Will Recommendation Algorithms Always Be Trumped by the Personal Touch?
Back in the early days, Amazon recommendations were decided by human beings. The company went on a hiring spree, bringing in editors to manage the site. Typically these were not the kind of West Coast engineers that typified the company; rather they were Manhattanites, steeped in the world of books and culture. Well-read, opinionated, they would read hundreds of books and write individual reviews. They were the gatekeepers that managed what was already (and self-consciously) the world’s biggest bookstore.
It wasn’t to last. Amazon wanted an automated solution, one that could scale up, that was measurable. But getting it right was proving more difficult than expected. It turned out getting recommendations that made sense was a hard problem. One staffer, Greg Linden, went for a different angle. Prior to him the main approach had been to look at an individual’s purchasing history and extrapolate out from that. Linden took people out of the equation.