Stephen Colbert had a great bit on his show last week on how retailers comb though massive amounts of customer data to predict and influence consumer behavior. In this particular example, Target used changes in a teen-age girl’s buying patterns to predict that she was pregnant, determine her approximate due date, and schedule baby-related coupons to show up in her mail. The big downside to this approach was in this instance, it was the influx of these coupons which tipped off the girl’s father to her condition. Note: this is what Target is doing. God only knows what Facebook and Apple are up to. Turns out there’s an awful lot of money to be had in 1. selling people stuff they don’t need or 2. convincing people that one brand of stuff they need is much better than another brand of the same stuff. There’s a reason the word “conversion” is used by both marketeers and cult leaders. Teasing insights in consumer behavior from the purchasing histories of millions of customers has driven some impressive technology development.
Now, the technology underneath Target’s approach isn’t that much different than combing through reams of genetic data to figure out what treatments a patient should get. (If you’re being picky the biggest difference is that the marketing problem is a “big N, small p” statistical problem, and clinical genomics is a “small N, big p” problem.) The genomics stuff has just been a whole lot less profitable so far, as evidenced by the somewhat abysmal returns of the entire pharmaceutical industry over the past decade or two. However, you’d think that a few of the people involved in cutting-edge big data technologies would find their way into the life sciences, and even into the even less profitable early-stage research end of the business. I’ve been trying to hire a few people like this, and have been pleasantly surprised at the resumes showing up in my Inbox. Even in today’s good economy where software engineers are enjoying their pick of positions, I’ve been able to find a few really good people who are psyched to come to a non-profit start up to do the “big data” back end processing genomic analytics stuff. It seems that’s what the real men, and even some real women, want to work on.
However, there’s a certain type of person who I haven’t found yet. See, at Target, they not only figured out how to predict a woman’s maternal status, they figured out how to sell her stuff while mostly not creeping her out. My initial story is interesting because it’s actually an exception to the more subtle ways these companies usually work. Now, if you’ve ever been around a first time mother you’ll know they are easily creeped out (in a good way!). Having a corporation send a woman baby coupons before a she has told her family she’s expecting might possibly be perceived as a bit creepy. So, what Target typically does is figure out when the baby is coming, and then send over a good mix of lawn mower and garden hose ads along with the baby stuff they really want to sell her. I don’t know who came up with that strategy, but it wasn’t the type of engineer I’ve been hiring.
Turns out that the hardest problem that Sage Bionetworks faces isn’t actually the science, although that’s pretty hard. But what’s even harder is trying to create an environment where the scientific community as a whole is more open with their data and analysis techniques. It’s a tricky problem because scientists mostly benefit (win grants and tenure) when they publish their own work and have their name come first on the author list. Hence they tend to have a lot of interest for other scientists to share things with them that further their own research, but limited incentives to make their own research truly reusable by others. There’s a bit of a Prisoner’s Dilemma going on here, with not only researchers, but patients benefiting less due to the restricted research productivity that comes from a community of scientists that refuses to share effectively.
This problem seems almost impossible to solve. One thing that gives me hope is the counter-example of open source software proving that you can have communities where enough people open their work up for others that additional developers keep buying into the system. The culture of the open source community maintains itself, and certain software tools, e.g. GitHub have what Joel Spolsky terms the right “social interface” for helping to propagate behavior that is in the community’s, rather then the individual’s best interest. Or, as another example, at some point eBay figured out that people would actually sell and buy stuff from complete strangers over the internet if eBay put little stars next to their screen names to indicate how trustworthy other anonymous random people thought the buyers and seller were. Really, stop and read that one again because it seems more and more ridiculous the more you think about it. Somehow it worked to create a culture where most people don’t rip each other off.
There’s been a big influx of people from the “hard sciences” into the life sciences over the past decades, from physics, mathematics, chemistry, computer science. That’s been good for the field. What I haven’t see enter the field are the newest ideas from psychology, economics, and marketing, and I think importing concepts from these “soft sciences” could be just as valuable in driving the field forward. Over on the Sage Bionetworks site you’ll find a job posting I wrote for a UI-focused software engineer. Sure, there’s some coding and design involved. But the real tricky and interesting bit is finding someone who can combine those coding skills with some innovative insights into the problem I’ve outlined in this post, and using the software to create leverage for social change. That’s the person I really want to hire next.