Facebook vs. Google Analytics – How to evaluate Facebooks performance?

Facebook vs GA

Facebook is playing an important role for most performance advertisers. However, understanding how Facebook contributes to the overall conversions is one of the trickiest questions. First of all, Facebook encourages you to look at the Facebook conversion data while Google is telling you a whole different story using the last-click based Google Analytics data. The visualization below explains the issue.

image1

The topic is complex and we wrote multiple articles on the topic of double counting, see How to spend $520 000 on Advertising and Accomplish Absolutely Nothing and why you can’t match Facebook conversions to Google Analytics data.

In this article, we’re going cover learnings we had from clients on how to measure the impact of spend on Facebook data on Google Analytics conversions.

A few steps are necessary to get good insights into your the contribution your Facebook campaigns have on overall conversions.

Get all the data in one place

First, you want to make sure that you have all the data from Google Analytics (conversions) and Facebook (costs) in a database table or in a spreadsheet with a daily granularity. To get started really simple you can just export data from Facebook in CSV/Excel format and do the same from Google Analytics. If you would like to get the process a bit more automated and avoid wasting time on manual operations, you can get a variety of different tools (a comparison can be found here).

The outcome of your extraction should be two tables which look similar to what you see below.

conversions

Table with Google Analytics Conversions

 

spends

Table with Facebook spendings

 

Get your insights

Now, we are going to look at the impact the Facebook spend has on Google Analytics conversions. We are now looking at how the media spending on Facebook impacts conversions on Google Analytics. Over time we have discovered that the strongest relationship between spendings (Facebook) and conversions (Google Analytics) is if we compare the spendings with paid search brand campaigns, also and organic search conversions.

Basic

That’s why in this very simple performance evaluation we are going to look at the data on weekly levels to understand what our weekly Facebook spend should be.

We are going to look at the data in a bubble chart:

  • The entity of the chart (one bubble) = 1 week (Important: read the note below on how to determine the best entity)
  • On the X-axis we look at the Spend on Facebook
  • On the Y-axis we look at the Conversions we have on paid-search Brand campaigns and Organic search conversions

 

Brand and Organic vs. Facebook

Brand and Organic conversions vs. Facebook spendings

 

Now, these insights on the data set give a clear indication that we have an upward sloping trend on brand and organic conversions in the weeks with increased Facebook spending. Based on these insights the advertiser would ideally spend between $2000 and $3000 per week on Facebook.

The reason why we are choosing Brand and Organic conversions is that in multiple data sets we see the strongest correlation between Facebook spendings and these outcomes based on our client’s data. This may be different depending on your customer journeys.

An important note here is that the defined entity really depends on the product you are selling or the conversion you want to achieve. If you have a product which is simple or is an easy sell the journey is shorter, whereas if you have a complex product you will have a longer customer journey.

So while for a luxury hotel (conversion > $ 5000) you might look at a 30 days conversion window, it is probably wise to look at the impact differently.

Advanced

To determine the effect of spending on revenue you can use a set of linear regressions. First you probably also want to clean data from the effects of special events (e.g. Black Friday or singles day, 11.11). You can do this by removing revenue that is larger than 200% of the median value of revenue in the preceding 15 days and the following 15 days.

The best is to run a linear regression model separately for consecutive overlapping periods of 35 days (or longer, in case you have a really complex product you are selling). The result is a set of coefficients for Facebook for each day. With this approach, you can clean data from seasonal effects and you can better deal with scattered data.

These regression coefficients show the effect of an increase in the costs of Facebook on total revenue (at a daily level),

These coefficients can be used to calculate relative shares of revenue that can be attributed to Facebook.

 

Wrapping it up

The John Wanamaker quote

Half the money I spend on advertising is wasted; the trouble is, I don’t know which half.

Sums up every modern marketer’s biggest problem quite nicely. With the walled garden approach of many large media companies, there is a tendency to look at data in silos rather than trying to understand the impact of spend on one channel to an outcome on another channel. I hope this post helps you to maximize the performance on your Facebook campaigns.