Viral marketing, randomness and the difficulty of controlling growth in social media September 13, 2007
Posted by jeremyliew in communities, Consumer internet, social media, social networks, viral, viral marketing.trackback
I recently met the CEO of a company who claim to be one of the most popular social networks in Turkey with several million monthly visitors from Turkey. This happened by accident – the founders are Americans who have no prior connection to Turkey.
This is just one of many examples of how difficult it can be to predict or control the growth of viral social media. Google’s Orkut, is a better known example – a social network started by a Turkish engineer working in the US that now dominates in Brazil and India. Friendster and hi5 fall into this bucket as well. As I’ve noted before, the online advertising market in the US is bigger than that in the rest of the world combined. The senior management of these companies know this, and all would love to see more US traffic, but it is now beyond their control.
The reason is the mathematics of viral growth. If the viral coefficient (the number of additional members a new member brings) in a population is less than one, it grows but eventually hits a ceiling. But if the viral coefficient is greater than one, it grows unbounded. Although your social media property may start in many different populations, it will come to be dominated by those with the highest viral coefficients.
This is best demonstrated by example. Consider a new social media property with 30 members, 10 each from three distinct populations; call them Pinks, Purples and Greens. Suppose that by luck and because of the initial users, the viral growth coefficient for the Pinks is 0.6, for the Purples is 0.9 and for the Greens is 1.2. Watch what happens to the populations over time:
All three groups are initially equally represented, But already by time period 4 the population is more than 50% Green. By time period 10 it is more than 75% Green and probably considered a “Green social network”. By time period 16 less than one member in 10 is not Green. For Green now substitute whatever national, language based, religious, racial or other demographic grouping that you choose.
These evolutions can happen very fast since a time period is the time it takes for a new member to invite more members. 2-8 weeks might be a reasonable assumption.
This example is vastly simplified, of course. Viral coefficients vary over time within and between groups. Viral coefficients also don’t vary quite as much between groups; the same underlying feature set is being exposed to all groups. But it illustrates the point that randomness can play a significant role in the eventual makeup of a social media property’s user base.
I posted about a similar finding in May, how you don’t necessarily get the wisdom of crowds, but sometimes just the crowdiness of crowds. It was based on a NY Time’s article that showed how randomness can have a big impact on the most popular songs for a crowd when popularity information is public.
Lightspeed Venture Partners Blog 
So what should an entrepreneur who finds himself/herself in a market where traffic is growing rapidly but where the advertising market doesn’t support monetization (pick your favorite developing economy with an immature ad market) do? Is it better to keep growing and “become big in country X” and hope that the advertising market develops or someone buys you or is there some other strategy? I look at markets such as the Philippines, Vietnam, and a few other places where traffic generally outstrips the ad opportunity.
Charles,
Part of the issue is that there is only so much you CAN do… One option is to keep burn low, open a sales office (and maybe even move the main office) to the geography where you’re big and wait for the market. Another is to get consolidated into a larger, more global entity that can better address the ad market. I’m not sure that you can just say “I’d like to be in the US now because that’s where the ad dollars are” and actually affect that.
Great post Jeremy,
Have you seen any research done on identifying users that have a higher “viral coefficient”? So you know what type of person (or even specific person) to target for your initial user base.
[…] Jeremy Liew has a great post on viral growth. I like what he says about the mathematics of viral growth. Jeremey says “If the viral […]
Jeremy:
I appreciate your efforts to clarify how viral growth factors can affect social websites, and how best to capitalize on this insight. Businesses are always trying to identify new markets, and the hot new growth of social sites in Web 2.0 is key to long-term staying power. I cross-posted on your piece to http://blog.innovators-network.org The Innovators Network is a non-profit dedicated to bringing technology to startups, small businesses, non-profits, venture capitalists and intellectual property experts. Please visit us and help grown our community!
Best wishes for continued success,
Anthony Kuhn
Innovators Network
I’ll be taking that into account if Sendside gets big in Turkey. I’ll have to invest in some translation on the fly platform
“Randomness” in my world mean disturbance factors around trends in variables. I don’t see any in the example: different group exhibit 100% consistent behaviour over time. The difference across groups, that’s parametric in the example, and it is likely to derive from some sort of structural difference. The theory that randomness pays an important part in this may still be true, but this reasoning does not even come close to proving it.
TAMING RANDOMNESS TO LEVERAGE VIRALITY USING THE LAW OF THE FEW
Malcolm Gladwell his “The Tipping Point” refers to an experiment on using a List of about 250 surnames from the phone book and asking people to see if a surname that is shared by someone they “know” (“know” is broadly defined as explained by Gladwell in the book). He reported the following results of the tests given to about 400 people:
1) 24 People surveyed know less than 20 people with surnames on the List.
2) 8 People know more than 90 people on the List
3) 4 People know more than 100 people on the List
The last two groups (2&3) are the “connectors”, which may be extended to include those who know more then 80 people with surnames on the list, assuming a “normal” (bell-shaped) distribution. The following may be inferred:
A) 24 People surveyed know less than 20 people on the List
B) 352 People know between 20 and 80 people on the List.
C) 24 People know more than 80 people the List
It may further be inferred as follows:
A) 24 people (or 6%) surveyed know an average of 10 people each (0-20) or a total of 240 times (1.2%) surnames on the List
B) 352 people or 88% know an average 50 people each (40-80) or a total of 17,600 times (88%)
C) 24 people or 6% know an average 90 people each (80-100) or a total of 2,160 times (10.8%)
The (viral) coefficient for connectedness may be defined as the ratio of (% of total of times “knowing” of names on the List) to (% of population surveyed). Computing further, the following may be derived:
• Viral coef of Group A = 1.2/6 =0.2
• Viral coef of Group B = 88/88=1.0
• Viral coef of Group C = 10.8/6 = 1.8
The above corresponds to the Pink, Purple and Green groups that Jeremy discussed above, with the lucky Green group well- populated with connectors (viral coef of 1.8).
If connectors are so influencing, they can be vital to the success of the social group. What can one do?
First find them.
Then try get connected to the connectors and the connectors will connect and reconnect to provide virality.
Where do you find connectors?
Connectors tend to be more visible, more connected, have more “friends”, send/receive more gifts/messages/points etc….
@ Danny and Amisare,
I am making a different point from the “Tipping Point” point. I’m talking about populations/groups having different viral coefficients, rather than about connectors vs other people. This can be quite a random thing for a particular group, dependent on hard to predict factors like the actual identity of the early adopters, specific features and how they appeal to different groups, etc. I think its hard to “control” these factors by “finding connectors” or pre -identifying groups with a viral coef >1.
Jeremy – Great post. I have been studying a little more about the networks you mentioned. I am not sure I buy into your theory of too many Brazilians/Phillipinos driving everyone else out, unless those groups represent a large majority of the total site membership. That analogy wouldnt hold out for a network like hi5 – they are a top 10 site in over 20 countries. Looks like they have global fragmentation and their virality might be dependent on how dense the local markets are.
Re: money – given that Facebook is experiencing business model challenges right here in the US doesnt mean that the global market is’nt lucrative. I think if these companies can become the top 3 trafficked site in their specific countries, they can take a significant share of the advertising dollars that is available.
Dann
Viral growth represents dangers and complications a business can enter and for want of a better description screw things up, without knowing anything about the market area.
Social networks and the resulting Viral marketing is an amazing method of not only generating traffic and leads but also acts as a great pre seller of products.
What ever the country you are in and whatever demographics your statistics and your testing shows, you have to adapt to your market!
What a Great Quote
6. Jeff Barson – September 14, 2007
“I’ll be taking that into account if Sendside gets big in Turkey. I’ll have to invest in some translation on the fly platform”
The (viral) coefficient for connectedness may be defined as the ratio of (% of total of times “knowing” of names on the List) to (% of population surveyed). Computing further, the following may be derived:
• Viral coef of Group A = 1.2/6 =0.2
• Viral coef of Group B = 88/88=1.0
• Viral coef of Group C = 10.8/6 = 1.8
[…] a tremendous amount of research and writing about virality on the web. This study shows how the success of your startup can come in unexpected locations – Orkut is a social network […]
Youtube Analytics – Youtube Insight
Youtube wartet neu mit einer Statistik-Seite fr seine Nutzer auf (auch Youtube-Mutter GoogleBlog berichtete darber). Es kann so errechnet werden, wann wieviele Nutzer von wo sich ein Video angesehen haben. Dies ist sicherlich sehr hilfreich fr smtlich
[…] importante de ese crecimiento se lo debemos al coeficiente viral, distinción hecha por el VC Jeremy Liew En pocas palabras el coeficiente viral mide la tasa de viralización de un sitio, osea […]