Graph Search’s Dirty Promise and the Con of the Facebook “Like”

Posted on: January 17, 2013
Posted in Social


We all know that the pressure for Facebook to monetize is massive and growing. Yesterday Zuckerberg fulfilled the promise he dropped at TC Disrupt to release a product that will finally compete with Google. To put socially relevant people, places, interests and photos all at your fingertips.

The truth however is that the link between query intent and your social interactions for interests and places is much weaker than FB wants you to believe. In computer architecture they call an out of date piece of data “dirty”. Accessing dirty data is bad, wasting time and causing more harm than good. And in this context, much of the structured data that makes up Graph Search is just that: totally irrelevant and dirty.

It turns out as much as half of the links between objects and interests contained in FB are dirty—i.e. there is no true affinity between the like and the object or it’s stale. Never mind does the data not really represent user intent… but the user did not even ‘like’ what she was liking.

How is this possible? Let me explain.

In the brand advertiser world CPMs have been the preferred measurement (people aren’t going to click an ad for Coke; instead its purpose is to influence you). For the past several years big advertisers on FB have actually been directing massive amounts of paid media to acquire fans. They quite literally bought likes.

Why? Early on FB made the case to brands that they must have fans… together with the ad agencies they convinced the Cokes of the world to spend money to be competitive (hey Pepsi is here too). Then, FB promised, something miraculous would happen. Your friends would see in their news feed you liked Coke!

So… FB convinced big advertisers to spend huge sums on CPA-like ad units whose sole purpose was to acquire fans. Ad agencies dedicated creative, planning and strategy resources to get the Cokes and American Expresses of the world to pay to have users click—almost 100% of the time because the user was promised some sweepstake or contest.

Recall back to all the past campaigns you’ve ignored where you could “like to enter” or “like to qualify”. They are literally everywhere and are always tied to fan acquisition. The numbers are shocking in magnitude: e.g. over the past several years AmEx actually spent about half of its ad spend on buying likes—tens of millions of dollars. Your friends didn’t just go to the American Express fan page and “like” the company for no apparent reason. They did so because they got something.

Across the board big advertisers were told to spend 50% of their ad buy solely on fan acquisition. This is a dirty little secret in ad agency land. Trust me. I’ve seen it firsthand from the marketer, advertiser and agency side. One direct effect of all this passive liking is an ugly messy data set with a bunch of implicit signals… that are wrong. What happens when your girlfriend types in “restaurants in San Francisco” into graph search and P.F. Chang’s gets spit out because it’s the most-liked restaurant. Was a bad Chinese chain the kind of serendipity you were looking for on your date? Didn’t think so.

Sure FB places check-in feature is another signal (beyond the like) I get it… but this isolated piece of structured metadata means almost nothing without massive scale and structure. So… someone in your social graph went to a restaurant and checked in. Wow! Stop the presses. Thank you FB for saving my night, I could not have eaten without you…

And basic math backs up how weak FB’s structured data is in spades. At launch yesterday FB claimed that one trillion connections have been made on the network to date. Great, that’s a lot of restaurant recs, right? Uhm, not really…  FB has 1 billion users. 1,000 signals x 1 billion is 1 trillion. So each user has logged on average 1,000 events/photos/places/things etc. IN TOTAL. And that’s gonna somehow predict where I want to eat? Most of that 1,000 pieces of data are your actual photos and friends.

That the mainstream population who live inside FB day in and day out expect more out of the service without paying is in a word… fitting. After all, FB users are NOT Facebook’s customers. The advertisers are. YES, Amex and Coke and the advertisers are Facebook’s customers, not you!
The truth is Graph Search deserves the exact disclaimer FB gave it… it’s a beta product. Through time, iteration, and effort it can and will be a useful tool for FB power users who are well connected, to find people and to sift through memories.

But the fact is we’re living in a web where services are unbundling, and social is unbundling too. You simply can’t roll up recommendations for people, places, and interests into a service that’s one size fits all.

Graph Search is no more a slam dunk for local / interest intent harvesting than me yelling across an auditorium for which doctor I should see for my cough… Offline we consult different places, people and resources, and you will do the same with social networks and web services online.

One response to “Graph Search’s Dirty Promise and the Con of the Facebook “Like””

  1. Anonymous says:

    Shocking and surprising to read about the level of AmEx advertising spending on Facebook! No wonder General Motors decided to stop throwing money away on Facebook ad’s shortly after the FB IPO last year.The point about dirty data is valid. I suppose that data quality is my field of expertise. "Dirty data" as a term includes outdated information. Dirty data also may be flawed in other ways. Best practices are to (in general) discard dirty data rather than "clean" it. Seeming lack of relevance is not dirty per se, but will require major effort to "mine" effectively. It isn’t so easy! Instagram may prove useful, assuming that it is cost effective to remove those wretched faux life filters that users blight the images with. I have no doubt that plenty of meaningful data can be extracted from an Instagram image! To do, at scale, seems less plausible to me.My example, of "dirtiness" and Instagram filters, is blunt, too specific. The better example, given by Steve Cheney, was of intent as indicated by online activity and actual behavior.

Tweet or Like this post.