<- Kitchn.io Blog

How can you "train" the Facebook algorithms to deliver the best possible results for your ads?


How can you "train" the Facebook algorithms to deliver the best possible results for your ads?

Recently we got asked this by one of our D2C customers who has been able to scale nicely by focussing on creative testing and automating most of his manual work: "My campaigns' performance fluctuates a lot! One week is amazing and another just average. What can I do?"

Our answer:

Fluctuating performance usually means one thing - the Facebook algorithm doesn't have enough (good) data. Either because you don't have enough conversions or you're not sending the (right) data back.

In the previous posts we've been already stressing out the importance of working together with Facebook machine learning, and not trying to micromanage it as Facebook algorithm by itself is the best performance marketing manager you can possibly hire. Now, I wanted to talk about the 3 most important things you can do to help the algorithm understand who your customer is and how to get more of them!

1. Volume!

Receive at least 50 lower-funnel optimization events on the adset level (or around 200 for the upper-funnel events). If you are optimizing your campaigns for lower-funnel events such as Purchase or Subscription make sure you have rather few adset with enough optimization goals to exist the learning phase, than many with just a few of them. Targeting by itself doesn't matter that much if your adsets receive too few events. The best possible targeting you can ever achieve is feeding enough events to the machine, so it will find more people similar to those who bought, and, ultimately, stabilize your campaign performance.

2. Ratio!

Your daily budget/bid ratio should ideally be around 10. Show Facebook that you wanna scale and have budgets to achieve your goals. The algorithm needs to see that with this budget and a bid it's possible to achieve the absolute minimum to deliver stable performance (10 per day will be just above the minimum required 50 per week). In my experience, putting on a significantly higher daily budget (10-15 times hire on what you are planning to spend) and restricting the bid usually delivers the best results with a stable number of events per week.

3. Consolidate!

Combine some of your adsets based on the targeted persona. If some of your adsets are targeting the same persona AND are optimized for the same event, it might make sense to combine them to get more data on the adset level and let it run for months, instead of creating the new ones. The more events you will get on the adset level, the better the predictions for the future customers will be. In my experience, reactivating old campaigns with 1000k+ events will most probably work better than starting a completely new one.

It might not be easier in the beginning, and you will need to start out a bit more aggressively, but being consistent in training the Facebook algorithm is a good investment of your time and effort. that will bring the results back.