Red Flags: View-through Conversions

Reading Time: 2 minutes

Why is it important?

This test is related to the concept of a “conversion window”, which defines what purchases on the website will be attributed to Facebook. 

The most simple interpretation of the “purchase” figure is the number of website visitors that came directly from some Facebook ad, and purchased an item. While this interpretation is simple, and good enough for some cases, the number of purchases counted by Facebook is a bit different than that. 

When counting conversions Facebook counts the following:

  • Purchases that happened within 7 days of a clicks on any Facebook ad,
  • Purchases that happened within 1 day of viewing the ad (if the ad was not clicked).

The purchases in the second case are called view-through conversions, where view comes from the fact that these buyers saw the ad on Facebook within 24 hours before buying, but have not clicked on it. This is relevant, because it can be argued how relevant Facebook (or any other channel) was in case of a view-through conversion. 

Two main arguments are:

  • If the ad was relevant to the user, they would click on it. 
  • Just because the ad was displayed, it does not mean that the intended user saw it. It could be the case that the user did not see the ad on the screen, or even that it was rendered below the fold, so it wasn’t visible.

For these two reasons, we consider it good practice to check occasionally how many purchases are being gathered from these view-through conversions. If too many total purchases are being counted due to views, that might imply some problems.

What if the test fails?

The test compares the total number of purchases in the last 30 days, to the total number of view-through conversions in the same period. If too many purchases are being counted from view-through conversions, then the test fails. 

The justification is that in case of a view-through conversion, there could be a third channel responsible for the purchase (email, organic search,…) and this distorts the real data.

Recommended Posts