Creating a Merit-Based Music Economy: Compulsory or Blanket Licensing for Interactive Subscription Services
Summary | Contents | Previous | Next



C2. Breaking the Promotion Bottleneck, Ending The Muzak Effect

If mass media created the Muzak Effect, with its insidious distortion of commercial music culture, then custom media promises to return music culture to its natural balance. The problem with mass media is that it requires the same program to be accepted by everyone in a group, leading to the success of tolerable music over favorite music, and the resulting overall cultural distortion. The solution is to find a way to deliver customized programs with the efficiency and success of mass media.

Computing and telecommunications technologies have now provided the first opportunity for this to happen widely, and it's called "mass customization" in the marketing business.

Mass Production and Productivity: Mass media have become very successful because of their efficient productivity. For a relatively small cost per audience member, information can be multiplied from a single source to a very large number of people. And because broadcasting usually has a fairly fixed overhead, and profit is generated per audience member, larger audiences generate more profit.

It is the media equivalent of the revolution in mass production started by Henry Ford's Model-T assembly line. It's a lot easier (and therefore cheaper) to make a standardized product for a lot of people all at once than to make different products for each individual customer. Customized products might still be higher quality in some ways, but they are much lower in productivity (the number that can be created with equal cost is much lower), and so custom-built products have tended to remain much more expensive compared to mass-produced products.

Mass production brought many kinds of products to the masses that were previously out of reach because of their high cost. Mass media did the same for information and entertainment, improving on the much lower productivity of live appearances and geographically constrained meetings.

Mass Customization: Mass customization combines the efficiency of mass production with the personalized quality of custom production. It's a completely unprecedented capability, with truly revolutionary potential. Using a computerized database of individual customer information, computer-controlled manufacturing can produce custom products with an efficiency approaching standardized production. In short, machine-enhanced manufacturing has been supercharged with machine-enhanced variability.

The technology to personalize consumer products and services is becoming widespread. It starts by providing standard options, from which each customer chooses a personal combination, and that "multi-standardization" is what makes the process efficient.

The same strategy can be applied to the media world, by building customer profiles for media services. That lets them personalize the choice of content efficiently enough to keep costs comparable to mass media. Deliver that personalized content using the point-to-point technology of the Internet, and you get a fully functional platform for mass customized media -- personalized content, combined with personalized delivery.

Some people object that the Internet is not ready to reliably support personalized music streaming on a very wide scale yet, and they may be correct -- today. Bandwidth costs for online systems are still high, and the price most people would be willing to pay probably does not cover the full costs of delivery in today's market. However, one of the most consistent patterns in both computer and telecommunications businesses is the increase of computing power and transmission bandwidth per dollar, over time.

In the computer processor business it is known as Moore's Law (processing power doubles about every 18 months), and in the telecommunications business, bandwidth efficiency has increased at an even greater rate. It is very likely that delivering custom media online will become cost effective in a finite period. It can be debated how long it will take, but most agree that it is not a question of "if" but "when."

Personalized Radio: What this makes possible is music programming very similar to a radio program, but individually personalized for each listener, breaking the dependence upon group-wise programming that leads to the Muzak Effect. This personalization needs to include the same "novelty with familiarity" as broadcast radio, balancing freshness with an environment of favorites.

A key feature of this personalization is how new music is selected to play along with the music that listeners already list in their preferences. If we want to avoid hobbling this system with the same bottleneck as mass media, we have to avoid using a small number of human decision-makers to make these choices. The ideal service systematically identifies new music based on listener preferences, and includes some of it in the live program, on a quasi-random basis. These systems are often referred to as "auto-recommendation" systems, and there are several methods that can be used.

Collaborative Filtering: One of the most interesting is called "collaborative filtering." This process collects the music preferences of many individuals together into a single database and then compares likes and dislikes. If fans of Artist A generally also like Artist B, then the system will sometimes play Artist B for fans of Artist A, even if they haven't explicitly chosen Artist B. This can even be done song-by-song, instead of artist-by-artist. This is a systematic way to extend the "word of mouth" process of individual recommendations between people who are not acquainted with each other directly.

Sounds-Like: Another method with some intriguing possibilities is called a "sounds-like algorithm." Music is broken out into measurable variables that correspond to the ways music actually sounds to humans. It's then analyzed according to how similar it is to other music in the database, so when a fan chooses one song, similar songs are sometimes played as well. These algorithms are in the early stages of refinement, but their special advantage is that they can make recommendations without requiring a critical mass of explicit fan preferences. Collaborative filtering requires music to have a certain number of fans before it can start reliably recommending that music to other listeners.

Artist recommendations: One may also ask artists directly which other artists they feel influenced by, as well as which other artists they enjoy but who don't directly influence their music. These two kinds of artist recommendations can be very helpful in navigating the music world, especially if the recommendations can be used in both directions, so that a lesser-known artist who feels influenced by a better-known artist will be occasionally programmed for fans of the better-known artist.

Word of mouth: Aside from these systematic recommendation methods, there can also be forms of word-of-mouth made easier in an integrated system. Song and playlist recommendations might be passed from person to person. One person's preferences might be chosen by a second to add to their own preferences. And explicit searching through preferences of a selected artist's fans might lead to the discovery of new artists to sample on demand.

Traditional methods: In addition to these new methods, more traditional methods like editorial recommendations and sub-genre classifications can still be useful, even though they have an inevitable subjective skew, and in the case of editorial judgments can be constrained by the star bottleneck. They provide a standard, predictable place to start, for listeners who haven't yet explored much music, and they have their place in any full-featured system.

Having all of these methods integrated together, with flexibility to choose among them, gives individual listeners tremendous control over their music experience. At the same time, it systematically exposes a very wide range of artists to their appropriate fans.

Most of these methods are currently still not fully developed, and require more fine-tuning before they reach their ultimate potential. They have shown some real promise, but for now there are none that completely fulfill their intent. Again, it is basically a matter of time and resource allocation before these kinds of tools reach their full flower.

Imperfection: It should be noted that no recommendation system, whether human-mediated or automatic, could ever be "perfect" for an individual listener's favorite music, even if ideally implemented. People's tastes evolve over time and are unpredictable. Even though these sorts of systems can get pretty close, they will always make mistakes. But as long as the service allows the listener to skip those mistakes easily, they will not seriously degrade the overall music experience, and can be forgiven.

In the broadcast radio world, mistakes are not allowed, because the only way listeners can skip them is to change the channel entirely. But once you add moment-to-moment control to the experience, mistakes become much less damaging, and tend to be forgotten in the midst of all the other good music. And while skipping mistakes, the system can collect that information and get better at avoiding mistakes over time. Personalized systems can learn from individual users' interactions with them over time.


Summary | Contents | Previous | Next