Direct marketing discussion question
Kai_Kaye
Tastearticle.docxThe subject of Tom Vanderbilt’s “You May Also Like” (Knopf) is taste, the term he uses for whatever it is that guides our preference for chocolate over vanilla, taupe over beige, “The Bourne Supremacy ” over “The Bourne Ultimatum”.
Taste is not congenital: we don’t inherit it. And it’s not consistent. We come to like things we thought we hated (or actually did hate), and we are very poor predictors of what we are likely to like in the future.
Taste today is a big business. The science of preferences dates back to the origins of the advertising and public-relations industries, but the Internet has provided it with a vast new field of operations. Compared with television, which basically had advertisers throwing tomatoes at barns labelled, for example, “Women eighteen to thirty-four,” the Internet is a precision instrument—as we all know from the lists, ads, and pop-ups on our screens that seem to know who we are and what we might be of a mind to pay for.
And they do know, sort of. Vanderbilt talked to a number of people whose job is to come up with the algorithms, derived from the staggering amount of data collected from clicks that produce a taste fingerprint for every consumer using a Web site or an app. He finds that, in the past several years, online marketing strategies have become extremely sophisticated.
With television, even after we purchased the Kellogg’s Frosted Flakes or Popeil’s Pocket Fisherman or whatever product was sponsoring our show, we kept seeing commercials for it. That was a waste of our time and, much more important, of the advertiser’s dollars. Algorithms aren’t supposed to generate recommendations for products we’ve already bought (though we still see a lot of these). They also aren’t supposed to recommend products simply because millions of people have bought them. Netflix once made this mistake, which is why you were constantly being invited to watch “The Shawshank Redemption” (and probably did a few times before catching on to the game).
Netflix learned, further, that recommendations shouldn’t be based on what viewers say they watch, since people over-report the number of foreign films and documentaries they claim to enjoy after a delicious foie-gras paired with a fun little Riesling. So the company now tries to figure out what we want to watch based on what we actually have watched. And not only does Netflix know what we have watched; it knows whether we watched the whole thing, and, if we didn’t, exactly where we stopped.
Then there’s the Internet spectacle known as “customer reviews.” This is, let’s face it, an open sewer. Once, when venturing out to buy a much needed tube of superglue, we went into the store, eyeballed the packaging, and made a guess that the niftier presentation, combined with the most plausible price, correlated with the gluiest glue. (A lot of us still buy wine this way.) On the Web, we have instant access to the unsolicited opinions of hundreds of superglue buyers (mostly pseudonymous, one of the worst things about the Internet), from the adhesives wonks who post “read more” commentaries on molecular compounds to the one-star hotheads on permanent caps lock and to hell with spell-check.
We don’t want to, but we often find ourselves identifying with the hotheads. We want to know, if things go wrong, just how bad it could be. This gives a single sufficiently radioactive bad review a blackball effect—which is, of course, the most fervent hope of the person who posted it.
According to food writer Ruth Reichl: “Anybody who believes Yelp is an idiot. Most people on Yelp have no idea what they’re talking about.” Customer reviews appear to be governed by a combination of pack mentality and “My water glass wasn’t refilled promptly!!!” narcissism. Reviews tend to be asymmetrically bimodal; they form a J-shaped distribution, with many high ratings, a smaller number of low ratings, and not much in between. The higher number of high ratings may reflect “positivity bias.” Studies show that if the first review is a rave subsequent reviews are more likely to be positive. If you are selling a product online, it makes all the sense in the world for you to have a friend post a positive review the instant the page goes up. We can often tell—shopping for books on Amazon, for example—when someone has taken this wise precaution.
One explanation for the low proportion of mid-range ratings is that the tiny fraction of customers who bother to write reviews do it because they had either an exceptionally good experience or an exceptionally bad one—which is, by statistical definition, not the experience you are going to have. Reliability is also compromised by the phenomenon of ratings inflation, the result of allowing sellers to review buyers as well as vice versa, as happens on services like eBay and Uber. It’s all a mess. But, assuming the wisdom of crowds, it’s probably not that much more untrustworthy than the advice of the salesman in the store, and it beats staring at the label.
