The Facebook stats were looking pretty unreal.

 

WebSavvy has been managing a few facebook campaigns for select clients.  Facebook offers some tantalizing targeting opportunities.  Getting ads in front of “people who purchase online” is mana from heaven for any e-commerce business.

I set up some pretty specific targeting and let loose a campaign aimed at  previous buyers.  I expected some decent results as I loaded up Facebook reporting, but I was not expecting the numbers that popped up.

 

FB - Before

164 conversions in the past 7 days with a staggering conversion rate of just under %20.  The best part was the CPA – under $10.  Start the party, call the client, we’ve found silly good ROI.

But some things are too good to be true.  Dug a bit deeper into Facebooks reporting options and editing the columns I noticed this default in “actions”:

attribution window

By default Facebook is considering any sale that comes within 1 day of seeing an add in the news feed to be a Facebook generated conversion.  Not clicking the ad mind you, just the ad appearing in the news feed.  A news feed that gets flicked past faster than a magazine at a supermarket checkout.

Facebook also considers anyone who clicks on an ad, then buys up to 28 days later a Facebook generated conversion.  An awful lot of things can happen in 28 days to motivate a purchase.

Changing the attribution window to something more immediate should result in more accurate data:

What it should be

1 day after a click is the closest you can get to the actual click with Facebook reporting data.  You’d expect a drop in conversions, and you’d be right.

FB - Conversions After

4 conversions, down from 164.

All of a sudden that CPA doesn’t look so stunning, and the conversion rate remains staggering, but in a whole new way!  The appeal of Facebook for this client has dropped considerably, although more testing is required before its considered a lost cause.

Facebook has some very specific and drool worthy targeting options, but at the moment Facebook reporting data has to be scrutinized very closely before drawing any conclusions.