How advertisers risk wasting money on Facebook Ads

Facebook offers ultra targeted advertising to third party developers. Advertisers should be vigilant in dumping their ad-spend on pricey Facebook ads.


Today Facebook released its so-called Audience Network to developers outside Facebook. This means that developers of, e.g., of Candy Crush can now sell Facebook ads (presumably provided that the app requires Facebook login) specifically targeted to the given user.

We should discuss first how this targeting happens, though I’m pretty sure everyone already kind of knows. Facebook knows A WHOLE LOT about you!! (!!) Everything you’ve ever liked, the interests you’ve listed, the places you’ve said you’ve lived, and probably even the photos you’ve uploaded Facebook has analyzed.

I’m not sure people really understand how powerful machine learning can be here (I don’t have time to enumerate all the ways that Facebook can align these traits to buying behaviors) but with some money (which it has) it’s pretty much golden. Anything it wants to know it can figure out via (empirical) simulation or historical study. Best of all, Facebook has nearly day-by-day histories of its users — which, let’s be honest, is all there really is to know about you (at least as far as advertising goes).

At risk of belaboring this point, kindly contrast this to typical online advertising. In typical online advertising, where at best advertisers can (try to) identify users using unreliable indicators of uniqueness like cookies (y u delete ur cookies internet better w/ cooOOooOOOOkies!), IP addresses (ugh hotspots) new HTML5 magic (cooookie never dieee), or side channel methods (i guess wat kind of computer u have), “useful” targeting is a challenge. I won’t define what I mean by useful, but the word has use.

AND, it turns out, advertisers are willing to pay a whole lot more for these Facebook targeted ads than they are for regular ones. In some ways, this makes perfect sense. For instance, 31 year old dudes are not probably going to go buy Tampax after seeing a Tampax banner ad during their Candy Crush marathons. If you can’t be sure that you’re not showing your Tampax campaign to 31 year old dudes, then you probably shouldn’t have to pay as much, because sometimes you might show your Tampax campaign to 31 year old dudes, who we agreed are probably not going to go buy Tampax.

But advertisers need to be vigilant in paying higher rates for these ads. Let us repeat the tired BUT STILL SOMEHOW NOT OLD adage

correlation!=causation.


Indeed, Facebook will be very good at placing targeted ads. And it may be that these targeted ads are associated to higher rates of purchasing or whatever result the advertiser seeks. But unless done carefully, which in online advertising it’s often not—and even worse Facebook could face a perverse incentive to not do this carefully—Facebook will most likely be ripping off advertisers by merely directing ads toward users with already high propensities to buy.

Consider an extreme case of this. Suppose you’ve put some cash on your kitchen counter: You’re buying a XBox tomorrow!! Yayyy! You have all the characteristics of a person who is likely to buy an XBox. Facebook knows you have all the characteristics of someone who is likely to buy an XBox, so while you’re playing Candy Crush Facebook serves you an ad for an XBox. When analysts tracking the campaign see the ad served to you, they’ll think the ad made you buy. But we agreed that you were already going to buy: You would have bought even if you hadn’t seen the ad! If all purchases arise in this way, the advertiser wastes his money directing all his money at an ad campaign that actually doesn’t do anything. The advertiser could have done just as well by not spending any money at all!

Of course, not all purchases happen this way, but if enough of them do, then a lot of the ad spend will be wasted. This is to say that advertisers need to be careful in spending money on Facebook ads, for Facebook may be so good at identifying soon-to-be-buyers that the people to whom it directs ads don’t even need to see ads for them to end up buying.

So, advertisers, please be careful. Ask Facebook to justify its cost per impression: Are its ads causing users to buy, or do users who just tend to buy (independent of seeing ads!) also those to whom Facebook serves ads?