Kim Kardashian may be earning a nice chunk of change charging for Sponsored Tweets, but according to Yahoo's principal research scientist, marketers who pay celebrities to tweet are likely wasting their money.
Yahoo's Duncan Watts spent two months analyzing influencers on Twitter and found that a large group of people who influence an extremely small group are a more effective (and cheaper) way to share a message than enlisting a celebrity Twitterer.
Speaking today at the BRITE Conference in New York, Watts dicussed "influencers" — a term that is very popular these days, but hard to quantify. (Jeff Jarvis calls them the "false God of social marketing.")
In their 1955 book Personal Influence, Elihu Katzt and Paul Lazarsfeld set out to answer the central question about influence: "Who says what to whom, and with what effect?" But identifying influencers and their reach is much harder than one might think.
As Watts points out, simulation models suggest that apparent influencers are often simply "accidents of circumstances."
Part of the problem in identifying influencers is that the kind of data you need to collect is extremely hard to come across. Says Watts:
"You want to be able to correlate successes and non successes with individuals. On top of all of that, you have a deep ambiguity on what you're trying to look for in the first place. It's unclear even what to measure."
To actually track influence, you in effect "have to be able to observe the entire network of who talks to whom."
That gets a lot easier with a network like Twitter. In addition, URL shorteners are a godsend for brands and individuals who want to track sharing.
"Many people have adopted this protocol of tweeting shortened URLs," says Watts. "As a consequence of that, you have these unique codes that identify one event from another event."
To track influence on the social net, Yahoo's research team crawled 23 million active Twitter users with 1.5 billion connections over a two month interval. They looked at 231 million posts containing URLs from 8 million users. They found that 8 million of those posts contained bit.ly URLs from 500K users. And that is what they focused on.
Yahoo classified a "seed" as a user who tweets a URL. For every follower who subsequently posts that same URL, the seed accrues 1 point. They repeated that measurement for followers of followers to obtain the total influence score for that URL. They then computed an average influence score over all the URLs that each active user put on the network.
What they found was that almost all "cascades" are small and shallow. A tiny fraction are large and propagate up to 8 "hops," or sharing. But even large cascades only reached thousands.
The Yahoo researchers found that individual influence on Twitter is highly unpredictable. The biggest predicator of influence is celebrity. But many of the highly visible celebrities on Twitter are very expensive ways to tap into consumers.
For example, as I've written about before, Kim Kardashian commands $10,000 for a sponsored tweet. But Watts' research found that there is too much randomness in predicting influence on Twitter to make that kind of investment profitable for a marketer. Watts says that while on average things work how you might expect, it is hard to predict the future influence of one individual.
"Targeting many seeds can improve predictive power greatly."
That's because tapping into the networks of individuals who influence on average just one other person can target more people than one hit sent out to a large group at once. It can also be a lot cheaper. Says Watts:
"Influencing one other person is clearly not what many people have been looking for [on Twitter.] But ordinary influencers are still influencers... Combined with mass media this could be very powerful."
Of course, accessing a large group of individuals who might have more power over their small groups of close contacts might not cost as much as hiring Kim Kardashian. But convincing them all to share your message is a whole different issue.