Measure and react
Once you accept that your intuition about how people behave is inherently flawed, then you really need a different model for learning about the world. Everything becomes data driven in a real-time, reactive way. A classic example in the Web industry is what we call bucket testing, where you might say, “I don’t know what to put on the front page of Yahoo!. I have very good editors who have plenty of ideas, and they can generate a pool of good candidates.” But if we want to optimize this, we actually have to go and show these different combinations of articles to buckets of people. Within a few minutes, we’ve got a million clicks that we can use to tell us which articles are getting clicked on more.
We can do the same thing for the display of advertisements, for the design of pages. All sorts of design parameters and choices that were once within the purview of intuition, of experts, are now tasks that can be distributed to the user population and learned empirically in real time.
This kind of measure-and-react strategy, as I call it, is particularly powerful on the Web because the numbers are very large and the cost of generating multiple versions is very low. But, in principle, this is something that could be done in the offline world as well. We see it in the fashion industry with Zara and in the casino industry with Harrah’s or in retailing, where you can systematically rearrange product positions on shelves in stores.
No comments:
Post a Comment