The Google FeedGen team has added a guide for optimizing feeds using BigQuery. FeedGen is ideal for up to 30k items; for larger product feeds, it’s recommended to use the BigQuery option.
Start Small First
In my opinion, it’s always good to start small. Try optimizing a smaller number of products and run an A/B test to see the results. For proper A/B testing, I use the Google Shopping A/B test tool from Adchieve, but you can also use Google FeedX, which is free but slightly more technical. If the results are positive and you want to optimize more products, then use the BigQuery option. Otherwise, it might consume a lot of time without significant performance gains. When a potential client requests to use Google FeedGen, I always first check their product feed for improvement opportunities. If there’s no headroom, I suggest other optimization methods.
Costs
The Vertex AI costs for this approach should be less than those of FeedGen since titles and descriptions are processed in batches, reducing the frequency of static prompt evaluations. Notably, you must choose between the Flash and Pro models, which have a price difference of 10x (as of July 2024). Additional BigQuery costs should be minimal for typical product feeds.
