AI in Fertility: It’s Still Rock and Roll To Me

Late-night thoughts on innovation in IVF while listening to The Wild, The Innocent and the E Street Shuffle. A first draft.

My questions:

1. Do I really get what “artificial intelligence” is, and how it gets us closer to a million IVF babies a month?
2. When will it answer the “so what?” test?
3. How do we go from “AI can do” to “AI is doing”?
4. How much does it cost to get there and who is going to pay for it?

“AI” has been used to describe:

1) database management with massive data sets
2) complex regression
3) mechanism-agnostic correlation/causation discovery from previously unmeasured but still observed phenomena

Of the three, only the last seems “intelligent” in so far as it leverages data we have been collecting and acting upon for decades to take us places we never would have thought to go, when the computer digests every data packet generated by human, chemical, machine, media, gamete, zygote and embryo during the cycle, then taps us on our shoulders and says, “hey, look over here,” pointing to a relationship we never would have noticed, much less measured.

Which creates a new knowledge on/off ramp in the IVF superhighway system, creating an isolatable and optimizable step in the blastocyst manufacturing process.

And with each new data on/off ramp, a more sophisticated way to Engineer The Hell Out Of IVF.

I gave two talks, one on the supply dynamics of the IVF industry, talking about how affordability and access to IVF is more than just waving a magic price-cutting wand to relieve the financial burden on patients, but also creating an ecosystem where the supply expands organically, which brings more and more patients into the fold — closer to home, and less life disruptive, the two non-dollar/pound/euro/yen/rupee parts of the Sable-Sirus score.

The ecosystem changes with each new tool we pull out of the standardization/automation/optimization toolbox: software that makes our decisions faster and no worse than we have done in the past (and eventually points out the mistakes we have made over and over again), hardware that works twenty-four hours a day without getting tired, QA’s and QC’s itself and reports to executive-functioning embryologists and doctors, who together tweak and polish and evolve a system of interlocking parallel and serial jobs — humans at the top, data in the middle, machines at the bottom getting their little polished titanium hands dirty — a system that can breathe in and out in size in response to how many babies the world needs and wants but cannot otherwise produce at a given time.

In other words, a sophisticated cell biology version of what every scalable, scaled industry, from cars to corn flakes, undergoes.

The trouble starts when the engineers go to lunch and the fund managers enter the room. IVF 2024 is still an oligopoly world, networks competing on everything except price, incentivized to enthusiastically grow but grudgingly innovate, or at least pay to innovate if returns from that payment don’t flip to accretive before the proceeds from Fund III or Fund IV need to be disbursed.

So IVF programs build brand and create proprietary patient supply chains, from referring doctor groups, employer-coverage agreements, or through Tik Tok awareness campaigns, but not really by competing on price (minus a few exceptions.) And this is not only a US phenomenon; IVF the pricing band is quite narrow in every country.

This equilibrium but not equitable pricing model keeps will continue as long as the barriers to scale, the system inputs that are in shortage keep the costs of production **and market entry** artificially high. Those inputs are labor (doctors and scientists), real estate (expensive laboratory space), and the inflated costs of pharmaceutical distribution for generic drugs.

Until these factors are addressed, the benefits of cheaper and more effective blastocyst manufacturing will go wherever the limited number of suppliers decide they want them to go — and the laws of freakonomics suggest that much if not most will go into increased margin and EBITDA inflation rather than scale.

My second talk addressed AI-IVF startup financing, and my message was not cheery. After a mini course in debt, convertible debt, private equity and venture capital, I shared my unwelcome observation that IVF tech companies might be the complex translocations of the start-up world: their returns too small and difficult to model for venture capital, their cash generation timeline too long for private equity’s 10 year fund artifact, and the value proposition too obscure to create an Excel model populated with defendable numbers.

Much more on this later.

The AI-Fertility conference is a late night club gig in between the stadium tours of ESHRE and ASRM. It was dinner with Simon and David and Alison and Rachel and Rachel, it was telling myself to shut up and learn when talking to Klaus and Christos and Charlie and Kathy (or anyone connected to an embryology lab really), being thankful to Hyejun for traveling a lot further than I did to share her work in Korea, to Eduardo and Richard for getting right to the point and asking me where we are and where we’re going, and most of all for having the opportunity to re-connect with Zev Rosenwaks, who taught me more about IVF than anyone has ever taught me about anything.

Thank you Nikita and Christina for organizing all of this.