Our post on the Optimal startup burn rate and the Kelly criterion was extremely popular as it provided an actual model to help people think through startup burn rates. Hopefully a post with an actual model to help people think through viral marketing will be useful, too.
First, go here and download “viral_tuning.xls.” Click the down arrow sign on the iDisk folder (you will need Safari on Mac or IE on Windows as Firefox does not seem to be playing nice with iDisk), which will download the file for you. Make sure that you have enabled macros (from the Excel menu: Tools -> Macro -> Security; the latest version of Excel for Mac does not support Macros). For those of you who don’t have Excel, you can find a neutered version of the model on Zoho here (no Macros).
The BLUE cells are inputs (can be edited). The BLACK cells are formulas (don’t touch them, unless you are checking the formulas for accuracy). Play around with the BLUE cells to see what it takes for a service to be viral. Then click one of the gray “SOLVE” buttons above any of the steps in order to optimize virality for just that step [assumes the other variables are held at a user-defined constant]. In this example, I have taken the registration flow of a typical social networking site.
Step 1: A user shows up at your site. What percent complete registration?
Step 2: You offer the user the opportunity to invite friends. What percent use the import tool (tool to grab contacts from Gmail, Yahoo! Mail, Hotmail, whatever social network, etc.)?
Step 3: How many friends on average do users invite? The emails go out to that many users and then you go back to Step 1.
The purpose of this spreadsheet is to help you think through just how hard it is to make a viral product. You should lay out the flow a user goes through when interacting with your product in steps just as I have done here. Try to understand the key drivers that encourage people to get their friends to visit your site — and optimize around those by developing and testing hypotheses in order to experience viral growth. The steps need not happen online (the same underlying principle applies for offline retail), but when the entire flow is online it’s obviously easier to measure.
The flow outlined in this model was key to building a number of social products on the web today — but consumers are tired of this strategy. Still, it should help you think through what virality means and just how hard it is to achieve a truly viral product.
5 Comments
June 29, 2008 at 6:27 am
For some reason Firefox 3.0 on Macintosh cannot access my public iDisk. Try Safari on Mac or IE on Windows. Sorry about that.
June 29, 2008 at 6:29 am
Lesson learned: Viral products don’t require registration
June 29, 2008 at 8:20 am
You didn’t factor user churn rate. Depending on the product, churn rate could be enough to dampen the growth from becoming viral.
June 30, 2008 at 2:55 pm
Eric,
Not so sure about viral products not requiring registration. Viral videos, sure. Viral jokes, hell yeah. But actual products? More likely than not they’ve required registration. 280North may be an exception, as may be Posterous, but Tumblr was one of those things with a very easy registration process that I signed up for and well, never quite used. I may have just done the same thing with Posterous, though I thought it was cool, I don’t think I’ll be a devotee.
July 1, 2008 at 9:41 pm
My experience suggests that the process of optimizing “virality” requires an engineering mindset *combined* with product sensibility. For example, the mathematics will suggest that driving email invitations sent will drive virality, but there is a danger associated with getting overly aggressive with sending invitation emails. The danger is that your company will be considered “spammers”.
I believe the best way to ensure you are optimizing the system while not getting overly aggressive is to measure and A/B test just about EVERYTHING, and always take the end users point of view. For example, you not only want to measure how many people import their address book, but also how many of them send the invite to everyone in their address book, which is likely a signal of accidental inviting, and should be considered “spam”. You will also want to add just about every variable you can think of to the mix in order to test. For example, you should be A/B testing every invitation that gets sent out, and understanding the open rates, the click through rates, the registration rates, deliverability rates (does it land in the spam filter), etc.
What most people don’t realize is that these “viral systems” each have “subsystems” that should be optimized and that the optimization of each subsystem (subject line on email, landing page, etc.) can have large impacts on the overall system performance.
Finally, I am a big believer that every web company should view their site as a “system” that can be measured and optimized, and therefore I am constantly recommending the book “Business Dynamics: Systems Thinking and Modeling for a Complex World” by John Sterman. It is expensive book, but it will change the way you think about the world, and this style of thinking can make the difference between success and failure for a start-up.
Leave a Reply