Slashdot links to this story about company that has made a 'dynamic pricing model' for selling MP3s over the internet. (Amazon is hiring their services.) It's a demand-based system: the price goes up as more people buy.

This kind of dynamic, human-free system is fascinating: like Dell's method of managing component bottlenecks in laptop sales by automatically discounting, say, 80 gig drives if the 40 gig model is going to take longer to arrive.

What's peculiar about this case is that it only needs to manage demand - supply isn't subject to any of its usual restraints. A change in supply is usually a long-run thing: companies, seeing profits to be made when the price of something goes up, have to enter the market with all that implies in shifting to different production regimes. This system is devoid of any of this. All costs are pretty much either sunk or fixed, excepting the miniscule price of the few electrons they have to pump down the tinternet tubes. Given that lowering costs is supposed to move the supply curve to the right, the quantity supplied can be essentially infinite.

It would be really interesting to see how they've implemented the program. It can't work analytically - that is, there's no equilbrium between supply and demand to find. All it can do is incrementally put the price up and wait for a pre-defined period of time to see how the download number changes, and by how much (i.e. how elastic demand is.)

The price of a download in this case is capped at 98 cents: I'm not sure how many people will actually be affected by such small changes in price at all. But songs start off at zero cents, the article says. This just doesn't seem to make sense to me. If fans are expecting a new mp3 single, the first lot will be getting it cheap. Oh, I suppose it depends on how they implement the price-setting algorithm.

If I was doing it, I'm not sure it would help to think about a standard demand graph. Yes, given enough of a price rise, sales will drop off. But the aim is to find a method for changing price in a way that maximises profit in a defined time period. Given that, such a model isn't going to care whether its 'moving along the demand curve' or whether the demand curve shifts to the right because there's just been an advertising campaign. Its going to be interested in extrapolating from known previous changes in demand only.

The company in question will be able to get an enormous amount of very direct data to analyse to achieve this. (Although, again, I'm still sceptical that such a small price difference will have much impact on buyers, but I could be totally wrong.) Also interesting: this draws attention to the fact that, usually, changes in price are made one by one, by real human beings - despite the fact that yer supply and demand curves suggest its a natural process, with people just abstracted conduits. And maybe they are: but there's a big disparity between the expensive mathematical methods large corporations use to work out their price and production level, and how prices are set in small human-scale markets. And more difference still between these and automated price-setting, actuarially groping about for the moment-to-moment profit maximum.