I’ve been following the 23andme saga with dread and fascination and frustration since it hit earlier in the week. If you’re not up to date, David Dobbs has the canonical collection of stories to get you there.

This is a post about business models and culture clashes. It’s long. I should also start by disclosing that my former non profit project received funding from Anne Wojcicki’s philanthropic foundation, that I hold some shares in small startups that could benefit from the success of direct to consumer genomics (though not in 23andme) and that I now work at a non profit that is premised on the idea that personal access to data will fundamentally change research. My 23andme data is online for all to download.

So I could personally benefit if 23andme survives and thrives, and I personally admire Anne and her work, even when I criticize some of the choices she makes. It’s hard as hell to be a CEO and she deserves credit for icebreaking, vision, and perserverance.

That said, I am deeply frustrated by the simplistic narrative of OMG FDA BIG GUBBERMINT SILENCING DARING ENTREPRENEUR. It’s not that simple.

23andme has been trying to bring a technology business model, and a bayesian data culture, to the genetics world for six years. Both of those were gambles. And I think the problem they’re in right now is a direct consequence of those gambles not paying off yet.

First, by a technology business model, I mean the idea that you go get all the users you can get by giving your service away for free and figure out a way to make money later. This is no different than social networks business models. Facebook, Twitter, Instagram, Snapchat - all of them started with this concept of acquiring the most users and then working out a model for revenue. And all of them converged on advertising, because for all the promises of the new network, selling people sweaters worked better than anything else.

As a business model for genetics though, it’s different, because your cost isn’t just distributing an app and paying Amazon cloud fees. 23andme has been selling their direct-to-consumer genotyping kits for six years now, always at a deep discount to their actual price. They exacerbated that financial model when they dropped the monthly subscription requirement last year. If they can’t find a way to make money soon, they’re going to wind up in the pets.com box - you can’t sell a dollar for fifty cents and expect to make money on volume.

I would guess that the hope was that population genetics would yield a business model that wasn’t regulated. If you could get enough people in the door, then your internal analysis could inform efforts at pharma, at insurance companies, and more - without ever having to market your spit kits as medical devices in the regular business model. Their push to hit a million users indicate this remains a goal.

But since they’re not doing that yet, it seems they’re not at the magic number. That means going the traditional route, which is to develop and rigorously validate a test, get it approved by the FDA, and then market it at an incredible markup. That’s what 23andme tried to do in this run. They invited the FDA in the front door, and at least so far, haven’t been very good about serving them tea. We’ll find out more about why later.

But I posit the data culture that permeates technology companies is at the root of the frosty relationship between the FDA and 23andme. It makes it very hard to pivot from a model based on loss leadership and data mining to one based on regulatory submission.

That “traditional” submission to the FDA would be of a very specific kind of analysis based on randomized controlled trials. It is designed to keep bad things from happening to people, not to make sure good things happen to people. As one of my favorite papers lays out, parachutes would not receive FDA approval as a gravity-resisting device.

Modern tech culture doesn’t work that way. Bayes’ rule is about probability. It’s a different way of knowing that you know something, and it’s one in which there is far more tolerance for uncertainty than the FDA is accustomed to. And there’s been statements by FDA officials that they are deeply uncomfortable with that. Note in particular Janet Woodcock’s statement that causal inference needs a “level of rigor” - and the date in late April. It’s right before 23andme cut off communication with the FDA.

The FDA isn’t saying that 23andme needs to shut down, or stop giving people data, or stop providing spit kits. It’s saying their causal inference doesn’t hit the level of rigor that allows it to pursue a technology business model in the health regulatory space. That’s a gamble 23andme was always taking, that the FDA wouldn’t know how to regulate it in time for it to make money. It’s not an overreach, or a screw job on patients. It’s just business.

And I hope this isn’t the end of it. We need DTC screening. It helped me. It’ll help many others. But until the FDA learns how to deal with Bayes’s rule and its discomforts - and until DTC companies figure out a business model that isn’t based on massive loss leadership - we’re going to keep coming back to this clash of culture and business models. Both sides need to make some changes if we’re going to avoid doing this over, and over, and over.