The prospect of going viral has fascinated entrepreneurs and observers since the success of Youtube, Facebook and more recently, Snapchat. Though many have tried to replicate the word-of-mouth enthusiasm that pushed those companies to billion-dollar valuations, most have failed. Marketing professionals write about virality as a holy grail of user acquisition, but the idea of studying it scientifically seems quixotic given the rarity of success. After all, science requires replicability. Nevertheless, academic and professional study has grown around the subject in an attempt to make virality more predictable and understandable.
Entrepreneurs, venture capitalists and academics have written about how to measure and project growth from virality, as well as how to combine it with other forms of marketing when a product isn’t “fully viral.” In my own experience creating a “partially viral” app, I’ve created a framework to think about ways to increase the virality of a product analytically. Improving virality can be thought of in the same way as any other user acquisition channel: as a pipeline management problem.
David Skok provides a nice overview on the virality coefficient. The virality coefficient is a measure of how many new users an existing user directly acquires. For example, if each user of an app sends out 5 invitations to friends and 10% of the friends accept the invitation, then each existing user directly brings in 0.5 new users. The virality coefficient is 0.5. Typically, most apps can measure the virality coefficient as the number of invitations sent multiplied by the conversion rate of invitation to usage, but this can vary depending on the exact word-of-mouth mechanism of the product (more on this later).
As Skok notes, in order for something to go fully viral, the virality coefficient must be greater than 1. Skok provides a nice excel model to show this, but there’s also a simple mathematical explanation. If any existing user directly brings in 0.5 new users, then those 0.5 new users will bring in 0.25 additional users, who will bring in 0.125 users, etc. This is a geometric series whose sum converges when the virality coefficient is less than 1 and diverges to infinity when the virality coefficient is greater than 1. Going viral simply means that you can expect to get an infinite number of users if your product existed for an infinite amount of time, ignoring the inevitable capacity constraints and overlapping invites. When the virality coefficient is 0.5, the sum converges to 2. Every existing user will eventually be the equivalent of 2 users once all the invites have been sent, assuming users only send one round of invites.
For products not lucky enough to go fully viral, Rahul Vohra has written about the hybrid model that combines virality with traditional marketing channels to amplify growth. When the virality coefficient is 0.5, the product might not be fully viral, but when a user is acquired through a traditional channel, that user will eventually be the equivalent of two users (the lifetime value is technically less because of discount rates and cycle times, but we’ll simplify in this discussion). The lifetime value of acquiring a user becomes much higher, allowing marketers to acquire paid users at a higher price.
Pipeline management for virality
Instead of investigating how virality can work together with other user acquisition channels, I’ve dedicated my time toward increasing the virality coefficient. Bringing a virality coefficient of less than 1 to greater than 1 makes later marketing efforts more effective, but any increase in the virality coefficient can add value. For example, if the virality coefficient goes from 0.1 to 0.2, then each user becomes the equivalent of 1.25 users instead of 1.11. Increasing the virality coefficient from 0.8 to 0.9 actually brings the value of each user from 5 to 10. Graphing the geometric series shows that entrepreneurs get considerable value from increasing the virality coefficient when it is between 0.5 and 1.
Increasing the virality coefficient requires breaking down the steps involved when a user shares the product with friends. In a typical product, marketing professionals might describe the virality coefficient as the number of invites sent times the conversion rate of accepting the invite.
Examining the process as a more detailed funnel reveals areas of improvement. For example, sending invites involves the user clicking a call-to-action, selecting her friends and sending through a sharing medium, such as text, email or social media post. Accepting the invite involves reading the text, email or post, choosing to investigate the product and ultimately converting to a user.
Each of those steps can be examined analytically. Optimizing the call-to-action can be one route to increasing the virality coefficient. Making access to contacts easier might increase the number of friends a user selects. Furthermore, invites sent through email might convert better than invites posted in a newsfeed, or invites sent through text might be more likely to be clicked on than invites sent through email. Even the message included with the invite could make a difference. Limitations do exist: no amount of A/B testing colors on a call-to-action button will get users to share a product they hate; however, if a product had a virality coefficient of 0.7 with users inviting on average two friends each, then it’s worth looking at how to get users to invite three friends each because then the product achieves full virality, assuming conversion stays the same.
Though cleverlayover, as a travel company, doesn’t focus on virality, I recently built a politically-themed-but-non-partisan video generating app that allows users to share videos they’ve created. This gave me a unique opportunity to examine the pipeline with real-world numbers. For this app, users create videos and send the link to friends through a variety of social platforms. Then, friends can choose to create a video of their own. Rather than invites, the sharing mechanisms here are the videos users create. After launching the app and getting some initial press, we were able to see a virality coefficient of 0.36, not high enough to go fully viral, but high enough to be encouraging.
Examining the pipeline in more detail reveals that the virality coefficient can be described as a product of the following: the number of videos each user creates, the percent of videos shared by the user, the number of views received by a shared video and the percent of new video viewers that decide to make their own video.
Note that in this sharing cycle, each user might send out more than one round of “invites” when they come back and create another video one week later, but for simplicity, our analysis assumes that users churn after their first day of app usage. The long-term retention rates for this app have yet to be proven out, so the numbers we use here reflect only a user’s initial engagement.
My friend and fellow Harvard Business School graduate Charles Hornbaker was kind enough to analyze the data for us. His analysis revealed:
|The number of videos each user creates||3|
|The percent of videos shared by the user||20%|
|Number of views received by a shared video||6|
|Percent of new video viewers that decide to make their own video||10%|
Multiplying these numbers together gives the virality coefficient of 0.36. In order to increase the coefficient, I have several options. I could try to increase the percent of videos shared by optimizing the sharing process. Making the process of sharing a video easier and a more prominent option can be a pretty standard exercise. However, at 20%, I consider the number pretty high already, and I assume the upper limit might not be that high given that users might edit and refine a video before sharing. The rough drafts are counted in the aggregate videos per user, meaning that a certain percentage of videos wouldn’t be shared anyway.
Later on, I might revisit this part of the funnel, but the number that stood out to me was the percent of viewers that decide to make their own video. A 5 percentage point increase here would be more effective than the same increase in the sharing rate: moving the percent of videos shared from 20% to 25% changes the virality coefficient to 0.45, but increasing the percent of viewers that decide to make their own video from 10% to 15% pushes the virality coefficient to 0.54. In addition, based on our initial user testing, I believe the percentage of users who would want to try the app once they’ve seen what it can do has the potential to be much higher.
There is a compelling case to focus on optimizing this particular part of the funnel before working on other marketing efforts. Right now, acquiring one user brings in 1.56 users eventually, but increasing the percent of viewers that convert into a user to 25% would make the virality coefficient 0.9. Acquiring one user becomes the equivalent of acquiring 10 users; this improves the effectiveness of each marketing dollar by a multiple of six (again, ignoring discount rates and assuming immediate churn). Instead of spending $10,000 to acquire users through paid channels, I could spend $8,000 running experiments with $2000 to acquire paid users and still come out ahead.
Not all products are shareable, but when a product has a virality coefficient of 0.6 without any incentives, it wouldn’t be a stretch to imagine the coefficient could be pushed higher through optimizing the sharing process. That growth can make marketing dollars significantly more effective. Though an increase from 0.6 to 0.8 might seem small, it actually doubles the value of each user (ignoring discount rates). Furthermore, breaking down the virality coefficient into a funnel that describes how the user shares the product can reveal that specific parts of the funnel are particularly impactful. If users typically invite one friend on average, then getting each user to send even 0.33 more invites would have the desired result of changing the virality coefficient from 0.6 to 0.8. Marketing professionals should investigate the potential to increase the virality coefficient before focusing on other channels, and they should consider applying their well-tested conversion techniques before resorting to paid incentives for sharing.