allow Google to make better judgments about product similarity. Accurate, descriptive product_type attributes allow Google to make better embeddings that allow for better generalizations.
Here’s what Google says:
“Here are some useful guidelines on how to improve the effectiveness of key Smart Bidding signals in your Shopping campaigns:
[…] Categorizing your products in 3 to 5 levels via the attribute ‘product_type’ following an arrangement similar to your website will help improve your impression share. (Nils: and predicted conversion values)“
SOURCE: https://shoppingsolutions.withgoogle.com/expertise/smart-bidding/
Don’t let bestsellers that go out of stock ruin your campaign performance. Help Google match them to similar products via descriptive product_type attributes!
– Nils
PS: If you are nerdy like me and like to scan scientific papers that describe these mechanisms, here’s a good read: https://storage.googleapis.com/gweb-research2023-media/pubtools/6407.pdf