Sonibyte Podcasting Blog

Understanding Podcast traffic statistics--Part II, File Downloads

In our last post, we gave a brief overview of how podcasts worked and defined some terms.  Now, it is time to get into the dirty details of looking at podcast traffic logs.

The first place to look is our file download logs.  The podcast sound files for our modest in-house blogs were downloaded 2,334 times on January 16th.  That seems pretty impressive, as it suggests we had an audience of more than 60,000 people per month.  If we group our results by IP address, we can see that  1,949 distinct IP addressed downloaded from our site.  This is because a number of IP addresses downloaded files from us, more than once.  311 of our 1949 IP addresses download more than one copy of the same file.  Their 696 downloads are about 30% of our 2,334 total downloads.

Why were particular users (IP addresses) downloading the same file over and over?  For instance, IP address 68.46.125.236 turns out to be Comcast--a major internet provider.  (It is easy to look up IP addresses using reverse DNS lookup from sites such as this.)  The 7 downloads of our reading of the poem "Dreams By Langston Hughes" could have been from a single user hitting the download button 7 times.  However, it is more likely that seven of our listeners are using computers served by Comcast.  When these users think they are connecting to us, they are actually connecting to a Comcast server that feeds their request through a different server and to us.

Note that Comcast also shows up also as IP address 76.27.59.224 and several other addresses.  Major Internet Service Providers (ISPs) such as Comcast, Verizon, and Speakeasy feed their traffic through many different IP addresses--as do portals such as Yahoo!, AOL, and MSN.

We also see multiple downloads from IP addresses such as 202.108.23.56, which was responsible for a total of 80 downloads, spread over 21 different files.  A reverse DNS on this address tells us only that the source is in China.  We can assume that these requests are coming from a Chinese internet provider similar to Comcast--one who has not fully disclosed its ownership of this server.  Since our "house" podcasts include English readings of the Bible, English readings of the Koran, and English readings of both classic and contemporary poetry, I'd guess that a lot of Chinese people are listening to them as part learning English.  It is nice to know that we are doing some good for the world!  By the way, lest you think that only Chinese people want to learn English, we also had at least 3 downloads from 83.202.100.211, which is Wanadoo, in France.

Of our 2,334 downloads, all but 41 were done by "agents."  These agents are also called "podcatchers."  There are twenty podcatchers listed on PodcatcherMatrix--and this nicely organized site helps you compare their features and functions.  The main function of a podcatcher is to watch the RSS feed of a podcast for new episodes.  When a new episode appears, the podcatcher grabs it and downloads it to the user's computer.  They can also provide a license or password authorization (when it is required), allow searching for new podcast feeds, and manage sychronizing the list of episodes on a user's computer with the list of episodes on her or his iPod or other portable player.

iTunes was responsible for more than 50% of the downloads on our site that were handled by agents.  Mozilla (the agent name used by Firefox browsers) added another 13%, followed by another 3% from download agents such as Doppler, Juice, and NS Player.  Unfortunately, about 34% of our downloads come from what are commonly referred to as "bots."  These are programs that scan the Internet looking for useful and interesting information.  When they find "content" such as our sound file, they greedily download it and store it away for later review and digestion by their owners.  Some of these bot downloads probably are from legitimate search firms or podcast directories.  The rest...well no one really knows where all the data absorbed by bots, really goes.

(By the way, remember that the RSS document was 64% iTunes?  Once you subtract the bots from the file download log, we get 77% of downloads via iTunes.  Our RSS log also gets bot requests--about 10% of our requests come from them.  Take these out and we get an adjusted percentage of iTunes RSS request of about 70%.  This is close enough to make us comfortable.)

If we subtract the 803 downloads from bots, we are left with a net of 1,531 downloads to listeners.  Of course, some of the multiple downloads may be due to retries (when a download fails and a user tries a second time).  I have seen an estimate on Jason VanOrden's blog that retries account for between 5% and 20% of all downloads.  I think I'll use the 5% number in our situation, and claim that we had 1,531 x 95% = 1,454 downloads on January 16th.

Short of getting users to install some kind of tracking software (possible--Volomedia has a pretty cool one), there is no way to know how many of these downloads were actually listened to.  But, no one knows how many people actually read newspapers or listen to programs that are broadcast on radio or TV.  Still, we'd like to know a bit more about our listeners, including how they found our podcast and whether or not they have subscribed to it.  We'll try to get more data on this from looking at our RSS request log, in my next post.

Understanding Podcast traffic--Part I, Some Basics

It is hard to make sense of the data we see on podcast traffic.  This is the first of a series of four posts, that will offer as much help as we can, on understanding and interpreting podcast traffic statistics.

Like most other podcast publishers, we look routinely at our file download logs and the records of how many people request our RSS feeds and those of our major publisher customers.  However, we often end up unable to answer fairly simple questions such as, "How many people listen to our podcasts" and "How many people subscribe to our podcasts?"

Our research is hampered partly by the way podcasts are delivered and partly because of the general anonymity and obscurity of Web interactions.  In the past, podcasting was a sideline and the numbers didn't matter that much.  Nowadays, many podcasters are starting to monetize their feeds by inserting ads into them.  The prices paid for these ads will be driven by how many people listen to these messages, and that increases the value of well-documented podcast traffic statistics.

We host feeds for a number of major magazines.  We can't/don't disclose their statistics.  Fortunately, we also publish several podcasts of our own, "Poem of the Day," "Daily Bible Verse," and "Daily Koran."  Therefore, we can and will share these numbers.

To keep things as simple as possible, I decided to focus on a single day in January.  I chose January 16th of this year, because it is not a holiday, it is not part of a holiday week, it is not a weekend day, and no huge national disaster or international event occurred that day.  (BTW, it also happened to be my sister's birthday!)  If you are analyzing your own logs, be sure to watch out for factors like the day of the week or the time of the year, as they will definitely affect your traffic patterns.

Before I start, I'm going to give an overview of how podcasts are distributed--just in case you forgot (or never bothered to find out!).  A podcast is a series of sound file (we are ignoring video, for the moment) on a particular topic.  Examples are radio shows (one of the first things distributed via podcast), comedy sketches, discussions of new gadgets, articles on news or politics, poetry readings, serial novels, Spanish lessons, and so on.  Most podcast listeners download podcast sound files to their computers.  They then either listen to the podcast episode on their computer or transfer it to their personal listening device (e.g., their iPod).  (For a great short description of how podcasts work, please look at this article from the University of Leeds.  For an even shorter version, see this post on podcast411.com--which is a great all around site for info on podcasting.)

When a user downloads our file, his or her computer has to establish a connection with our file server and request the file.  As a result, we get to see the user's IP address and what type of downloading software he or she is using.  (For instance, about 50% of those downloading our files on January 16th were using iTunes and about 13% were using Mozilla.  About 64% of those requesting our RSS documents were using iTunes--a difference we will talk about, later.)

How does a potential listener find out that we have a juicy sound file waiting to be listened to?  There are two paths to this knowledge:

1.  Some listeners could remember where we keep our files or know we have a Web page that links to our files.   Despite hard looking, I can't find much evidence that these people exist.
2.  Most (or all) listeners discover our files via our RSS document or "feed."  Our RSS document describes our podcast in both general and technical terms.  It says what category our podcast belongs in, supplies a text summary of our product, and shows where a "cover" photo resides.  It also lists the most recent episodes in our series of podcasts.  Most people examine RSS documents with an RSS "reader."  The best known of these for podcasting is Apple's iTunes reader.  (We will talk more about readers and podcast directories in a later post.)

After a user looks at our feed, he or she can choose to download one or more of the files listed in the feed.  If the user likes our content, he or she can "subscribe" to the RSS feed.  Once the user tells his or her reader that he or she wants to subscribe to a podcast, the reader will faithfully and regularly check our RSS feed document to see if we have posted any new episodes for our podcast.  When we update our feed by adding a new sound file, the reader will notice this and automatically download the new file to a subscribing user's computer.

You can see a page of podcast feeds for both our podcasts and those of our customers on our site at http://www.sonibyte.com/rssfeed and inspect a page that holds our Daily Koran feeds at http://www.sonibyte.com/rssfeed/35.xml.  For techie readers I'll mention that these pages are actually a transformation of our RSS document into an HTML page using some Java script and an XSL translation template.

Does this sound simple, so far?  Just wait.  You will see in our next three posts that it gets pretty difficult to decipher the relationships between listeners, subscribers, file downloads, and RSS document requests.

Is there an 80-20 Rule of Podcasting (a Pareto distribution)?

The 80-20 rule is also called the Pareto principle.  It states that "80% of the effects come from 20% of the causes."  Does this rule apply to Podcasting?

Our clients are major publishers.  Their writers crank out weekly or monthly columns and newsletters of consistently high quality, that focus well on their readers' interests.  When these articles are converted into podcasts, using Sonibyte's voiceover system, should they all garner about the same amount of attention (as measured by total downloads).  Or, will some podcasts turn out to be home runs?

We studied a set of 185 podcasts created by a selection of our clients during the second and third quarters of 2007.  These podcasts came from five different publications and included between six and twenty podcasts from a single "feed."  (Most podcast users subscribe to a feed of podcasts that are produced by a particular author on a particular subject.  They then automatically receive each new podcast article, when it is published.)

We counted the number of downloads each podcast received during the 180 days after it was published.  We then calculated the average number of downloads that each feed generated and examined how each podcast did relative to its average.  For instance, say the average number of downloads for a podcast feed over 180 days was 10,000.  If a particular podcast in this series had 15,000 downloads, we would give it a score of 1.5.  If it had 3,000 downloads, it would get a score of 0.3.

When we broke apart the downloads in this manner, we found they fell into three distinct groups.  About 10% of the time, a podcast in particular feed would produce less than 50% of the normal number of downloads.  About 5% of the time, a podcast would produce more than 50% as many downloads as normal.  (The best performing cast produced more than four times the normal number of downloads.)  However, most of the downloads--about 75% of the total--came from the 85% of podcasts that had ordinary performance.

This analysis suggests that the best strategy for podcasting (to stay with the baseball analogy) is "small ball."  Don't hope that a podcast will be a home run, and swing away.  Instead, develop a core franchise for your articles, and hope to steadily increase your number of subscribers and average number of downloads.  We will look more at how to build subscribers, in a future post.