An apparent Long Tail paradox

Several people have written with what appears to be a paradox. Some Long Tail markets (online music, for instance) seem to be more hit-heavy than traditional offline retail. For example, one emailer notes that the NARM is reporting that in digital catalogs:

• 0.2% of the catalog (2,600 tracks) represents 50% of the sales
• 1.5% of catalog (22K tracks) is 77% of sales
• And the tail: 88.2% of the catalog is just 7% of the sales

Odd, no? Actually, no. Don't forget that the effect of a massive increase in inventory is always to make sales appear more concentrated in percentage terms. So, for example, if you have 1,000 titles and the top 10% (100) account for half the sales, and then you expand that to 10,000 titles, the exact same sales of the top 100 now mean that it's just the top 1% that accounts for the same half of sales. It's actually no more concentrated, but the bigger denominator makes it appear that it is.

The right way to compare these markets is the way people experience them: not in percent of total inventory but in absolute number of titles. Here are two sets of comparisons for illustration's sake:

Title rank                  Percent of total sales

DVDs                       Blockbuster     Netflix   
1-10                         31.2%             13.1%   
11-100                     36.4%             24.8%   
100-1000                 19.8%             25.9%   
1000 on                    12.6%             36.2 %   

Albums                   CDs (Soundscan)  Rhapsody   
1-10                          59.1%               44.6%   
11-100                       5.4 %                2.0 %   
100-1000                  13.2%                6.1%   
1000 on                     22.3%              47.3%   

Let's do the concentration analysis two ways, first expressing inventory in percentage terms and then in absolute figures, so the difference will become clear.

• Blockbuster (total inventory: 3,000 DVDs):  80% of sales come from top 10% of titles
• Netflix: (total inventory: 60,000 DVDs): 80% of rentals come from top 5% of titles

Seen that way, it looks like Netflix's business is twice as concentrated in the hits as Blockbuster's, weirdly enough.  But that's actually just an artifact of the fact that Netflix stocks twenty times as many titles as Blockbuster. For a more illuminating comparison, let's do the inventory analysis again in absolute numbers.

• Blockbuster: The top 100 titles account for 69.4% of sales.
• Netflix: The top 100 titles account for 38.8% of sales.

Which is just what we would expect. Netflix's demand curve really is much less hit-centric than Blockbuster's. The top 100, which are the majority of sales offline, are the minority of sales online. Which is exactly what Long Tail economics would predict. Phew.

Comments

[Will Davies]

Putting the long-tail to one side for a moment, this is not in the least bit surprising. For a while, digital technology was more commonly associated with 'winner-take-all markets' than with long tails, the rationale being that it enabled infinite reproducibility of goods, at zero marginal cost. This, of course, is as much a feature of the IP regime and of the commercialisation techniques that publishers now tack on to their hits. But the fact that J K Rowling earns the same amount as several thousand moderately successful authors, or David Beckham the same as an entire league of semi-pro footballers, is not news-worthy any longer.

My understanding of the long tail has always been that the curve hits higher up the y axis *as well as* extending further down the x axis.

[Jitendra Gupta]

I am no sure I agree with the analysis here...The main reason is that we are comparing apples and oranges here. We are comparing a title based renting model (Block buster) with a subscription based renting model (Net Flix). With a subscription based renting model people see movies that they might have at the top of thier lists where as a title based model cannot do that if it wants to have repeat business.

Thanks. Jitendra

[Jitendra Gupta]

Correction on the comment above:

With a subscription based renting model people are forced to see movies that are not at the top of thier lists where as a title based model cannot do that if it wants to have repeat business.

[Justin Kownacki]

Regarding the above comments: perhaps it's just a case of semantics, but I don't believe anyone is "forcing" NetFlix users to see movies that aren't at the top of their lists -- especially since their lists are completely customizable and what's at the top of your list may not be what's at the top of mine. If the nationally popular titles are at the top of everyone's list and therefore not necessarily immediately available, you may need to temporarily satisfy yourself with your second choice... but it's still YOUR second choice. Meanwhile, Blockbuster essentially TELLS you what's at the top of the list because they MAKE the list (or the studios and distributors do).

In that sense, Blockbuster could just as easily say, "Okay, we're not carrying any new films whatsoever, but we WILL have 75 copies of The Godfather in the store." Sure, there'd be an outrage, and a huge loss of business. But eventually the model would stabilize, and you know what you'd have: a whole lot of people renting The Godfather. Why? Because it's available.

The Long Tail isn't just about what's popular or selling well, since that's as much a product of availability as it is popular taste. The Long Tail is about the ability to sell (or buy) what you want when you want it -- in essence, the value of on-demand. In that sense, NetFlix gets it and actualizes it in a way that Blockbuster, due to shelf space or geographical markets or payola, never could.

[David]

The most important fact with the NARM numbers is that these are tracks and not full CDs. How many different artists and genre's are represented in the 22K tracks? The retails singles market was killed by the labels more than a decade ago (the top retailers like Walmart, Target and Best Buy sell NO singles), so there is no way to produce a straight-forward accurate comparison between online and traditional music retailers.

[Jitendra Gupta]

Interesting comments!

If I were to choose one thing driving the difference between the sales & title rank distribution between BlockBuster and Netflix, it will really be the difference in the payment model. With the Netflix subscription based model, users explore more and therefore the big catalog comes into play. With BlockBuster customers pay rent per title and therefore are not looking beyond a few hits that made them come to the store in the first place. An apple to apple comparison to me would be comparing the trends of a subscription based model at BlockBuster (I don’t know if they even have it) with that of NetFlix. My guess will be that in such a case, we will see a lot similar distribution between NetFlix and BlockBuster.
In case BlockBuster had a subscription based model, another factor that could come into play will be the search costs associated with looking at a large number of titles. NetFlix by being an online store is able to leverage user profile and rent history/ratings and offer personalized recommendation. That will be really hard to do for a brick and mortar store.

[azeem]

Chris,

One way of expressing this would be the develop the long-tail coefficient which could express the shape of the area 'under the curve', ranging from tending towards zero (all sales from a single item) to 1, most sales from out in the curve.

If i=1 to n where n is the number of items in the inventory
and Xi is a particular position of item X in the inventory; i.e. X2=2/n (i.e. the second item in the inventory) and X6021=6021/n i.e. the 6021st item in the inventory.
and Yi is the proportion of sales from item i, expressed as a percentage or ratio from zero to one.

we then

sum (XiYi) for i=1 to n to give us a single number. A high number suggests good long-tailishness. A low number suggests non-longtailishness.

In practice, you'd be expressing the shape of a jagged line, since you only have a few datapoints. you'd also want to express the number of items in the inventory to avoid trivial examples.

I am sure there is a better way to express the long-tailishness of the curve, but you'dneed a real mathematician to do that.
cheers