Social Value

Deep in our underground hollowed-out volcanic lair, ninja scientists have developed an algorithm that I think will have a large impact on both science and commerce. We call it "Social Value."

Social Value is the amount of behavior that one person generates among their friends. An anology might be the ripple on the social pond. Let's say this person goes to see a movie, listens to a song, or plays a game. Now let's say that this person is influential. How much more likely are their friends to go see the movie, hear the song, or play the game?

Each person has a unique amount of influence in their social network, and on each friend. Maybe you are highly influential on Bob, but Steve doesn't care what you do. What the Social Value algorithm does is to add up all of your influence and put it in units we care about--like sessions, time, or dollars.

Now imagine that we also have a good sense of the person's own future behaviors, thanks to some predictive modeling. This means that for any one person, we're generating two values:

1) An asocial, future behavior number. Marketers call this lifetime value (LTV), but they don't strip out the social impact. This approach does.

2) Social value. An amount, independent of the product, service or activity (i.e. independent of the movie, song or game), caused by the social connections of the person.

Put together and viewed in the aggregate, these values show how much behavior is attributable to the service, and how much is attributable to the community around the service. We're finding that the typical split is about 75/25, where 25c on the dollar is due to community, rather than the product. This number goes up when there's robust community, and down when there isn't.

In gaming, we're seeing a 10-40% range for Social Value.

Knowing these numbers and who they are attached to has a strong and clear value proposition for the developers--find the high Social Value people and a) get more of them, b) retain them, & c) leverage them to cause others to spend more. People with big Social Value numbers can be thought of as "Social Whales."

Conversely--and this I hope to blog on in the future with some shareable data--there's a fun side effect of the scores when a person has a negative Social Value. That's a person who's presence in the network depresses the activity of others. The scientific term is "asshole," or in the game context, troll. I'm looking forward to seeing how firms use this number.

So, that's the commercial side of things. On the research side, I think this algorithm has a lot of applicability in the social sciences. After all, it doesn't care what the behavior is--voting, shopping, writing, singing, whatever.

 

 

 

 

Comments

[Andy Havens]

1.

That's interesting, but there may be a possible rub. Back in my old advertising days, pre-Internet, we wanted to know how much of new traffic into our stores was based on advertising vs personal recommendations. So we did some initial surveys and found out that around 25% of people who came in the door did so because of a friend, family member, etc. saying nice things about the service. We targeted those folks and did some deeper diving and found out that almost all of them had *also* seen an ad... so it's hard to say which was the driver. We also found out that many of the people who did the recommending had initially come in through advertising... So while the initial read seemed like about 25% of sales were dependent on recommendations, what we ended up feeling comfortable saying was that recommendation were *involved* to some degree in around 25% of sales. But that advertising was involved in closer to 95%. We also found out (in a related note) that it was almost impossible to move the recommendation bar by anything other than good overall service. That is, we tried "refer a friend" programs and upped the dough quite a bit, but that didn't have hardly any effect. If people liked us, they'd recommend us. Period.

[Dmitri Williams]

2.

There's a ground truth to the phenomenon out there and it may be 10%, 40%, etc. It's going to vary by context, platform, region, and a lot of other things. But what this approach does via machine learning is to compare cases where there is a social connection vs. those without, effectively controlling for the asocial factors in the model. So, when we find it's X% and the rest is the service, that's what it is.

We haven't done anything in regular retail, chiefly because the social graph data are too difficult to come by, and harder still to attach to PoS data. It may be different than online shopping, or not. And social sign-in may be part of a different phenomenon, or not.

As always, we'll let the data tell the story. And so far, that story for games at least is 10-40%, with a mean around 25%. It's early, still.

[joker]

3.

high school tribes forever... wonderful

[Rory Ocean]

4.

Haha nice use of scientific terms: 'a$$hole'

Social value definitely has relevance. Sometimes it is earned over time eg. if a person recommends a movie and all their recommendations have been great I'll likely see the next one no questions asked, and vice versa. In my eyes their social value either increases or decreases over time based on this. Celebrities, sporting stars etc. hold a certain amount of social value due to their presence in the media. This can take a long time to build up and be extremely valuable, but it can all be destroyed in an instant!