z on life

When you’re creating a web property that you intend to ‘go viral’ with, whether it be a Facebook app or full-blown destination site, you’re going to have to meticulously track the growth to understand how you’re doing, where improvements can be made, and so forth. Tracking the growth correctly is actually very hard and even the most rigorous analytics can sometimes fail to reveal the true growth rate.

There is one easy trap to fall into, and it’s one that I just did, so I thought it would be good to talk about. The trap is in not understanding that you may be locally viral, where being ‘local’ is defined relative to some super-specific demographic, or user-behavioral context. In English? You may only be viral for some small set of or narrow type of users that will quickly adopt whatever it is that you’re making. However, if that demographic isn’t sufficiently sizable, you may burn through all of those people and be left with no new users to acquire.

The worst part about this is that, early on, you may see excellent growth. My roommate/coworker Aman Gupta and I recently released the ‘Hotter Than’ Facebook app as a pet project. It took a few hours to build and was a fun experiment (it’s not by any means our day-job). When we opened the floodgates, we saw in the first week a doubling day over day. If you study viral, recursion, and logistic growth from a mathematical perspective, as long as the remaining population of potential users is large compared to the amount of current users, your growth per new user on day N should be a good indicator of growth per new user on day N+1.

So have we continued, then, to double day over day? No, in fact growth has slowed to basically nothing. Why is this? Well, from looking through the raw data logs, we can see that during our growth phase we had spikes of users in very specific demographics. Moreover, it is fun to look at who was actually sending out invites. It turns out invites were sent mainly by a few categories of users: older, international, small friend networks. Now certainly, there are plenty of older people and international people on Facebook, many with small friend networks, so shouldn’t we have spread like wildfire through these huge demographic categories?

Interestingly, no, and here’s why. We thought a bit deeper about this and it turns out these were all indicators of being a newbie Facebook user. The average Facebook power user is something close to a United States college student with tons of friends*. The farther outside of this age, location, and dense friend networks you get, the less savvy the users likely are. We were way outside of this.

So what does that tell us? What we actually probably got were users easy to trick into viral invite processes, and thus, our invite scheme was not effective, logical, and enticing enough to hit the mainstream. Thus, we prayed on a few ‘fish-in-a-barrel’ types for whom this was likely a very early application install in their personal Facebook lifecycles. These people were most likely not aware of the ramifications of spamming friends and were thus much easier to convince. We burned out those users quickly and eventually there were no new accessible users who would succumb to the virality. After day 7 or 8, the growth halted and the usage flat lined.

Moral of the story, you need to understand exactly why people are taking the bare and sending your invite. Is it appropriate only to a certain culture, age, or interest group? Or can it be generalized to the mainstream? Hopefully, when all is said and done you’re not just tricking newbie users like we were. If you cannot generalize to the average user, then you’re early growth may not be any indicator of future growth so don’t get excited until you have a diverse set of users who have proven acquirable across enough of the spectrum to make the pool of potentials huge enough for a macro-success. It’s crucial to know where are you viral? Is it local, or is it global?

*note: This definition of an FB power user isn’t backed up by any data and is more by ‘feel’. Thus, the reason for our failure is still a hypothesis, not proven fact. The point here isn’t to know exactly what users we attracted, but to understand that those initial users were not normal and their affinity for our viral hooks was not generalizable over the whole population.