A short study of this web site reveals that a hugely important factor for success in startup companies is finding ways to acquire customers at a low cost. In the Business Models section, we looked at the perfect business model: Viral customer acquisition with good monetization. However viral growth turns out to be an elusive goal, and only a very small number of companies actually achieve true viral growth.
In 2005, I invested in a company called Tabblo (acquired by HP in 2007), and had the good fortune to work with an outstanding entrepreneur, Antonio Rodriguez. Tabblo did manage to achieve good viral growth, but around the same time YouTube was launched and managed to achieve explosive viral growth. In the process of looking at these two companies, we learnt several important things about virality. This post digs deeper into what it takes to achieve viral growth, and examines the key variables that drive viral growth.
To give you a preview of this post, what you will learn is that there are two key parameters that drive how viral growth happens, the Viral Coefficient, and the Viral Cycle Time. To fully illustrate the arguments, I have included two spreadsheet models (embedded) that you can play with interactively to see how viral growth works. There is a risk with this level of depth, that some readers will find this too technical, and if you find yourself reacting that way, may I recommend that you jump straight to the conclusion, which is under the heading Lessons Learned towards the bottom of the article.
What we want to understand in these two models, is how the population of Customers changes over time. The first model that we will build looks in a very simple way at how viral growth works in the marketing world.
The Viral Coefficient (K)
Imagine you are starting a new company that plans to acquire customers through viral growth. You have several friends that you use to become your first customers, and they in turn start inviting friends to join, and those friends start inviting friends, etc.
The model at this stage has the following inputs:
| Variable Name | Description | Example Value |
| Custs(0) | Initial set of Customers | 10 |
| i | No of invites sent out be each new customer | 10 |
| conv% | The percentage of invites that convert into customers | 20% |
The first thing that we need to calculate is the number of new customers that each existing customer is able to successfully convert. This turns out to be an extremely important variable, and is known as the Viral Coefficient. The formula to calculate the viral coefficient is pretty simple: multiply the number of invitations by the conversion rate.
| K | Viral Coefficient | K = i * conv% |
Now lets take a look at how K affects customer growth as we go through the first cycle of viral “infection”. Our initial 10 customers will each send out 10 invitations, and successfully convert 20% of those (i.e. 2 new customers each). So the total customers after the first cycle will be equal to the starting 10, plus the new 20, which equals 30.
To fully understand the model, it’s useful to look at the second, and subsequent, cycles of growth. In the model above, only the new customers that were added in the prior cycle send out invitations. This is because it is highly unlikely that the entire population will continue to send out invitations every cycle. Every time I have looked at other blog articles or formula for Viral Growth, they appear to have gotten this part of the calculation wrong.
Understanding the impact of the Viral Coefficient
Now that we have the model built, we can play with the variables to see what effect they have. In the spreadsheet above, go to cell B6, and change the Conversion rate for invites (conv%) to 5%. This will make the Viral Coefficient less than 1. Now look at what that did to your population growth. Instead of continuing to grow, it grows to 20 people, and then stops.
What this tell us is very interesting:
The Viral Coefficient must be greater than 1 to have viral growth.
Further playing with the spreadsheet will show that increasing the viral coefficient by increasing the number of invites sent out, or the conversion rate, has a nice impact on how the population grows. Try this out by changing cells B5 and B6 in the model above. Later on we will talk about how to design your application to maximize these values.
The Second Important Variable: Viral Cycle Time
Antonio Rodriguez built Tabblo around the same time that YouTube was built. Both sites were viral, but while Tabblo was reasonably successful, YouTube exploded and amassed users at a rate that had not been seen before on the Internet. What was going on here?
To answer this question, we have to look at the Viral Cycle Time,(which we will refer to in formulas as “ct”).
The full viral cycle involves several steps that work in a loop:
The Viral Cycle Time is the time that it takes for this cycle to complete.
In YouTube’s case the Viral Cycle Time was extremely short: a user would come to the site, see a funny video, and immediately send the link on to their friends. Tabblo, on the other hand, had a much longer cycle time. A customer would post some photos on the site and invite their friends. The friends might see the photos on Tabblo, and like the experience and decide that they would use the site the next time they took photos they wanted to share. However, that is where the problem came in: it could take months before they next took photos, and decided to share them.
Later on this post, we will talk about how to optimize Viral Cycle Time – (see Lessons Learnt).
How Viral Cycle Time affects growth
To model Viral Cycle Time’s effect on growth, I searched the web, high and low, looking for a pre-defined formula. To my great surprise, there was no formula that I could find that correctly calculated customer growth, and showed the impact of Viral Cycle Time. What was also surprising, was that I did find several blogs showing formulae for viral growth, but in every case, they appeared to make the same mistake, which was assuming that the entire customer base would continue sending out invitations for every cycle. So I collaborated with my partner, Stan Reiss, who turns out to be a whole lot smarter than I am, and he helped me develop the fomulae that are used in the more sophisticated model for viral growth below:
A quick look at the table that shows the effect of varying the Viral Cycle Time shows that customer growth is dramatically affected by a shorter cycle time. For example, after 20 days with a cycle time of two days, you will have 20,470 users, but if you halved that cycle time to one day, you would have over 20 million users! It is logical that it would be better to have more cycles occur, but it is less obvious just how much better. A quick look at the formula tells the whole story. The Viral Coefficient K is raised to the power of t/ct, so reducing ct has a far more powerful effect than increasing K.
This explains why YouTube exploded at a faster rate than ever seen before.
Lessons Learned
There are a large number of interesting lessons to learn from the above models:
- Unless you have a Viral Coefficient that is greater than 1, you will not have true viral growth.
- The most important factor to increasing growth is not the Viral Coefficient, but the Viral Cycle Time (ct) which should be made as short as possible. This will have a dramatic effect on growth.
- The second most important area to focus is the Viral Coefficient (K). Anything that you can do to increase the number of invitations sent out, and the conversion rate, will have a significant effect on growth.
In addition to the above lessons that come from the model, there are some other important observations:
- Virality is not a marketing strategy that can be executed by the marketing department. It has to be built into your product right from the beginning. This is a function that needs to be thought through by the product designers and developed by the engineers.
- The most viral products are those that only work if they are shared. For example, Skype only worked in the early days if you got your friends on to Skype, otherwise you had no way to call them. If you have an application today, think about how you can make it social, where it would work better by sharing data with friends/co-workers. That provides a great incentive for customers to invite their friends/colleagues to use the application.
- To make the Viral Cycle Time as short as possible, we can apply the same thought process that we use in Building a Sales and Marketing Machine, where we look at what are the customers motivations and negative reactions as they flow through the viral cycle. For example, when I reach the stage where I have to enter my friends addresses, I will not bother to do very many if I have to look them up in another program, and copy and paste them one-by-one into the browser. You can solve this problem by providing me with Facebook Connect integration to invite my Facebook friends, and an adapter to import my email contacts. (Check out the “Share This” button at the bottom of this post as an example of how this can be done.) Getting at email contacts is easy with web mail clients like GMail, etc. – but harder with Outlook. However viral products like LinkedIn have created Outlook adapters that you can download. It is also feasible to get at that information via Outlook Web Access (OWA) provided you can deal with the security concerns.You should also be looking for ways to encourage customers to invite people at various junctures in their use of the application. And of course, you should be asking yourself the question: is the value proposition of your product really that compelling that your customers will want to share it with others?
Another great way to increase virality is to incent customers with a reward for every customer they successfully convert. Since this can result in an individual feeling guilty that they are making money off their friends, the best way to do this is to also provide the friend that is receiving the invitation with an equal incentive. Now your customer will feel like they are doing their friends a favor.
- Consider leveraging viral platforms such as Facebook, which have built in social features to let friends know what apps you are using. The wall, and status updates provide a great way for their friends see your app.
- Use A/B testing to figure out which approaches and creative presentations are getting you the highest conversion rates.
- If you are successful in creating a viral model with very short cycle times, watch out for what can happen. Several companies that have been lucky enough to achieve this have been shocked by the enormous need to scale server capacity. Fortunately with cloud computing offerings such as Amazon EC2 and S3, it is easier than in the past to scale on demand.
Hybrid Viral Models
Many entrepreneurs reading this post will realize that they may not have the means to achieve true viral growth (where they have a Viral Coefficient of greater than 1). Rather than giving up, it is worth considering a hybrid viral model. In the hybrid viral model, you make up for the shortfall in customers by acquiring those through some other means such as paid search, or SEO.
Model Limitations
The model above is pretty simplistic and does not take into consideration several real world phenomena:
- What happens when you grow so fast that you start to saturate the population. This has happened to several Facebook app developers. They experience very rapid growth, and then suddenly the growth dies. Andrew Chen has written a great blog post about this: Facebook viral marketing: When and why do apps “jump the shark?”. (Side note: I don’t believe that the equation that Andrew puts forward for simple viral growth is correct, as it assumes that the entire population will continue sending out invitations at each viral cycle. However his work on saturation of the population is very relevant for highly successful viral apps.) In case you are interested in where the term “jump the shark” came from check this out: Wikipedia: Jumping the shark.
- What happens if you have attrition in your customer base over time. An easy way to extend the model to take this into consideration would be to add a variable to model Attrition Rate as a percentage of the entire installed base at each cycle, and simply subract this from the total population at each cycle. This topic is nicely covered in this blog post by Andrew Chen: Is your website a leaky bucket? 4 scenarios for user retention.
- The customers that you have may send out more than one set of invitations beyond the initial set.
- etc.
Further Resources
Check out Andrew Chen’s blog, as he has written extensively on the subject of Viral Growth. For example, here is one great example: What’s your viral loop? Understanding the engine of adoption.
Acknowledgements and Thanks
My thanks to Antonio Rodriguez, the founder of Tabblo, who got me started on thinking about this topic several years ago. Also to Andrew Chen, whose writings on this topic are excellent. And to my partner Stan Reiss, who took my simple logic and turned it into an elegant mathematical formula.
Counting your entire user base in ongoing virality is certainly an issue, but especially in social games you do not see a viral coefficient of 0 for ongoing engaged users. You typically have two primary cycles, the install (early user) and engagement (ongoing user) viral rates, and modeling properly means keeping track of both. Having a product that can get that first invite quickly (YouTube) is critically important, but one shouldn't overlook the power of continued reason to invite over time – another area where YouTube excelled actually.
But this is an incredibly important point to bring to light, as it's really not talked about enough. It's so bad that I commonly see entrepreneurs not thinking about viral cycle time, and therefore automatically setting it to one month. Why? Simply because they are charting out the next year of growth and their excel spreadsheet is simply taking the “new users” from the month previous and multiplying by the viral coefficient.
Hey David,
Great thoughts. I appreciate your effort to explain your ideas from a mathematical/statistical perspective and respect your insistence in refusing to “dumb down” (for lack of a better term, as an individual's technical ignorance should not by itself imply low intelligence) your content.
I have one minor criticism though, and as a math major I feel obligated to point out what I believe to be an error in terminology: “This turns out to be an extremely important *variable*, and is known as the Viral *Coefficient*.” If my knowledge of mathematics should be trusted, please believe me when I remind you that the realms of variables and coefficients are indeed mutually exclusive, i.e. no one term can simultaneously exist in both categories. Though the value of the 'Viral Coefficient' differs from one use of your equation to the next, this 'variability' does not make the term a variable. I have a strong feeling this was just an unintentional slip-up, as you've named the term 'the Viral Coefficient' in the first place, but I'm confident most anyone with the technical nature you display would appreciate being informed of such a slip-up; such individuals often have an extreme distaste for…ahem…'technical incorrectness'…haha.
David. Fantastic post on this topic. I used to be the CEO of Going.com and we lived this daily. The incorporation of these steps helped accelerate are growth. The timeliness/viral cycle time component you mentioned i've never seen discussed before but it is dead on the mark/game changer. Evan
Lou, you raise an excellent point. The Viral Coefficient is definitely a variable, and one that you can effect by the actions you take in building your product. Unfortunately that term came before my article, and is already in widespread use. So rather than change the name, I stuck with the industry term. Whoever invented the term should may not have realized the ability to change the coefficient.
David,
Very nice synthesis of what we learned— and spot on with respect to the importance of cycle time.
I would caution entrepreneurs looking at this post to consider only one additional factor (which turned out to be a key difference between YouTube and Tabblo): the appropriate definition of an “infected user,” or as you call them, customers. In the case of YouTube, a customer was just someone who forwarded a link— and not as was the case with Tabblo, someone who created a video and uploaded it. As such the cycle time was indeed *much* shorter. You'd see the funny Will Ferrell video and immediately forward it to your office mates.
By contrast if you measured the cycle time as invite-> video uploader I think you'd find that YouTube's viral equation looks a whole lot different from what most people think.
If your money is coming from the size of you “creator” audience (be it Tabblo, Hotmail, or PayPal), it is really important to be measuring the appropriate type of infection in your virality metrics. For instance, in our case we knew there was a direct correlation between content creators and buying habits so that was the key loop to measure. In the case of Consumer Internet ad businesses (like YouTube), the eyeballs on the page are all that matter.
Again, great post though— I wish I'd had this in 2006!
David,
. I’m sure if Tali were around, she would have put her Harvard math degree to use!
Glad you suckered Stan into helping you with the equations
Great post – I love the takeaway that minimizing cycle time can be more powerful than maximizing the oft-discussed “viral coefficient.”
If you’re taking requests from readers, I’d love to see a post discussing different types of virality. You touched on this in your point about Skype, but I believe it warrants further discussion, since the term “virality” is thrown around without much distinction around type of virality.
I use these 3 non-mutually exclusive categories:
–“necessarily viral”: I need to invite friends in order to derive value from the product (e.g, Skype, Facebook)
–“inherently viral”: My use of the product exposes it to friends (e.g., Hotmail, any photo/video sharing site)
–“word of mouth viral:” The product is so impressive/cool that I’ll tell my friends about it, either through online or offline channels (e.g., Google, Hulu, Gilt)
I should note, though, that I believe the advent of the “stream” (ie, Facebook newsfeed and Twitter) has blurred the lines between “inherently viral” and “word of mouth viral.” An auto-tweet feature can make my friends automatically aware of a product I use, even if my actual use of the product didn’t touch them. (Foursquare’s auto-tweet is a beautifully executed example of this.)
I believe “necessarily viral” is the most powerful type (since it usually implies a network effect), but it’s also the hardest to pull off – if my friends aren’t interested in using the product, and I can only derive value when they’re on it, I’ll abandon it quickly.
In any case, I’d love to hear what categories you use, and what tips entrepreneurs should keep in mind as they build products aiming to fit in one of these categories.
Great article about viral marketing for startups http://bit.ly/629zU9
David – great post. I learned about the viral coefficient from Ted Dintersmith at CRV 10 years ago when starting Upromise and it has stayed with me ever since, but your explanation was far more sophisticated and complete (no surprise) than my simple rule of thumb: “make the viral coefficient > 1 to succeed in driving sustainable virality”. It's wonderful to have you contributing to the public dialog as entrepreneurs can learn so much from your wisdom and experience (never mind your friendly VCs!). Keep 'em coming!
Viral Marketing Math … interesting
Chris, thanks for the feedback. Your separation of the different viral types is very useful, and a nice topic for a follow on post.
Great Post .. I am just a newbie..but with high intuition levels of success…
Great Post .. I am just a newbie..but with high intuition levels of success…
David,
Great post! I've been thinking about and working on optimizing viral loops since 2004 when I led product management at Friendster. I'm familiar with Andrew Chen's posts and your post definitely adds to the body of knowledge.
Great point about most models assuming “old” users continue to invite friends at the same rate as new users. Nabeel raises a good counterpoint, too, but if, for the sake of avoiding an overly complex model, I had to choose between assuming that either all or none of the “old” users invite, I'd probably choose none as you did. Although, I guess it wouldn't be too complicated to treat the “old” users as a second segment and apply a different (much lower) viral coefficient to that segment if the nature of your business justified that.
Thanks also for highlighting the importance of cycle time and providing that equation. In other good online discussions I've seen on viral growth, I felt that there was a “time element” missing and your cycle time discussion addresses much of that.
I like your equation incorporating cycle time, but I think the picture is (unfortunately) more complicated than the equation indicates, since the parameters themselves vary over time (and don't have static values as the equation implies). So a product's viral coefficient and cycle time vary over time. This variation over time occurs for many reasons: changes in your invitation process, changes on your registration page, spikes due to positive press, launching new features, server downtime, inadvertent introduction of bugs in the viral loop, increasing saturation of the population, etc.
So, the “K” and the “ct” are themselves functions of time. Rather than a single equation that predicts the # of customers at time t from the # of customers at time 0, you have a recursive equation where the # of customers at time t is a function the # of customers at time t-1 as well as the values of K and ct from t-1.
Even with a very successful viral application, I don't think the value of K stays way above 1 for long periods of time. If it did, the equation shows that the application would saturate the population quickly and have it's growth rate diminish to zero (or the intrinsic growth rate of the population). I like to think of the periods of time when your K is above 1 as “growth spurts”, and you obviously aim to have many and longer growth spurts. Related to this, it's been pointed out that even non-viral growth is beneficial growth, so trying to maximize your K even if you don't get it above 1 is still valuable.
In some of the presentations I've given, I discuss viral growth, optimizing the viral loop, and an example from Friendster. My slides are posted online at http://www.olsensolutions.com/speaking
Thanks again for your rigorous, well-written, thought-provoking post!
Dan Olsen
CEO & Founder, YourVersion
http://www.yourversion.com
Great article, insight and math to back-up the viral marketing message.
I really like this as it will help me help sales managers, marketing managers and CEO's understand that saying the words viral marketing in the context of how they are going to market is very different from actually understanding the model and the metrics for success.
I also agree strongly that viral marketability is designed into the product, it's not something you can bolt onto and existing product or service easily….unless what is already created has such intrinsic value where sharing links is all that is required to expose it.
Mark
Dan, your points are one hundred percent on target, and I would love to build a more accurate formula for this over time. Thanks for adding in your views here. They enrich the discussion and add value.
I have also enjoyed looking at your slides online.
Best, David
Great Posting… here other useful Online Marketing “Mathematical” Formulas http://syndikomm.com/blog/?p=216
great analysis chris.
i'd also point out that “necessarily viral” usually implies a longer CT (because you need to wait for your friends to join for it to be useful). So while it may be more powerful in building the coefficient, but based on david's analysis that may not be the best tradeoff…
great analysis chris.
i'd also point out that “necessarily viral” usually implies a longer CT (because you need to wait for your friends to join for it to be useful). So while it may be more powerful in building the coefficient, but based on david's analysis that may not be the best tradeoff…
Can you provide a derivation of your growth formula?
nevermind, it's just a geometric series. got it.