Cutting the Long Tail Short?

6 Nov

On Tuesday I was lucky enough to attend a little bit of the Telco 2.0 event in London. James talks about it in a little more detail here. I was only there for a half day (and apparently missed an awesome presentation by the CTO of Amazon…) and while there was a lot of interesting content being presented and debated (in the specific Telco 2.0 Mindshare approach to promote interactive audience participation) I wanted to react on a specific presentation that I found both very interesting and deeply flawed. I will be fairly blunt in this post: since Page attempted a direct (and often quite bilious) refutation of Anderson's work, I'm pretty sure he'll be open to direct criticism himself.

Will Page is an economist at MPCS/PRS, a UK based music royalties management organisation. His presentation aimed at disproving Chris Anderson's Long Tail. If you're not in the know on Chris Anderson's Long Tail, you can check out his original Wired article on the topic, the excellent Wikipedia page, or even better, read Anderson's book in which the argument is laid out in a much more subtle way. The basic (but often misunderstood) tenet of Chris Anderson's approach is that digitization of content (esp. media content) changes the market dynamics as scarcity of shelf space and inventory costs are reduced/eliminated. In effect, it drives proportionally more sales towards low visibility/niche items (the tail).

Essentially, what Will Page set out to do was to prove that Chris Anderson was wrong by analysing the selling data for iTunes in the UK (Page didn't say that was his dataset, but everybody in the audience knew it was…) He used statistical methods for analysing distribution patterns to do this demonstration, which essentially came down to the following points:

  • Anderson only analysed selling volumes, never revenue, and it's the latter that's relevant to players in the market
  • Of the 13 million available tracks on iTunes UK, 10 million don't sell at all. If Anderson's theory worked, they should.
  • A comparative mapping of the iTunes sales and traditional selling patterns shows that the "head" items sell significantly more than they would be expected for a brick and mortar retailer. Again, according to Page, this is inconsistent with Anderson's "theory".

There are some aspects of Page's presentation I found interesting and a good challenge to Anderson. The revenue angle is obviously crucial (although it could be said that looking at revenue purely on the basis of what individual sales generate as Page did is also very "old economy" and definitely not telco 2.0) and was indeed not explored by Anderson. This is an area where more data and analysis can only be welcome.

The fact that head items are generating more revenue proportionally than they would be expected to in brick and mortar is also interesting and worth exploring. In my reading of the Long Tail that's not necessarily inconsistent with Anderson's views. Anderson is often caricatured into saying that there will no longer be a head, which he's not stating at all. ON the contrary, he says the head is necessary to generate tail sales, and that it's crucial enablers like search and filtering (recommendations, etc.) that allow the customer to tap into the tail. And indeed, I noticed that Page didn't comment on the fact that his very own graph also showed the tail items overperforming what would be expected in a brick and mortar distribution pattern. This may have been skewed by the fact that he couldn't really map zero sales, but it would have been worth a comment at least.

The one argument that I found very interesting from him was that Anderson theorises a lack of shelf scarcity when in fact the portal of online retailers act like a display of sorts and is indeed in very short supply. There's no doubt in my mind that this remains a crucially scarce resource and in that sense, Anderson was theorising a very utopian digital distribution that does not and probably cannot exist.

However, this very argument also suggests that iTunes portal strategy may be a strong determining factor in skewing the sales towards the head, just as Amazon's is notorious for encouraging tail sales. Again, Page didn't touch on this, potentially a major explanation as to why 70% of the stock wasn't selling.

And this in turns lead me to question iTunes as a relevant model to actually analyses these trends. iTunes is far from a pure and perfect market (assuming that one exists) and is in fact, to me, a very artificial market in at least two ways:

  • First of all, the iTunes business does not require for music sales to actually generate a profit. Apple makes its butter on the sale of iPod devices, not on the sale of music.
  • Perhaps more crucially, there is no price flexibility on iTunes. A normal market adjusts prices according to costs and demand, but on iTunes, everything is £0.99.

A crucial aspect of the supposed virtues of Anderson's Long Tail was that tail items were of lower cost, and therefore would be sold for lower prices and/or generate higher margins. Page uses a data-set where the concept of margin is irrelevant, and the pricing is fixed irrespective of costs or demand. I'm no economist, but it seems to me that both of these are fundamental issues with using iTunes as an example to refute Anderson's theory.

Now let's be clear, I'm not saying Anderson is absolutely right. I think he described – with what datasets he had available – an potential evolution in content distribution. The reason this is relevant to telcos is that many of them see content distribution as a way to generate new revenues and overturn the squeeze they're currently speeding towards where rising traffic costs will pass flat fee revenues. And to be fair, although many telcos have moved into content distribution, I have yet to see something that even vaguely resembles a long tail model (very large catalog, powerful search and powerful filters) in any of the telco initiatives.

So I'm all for trying to assess whether Anderson's model is actually verified in real-world experiences beyond the ones that Anderson himself analyses in the book. But sadly, Page's refutation is largely flawed, a lot more so than the source material he's trying to refute. I'm not an economist (and neither is Anderson) but Page is and that makes this presentation a little puzzling to me. It seems to me there has to be an ulterior motive behind his diatribe (besides the fact that he seemed to resent Anderson's success) but I can't quite figure out what it was…

4 Responses to “Cutting the Long Tail Short?”

  1. Steve November 9, 2008 at 5:29 pm #

    Benoit – when most adults find that the facts don't fit the theory, they discard the theory. You seem to prefer to discard the facts.

    Why not mention that the data fitted a lognormal distribution very closely, and didn't fit a Pareto curve, as Anderson predicted? That's the most interesting observation to emerge from this research, and not to mention it is quite an omission.

    Casting aspertions on the integrity of Mr Page (because his facts ruin your fantasy) is not adult behaviour either.

    Very disappointing post.

  2. Benoit FELTEN November 9, 2008 at 8:35 pm #

    Wow. I didn't think I'd move into the realm of personal attacks quite so fast, but since this is the first on my blog, I guess I should feel flattered.

    I'm not defending Anderson at all costs, I just see some pretty serious holes in Page's refutation. If you don't agree with that fine, but you seem at least as much a "non-believer" as you accuse me to be a "believer", and neither has anything to do with fact.

    I've discussed Anderson's theory at length with many people (very few of whom have actually read the book, I should add) and the conception that the long tail model just happens is silly and not asserted by Anderson anywhere as far as I can see. So disputing the theory with a dataset where the theoretical conditions for the model to be verified cannot happen seems pointless to me. Apparently, otherwise to you.

    Finally, where do you read that I refute that Page's dataset is not a log-normal ? I state pretty clearly that 10 million tracks in his dataset don't sell, which is pretty clearly in opposition with Anderson's theory, were the dataset valid to make that comparison. I just tried to describe what I saw in a way that would be understandble for anyone, not just statistics buffs…

    And yes, I tend to dislike people with hidden agendas and Page clearly struck me as someone who had one. And incidentally, reading the article in the Register Page is clearly not the only one. I don't understand why a refutation, if it needs to be done, cannot be done in a cold and analytical way. Page's refutation not only didn't convince me, but it didn't strike me as cold or analytical either.

    Maybe I read too much in there, but your post now suggests I may not even have read enough!!!

  3. Marlin May November 10, 2008 at 3:46 pm #

    Mr. Anderson comes to his own defense.

Leave a Reply