More bumpiness on the road to utopia

Back in 2002, I wrote a post about an article in the New York Times and an interview on Fresh Air, both by Jefferey Rosen.

The main point was this:

“(I)t’s notoriously difficult to find parking (at Oracle’s headquarters), and the space (Rosen) finally found was far enough away from the door that he had to walk something like 15 minutes to get to the building.

Here’s the big question: If Oracle can’t even reliably predict how many people will park at their own building — which is presumably why they haven’t built adequate facilities, and not because, say, Larry Ellison is a cheap bastard who doesn’t care about his employees much — how reliable do you think they’ll be at predicting terrorists (which is what they were touting their software for as a tool)?”

So now it’s 13 years later… and the state of the art when it comes to using software to make large-scale complicated predictions hasn’t advanced very much. Or such is the implication of this story from the radio show Marketplace, which is all about how UPS is having a more difficult than usual stretch of predicting their delivery times… Which in turn is because they’re being handed bad predictions from their customers about how much product they’re going to be shipping to end customers like you and me for Christmas.

The really interesting part of that, when it comes to looking back up the logistics tail, is many of the companies UPS is delivering for are companies who are mainly or solely online. Which means all the “magic” Amazon, Google, Facebook and others use for their “targeted” advertising… isn’t very good at predicting what, and how much, you’re going to buy.

“The problem that challenges both online retailers and carriers is the increasingly unpredictable shopper. Now that people can buy anything, anywhere, on their phones, retailers and carriers are having a hard time figuring out their next moves.”

That’s a much more systemic criticism than it looks at first glance. Because it reinforces the idea that the relationship between these companies’ advertising and enterprise management tools to how people behave in the real world is fairly… well, random.

This is also reinforced by Facebook’s page for your Ad Preferences. This page, algorithmically generated, is a concise list of what Facebook thinks it “knows” about you, from scanning your posts. And for many, many people I’ve seen, it’s laughably inaccurate. Which means the ads Facebook sells, oh-so-targeted at just the right people to respond to them (or so they tell the people paying to place such ads), are also laughably inaccurate in their targeting.

Which is part of why UPS can’t get your package to you in time.