RonAmok!

The adventures of an analog engineer and digital storyteller who studies emerging networks and their impact on the great game of business.
Jun 12, 2014

Those who follow this blog know that I like all sorts of data. Whether it be tracking the tweets of a company dealing with a public relations issue, or analyzing the social media posts of a public utility company during a blackout, I love finding stories in data.

For the past three years, I’ve been collecting social media data for the top 13 contestants on the television show, American Idol. I wanted to know if social media data could predict who was going home every week.

I started by capturing both Facebook and Twitter data. It didn’t take too long to see a correlation between contestants’ Facebook numbers and those being sent home each week. Although the results weren’t perfect, the accuracy of these Facebook-derived predictions proved more accurate than simple chance. For example, in Season 11, when the television audience voted to chop the Top 24 to the Top 10, the data predicted 8 of the top 10. And during the course of the next 13 weeks, the data successfully predicted 18 of 25 bottom-three (bottom 2 in week # 13) candidates, for an accuracy of 72%.

By the end of the first season, I had come up with a theory that the correlation to America’s voting had less to do with the total number of Facebook fans and more to do with the number of fans gained the night immediately preceding an elimination show. This method worked in all three seasons with one major caveat: it’s accuracy dropped with the number of contestants. The larger the pool, the easier it was to choose both the bottom three and who was likely to go home. However, as the talent pool got smaller, the method faltered, as proven by the fact that it chose the wrong winners in two of the three seasons.

Now, with three individual seasons of data under my belt, I thought it would be fun (yeah, I know) to compare the Facebook data from the top three performers from each season, side by side. I created the interactive bubble chart above for you to play with.

Note: Because American Idol ran for different lengths of time each season, I needed to find a way to align the data between seasons. Since all three data sets had the “last 11 weeks of data,” that’s the information that I used. Once the chart was built, however, I could see a flaw that needs to be explained. In its default configuration, Facebook Fans are on the x-axis, “Talking about” on the y-axis, and a number (1901) on the slider. Unfortunately, the Google Motion Chart wants that bottom number in years, yet in my data is actually weeks (1 through 11). So, when you animate the graph, note that 1901= week 1 for all three seasons, 1902 = week 2 for all three seasons, and so on.

One of the things I did expect to see was a year-over-year reduction in the number of Facebook fans, since supposedly the show’s television ratings are falling. The past three years of Nielsen ratings verifies that story:

2012 (Season 11) Finale: 20.7 million
2013 (Season 12) Finale: 13.3 million
2014 (Season 13) Finale: 10.4 million

So, when we compare the top three from each season, it’s no surprise that the total number of their Facebook fans have dropped:

2012 (Season 11) Sum of the Top 3 FB Fans: 574,531
2013 (Season 12) Sum of the Top 3 FB Fans: 153,586
2014 (Season 13) Sum of the Top 3 FB fans: 114,567

Not only is the number of FB fans per contestant is dropping, but the season-over-season differences are significant. For example, Season 11’s second-place contestant, Jessica Sanchez, accumulated almost as many Facebook fans in her first week (43,799) as Candice Glover, Season 12’s winner gained over her entire season (45,219). And except for the socially-popular Angie Miller from Season 12, Season 11’s 3rd place contestant, Joshua Ledet, had more fans than all four finalists from Season’s 12 and 13 (Kree Harrison, Candice Glover, Jena Irene, and Caleb Johnson).

But, as often happens, the data did reveal something unexpected. Although the total number of Facebook fans fell from season to season, the overall engagement of those audiences rose. Season 13’s final three (Alex Preston, Jena Irene, and Caleb Johnson) lead all nine contestants of in “talk-to-fan” ratios (Facebook’s “Talking About” metric divided by the number of fans). The data suggests that today’s America Idol fans, although down in viewership, are talking bout their favorite contestants more.

Dwindling television viewership may not be the only reason why Facebook numbers are down. The social media landscape is different. For example, Instagram may have drawn some audience from Facebook since it came onto the scene in 2010. The chart above shows the number of Instagram followers for all nine contestants after Season 13 ended. Both Alex (Season 13’s #3) and Caleb (Season 13’s winner) had more Instagram followers than they did Facebook followers.

So, what did I conclude from this three year experiment?:

  1. The number of Facebook fans gained the night between the performance and elimination night is a good indicator of who is going home for the first five weeks. But as the pool of contestants dwindles, so does the accuracy of the method. The method is ineffective in choosing the winner on Finale night.
  2. Although the number of Facebook fans is down year-to-year, today’s fans are talking about their favorites more than yesterday’s.

Someone who’d been following my experiment asked if I was disappointed that social data could not help predict the outcome of the contest. Absolutely not! In an age of big data, it’s comforting to know that humans and freewill still rule.

Probably the most important knowledge that I gained during these past three seasons was the ability to use motion charts. If you want to play with the motion charts for each season, you can view them here:

Season 11 American Idol Contestants Motion Chart

Season 12 American Idol Contestants Motion Chart

Season 13 American Idol Contestants Motion Chart

royaltyIn December 2011, Amazon opened the Kindle Owner’s Lending Library (KOLL), an innovative program that allows Amazon Prime members to borrow up to one book per month from a list of those enrolled in Amazon’s Kindle Direct Publishing Select (KDP Select) program. The lending library is unique because even though borrowers pay nothing for access to a book, authors are still paid by Amazon every time their book is borrowed. In addition, rather than setting a per-borrow price, Amazon left that decision to marketplace demand by promising to fund the program with at least $6 million during the course of the first year. Since the program began, final payment per borrow has been calculated by dividing each monthly allocation by the number of books borrowed in that month.

On KOLL’s first birthday, Amazon appears pleased with the results:

  • Amazon Prime members have borrowed over 3.5 million books
  • Authors have been paid $7.3 million (21% above Amazon’s annual commitment), which translates into an average of $2.04 per borrow.

Taking a closer look at allocations over the past year reveals that Amazon may be modulating the monthly amount in an attempt to keep author royalties above $2.00. After opening the program with $500,000 in December 2011, Prime members borrowed 295,000 books, which translated into $1.69 royalty per borrow. The company then increased its January 2012 allocation to $700,000, but the authors only received $1.60 for each borrow due to a 50% jump in borrows, as Christmas Kindles needed to be filled with content. For the next nine months, however, Amazon has held its monthly allocation steady at $600,000, resulting in 2.4 million books borrowed and an average royalty payment of $2.19 during that period.

total_books

In November 2012, the company demonstrated confidence in the business model by expanding KOLL’s reach beyond the United States to include Europe. Anticipating a jump in sales, Amazon increased its monthly allocation for the first time in 9 months, growing it from $600,000 to $700,000. If my assumption is correct, and $2.00 is indeed the magic royalty number, the company came pretty close to predicting European demand as 368,421 books were borrowed bringing the average royalty per borrow to $1.90.

Finally, Amazon appears to be anticipating another borrowing surge with the announcement that it is committing an additional $1.5 million to the fund over the holidays. The company has decided to add the first of that $700,000 to its December baseline of $700,000–bringing the monthly total to $1.4 million. If Amazon is trying to hit a royalty goal of $2.00 per loan, the company is expecting Prime members to borrow 700,000 books in December 2012, which would highest monthly figure to date (January 437,000) by more than 60%.

It looks as if Amazon’s experimental business model is paying off for both authors and the company. What do you think?

Aug 21, 2012

Yesterday, I got an email from NewBlueFX, a company that creates plugins for nonlinear video editors like my favorite, Sony Vegas. The email, entitled, “You Determine The Discount Sitewide Sale” described an interesting social marketing campaign that promised a one percent discount for every retweet of its message, with a sixty percent cap.

Here are the rules from the email:

  • The Tweet-a-thon starts Monday August 20th at 12:00AM PT and goes until Tuesday 11:59PM PT.
  • Total Tweets and corresponding discount to be announced Wednesday morning 7AM PT.
  • Discount can be used on any product available on NewBlueFX.com
  • Wednesday’s sitewide promotion expires Thursday, August 23rd at 7:00AM PT

I love this campaign from the standpoint that NewBlueFX clearly sees its audience as an asset. It also gives us the opportunity to play with some math to calculate potential dividends that the company is seeking to extract from those assets.

The company’s Twitter account, @newbluefx, has 928 followers (the asset). Due to NewBlueFX’s niche product line, this audience likely consists of specialized video artists who’ve made significant investments in sophisticated video editing software. Since like-minded people tend to follow each other on Twitter, NewBlueFX is betting that by encouraging its audience to spread its messages through their networks, that the message will ultimately find its way to the right people as opposed to (as traditional media offers) the most people.

The company has done something else by offering a discount for each retweet…it has established a retweet price.

Consider that the NewBlueFX online catalog contains 26 products, whose:

  • Average price is $127.64
  • Median price is $129.95
  • Mode price is $129.95
  • High price is $299.95
  • and a Low price is $49.95

By taking the median price of $129.95, New BlueFX has established a price of $1.299 for every retweet (1% of the median product price) its campaign generates–up to a cap of 60 retweets.

The model gets even more interesting if you compare it with advertising through traditional media. By capping its discount (and therefore the amount it is willing to pay for the entire campaign), any retweets generated over 60 will drive the effective cost per impression down–the opposite of what happens when purchasing traditional advertising.

Social vs. Traditional CPM

If the average Twitter user has 126 followers, each retweet has the potential to be spread to 60 x 126 = 7,560 people. Therefore, NewBlueFX has established that it is willing to pay $77.94 (60 x $1.299) for access to its audience’s audience. Had NewBlueFX decided to invest that same amount of money in someone else’s audience (traditional media), similar access would translate into a CPM (cost per thousand impressions) of $10.31 (1000*($77.94/7,560)).

Figure 1 illustrates the company’s CPM costs as a function of the retweets it generates:

  • If nobody retweets, NewBlueFx offers no discount, and therefore carries no advertising cost.
  • For the first 60 retweets, NewBlueFx pays a flat rate of $10.31 CPM, capping its campaign advertising expenditures to $77.94.
  • Then something very interesting happens. Because the discount is capped at 60%, every retweet that exceeds 60 is a bonus, essentially causing the company’s CPM rate to fall-off exponentially the more successful it is.

The more successful the campaign, the lower the CPM

 Figure 1: CPM per Retweet

This example shows the difference between social media and traditional media. In traditional media, your message will only travel as far as you are willing to pay for its distribution. In social media, your message will only travel as far as your audience (asset) is willing to spread it for you.

NewBlueFX recognizes that its audience is an asset that can pay dividends.

Does your company?