Thoughts on Pocket

I got an interesting email from Pocket webservice the other day. Pocket is a "read it later" service that integrates with your browser and/or mobile device to allow you to easily save documents for later reading. It has some nice built in features for reflowing common article layouts to read them easier and has pretty good cross device synchronization. According to Pocket, I was in the top 5% of their user base because of how much I read.

I am amazed. I don't read that much, especially not on pocket. I read maybe a few articles a week. My first thought was that maybe I'm not actually in the top 5%. Maybe pocket sent that to the top, say, 10% of their users. I wouldn't put it past them; it feels relatively obvious that the purpose of the email was to encourage me to continue engaging with pockets service. Unfortunately, there is no real way to check this theory, so I was forced to continue trying to understand how this could have happened.

Pocket counts itself as having a lot of users 17 million if you trust their website. That is a lot of users. Even if I fall in the top 5% of that list, I am still not really that close to the top of the list, there are still thousands of users ahead of me. In that way perhaps most real users of pocket are in the 5%.

The amount that different users read on pocket follows a power law. I have no data for this, but, based on my experience, as well as the conventional wisdom of the online HCI community I can be pretty sure I am right. In almost every website I heard of, user behavior has followed a power law distribution. In wikipedia for example, we have the 90-9-1 rule in which 10% of the workers do almost all of the work with 90% of the users simply "lurking". Of the 10% that works only the top 10% of them (1% of the overall population) actually creates new content, with the rest simply improving the existing content. In pocket this likely manifests as the top 10% of users doing 90% of the reading.

What this means as far as my 5% is that I might only need to be an "active" user to really hit the top ranks of pocket. I read a few pocket things every week, but if I get to compete with the 90% of users who signed up, used it for a week or two and then forgot about it, then of course I can be highly ranked. I guess, in this way thank god for power laws. Because of power laws, keeping along and being an active user of a service actually does put you in a special top-tier cohort of other actually dedicated users.

Unfortunately, none of this is really an answer. I can't actually know if their reporting is honest or not or how they did their math. Maybe they actually only considered "active" users and I still ranked highly. Maybe most people got this email. I cant really know, but maybe my thoughts might entertain, and if not they will certainly fill out my blog post for the week.

On that subject, I have one or two other thoughts I wanted to share. Everyone I know who uses pocket says that they don't read most of what goes into their queue. I think that part of the reason behind this is because there are two processes at play here. The first is the process by which you come across maybe interesting content. This process happens at random, with really no respect for what you are doing, and sometime you don't have the time, or attention to spare to an article. That's why services like pocket exist.

The second process is the process by which you decide you have some free time to read. In my opinion these are two relatively independent processes. Sure, some times I find something, add it to my reading list, and then think of it when I get home. But for really interesting things like that I wouldn't need a reading list.

For most things on my list, there are two independent Daniels. The first Daniel found it interesting, but not interesting enough to read now. The second Daniel was bored and is looking for something to read. The first Daniel gets the satisfaction of not saying "no" to a maybe interesting read without having to actually read it. The second Daniel gets the satisfaction of reading something interesting selected from the curated list that Daniel 1 put together. Its a pretty great scheme.

There are, however, issues with the two processes being independent. So there's a result in queuing theory, which is the study of queues. That results says that when you have a queue that is being fed at some rate, and independently emptied at another rate you can compute the average length of the list based on those two rates. So If I just measured how often Daniel 1 added books and how often Daniel 2 read books I could know how large I should expect my reading list to be. The problem is that this equation is only defined when Daniel 1 is strictly less active than Daniel 2. This makes some sense, but it also predicts that for most people (from what I can tell) they will add more than they will read and their reading list will grow without bounds. This feels like a problem to me. I don't actually know how to solve it, but it might be an interesting research idea to explore later.

One interesting aspect of the result I brought up is that the queue doesn't have a defined expected average length if the rate items are put in is the same as the rate items are taken out. I find this fascinating. If I read items on pocket just as frequently as I add items to pocket, even then I will not be able to keep my queue from growing (without deliberate non random action). I didn't really bother working through the math to understand what type of undefined we are talking about here. As far as I understand it simply means that there isn't stable equilibrium to how many items you should expect, which probably means you should expect it to grow without limit. I find this really interesting.