When Intelligence Is Cheap, Outcomes Become the Product
Team: Sam Cialek (CAS ’11)
About the Venture: Serif provides navigation infrastructure for health experiences by helping platforms guide users through personalized next steps and pathways. The product aims to function as an interoperable layer that health services can integrate rather than building their own routing logic.
Everyone selling software to businesses should’ve seen coming some version of the events that kicked off this year.
With OpenClaw, Claude Cowork, and other agent-based tools linked to better-and-better foundational models, intelligence is beginning to commoditize and it seems assured that many classic SaaS activities – project management, taxes and accounting, drafting contracts, etc. – will soon be done by these agents for the cost of continuously cheaper tokens.
This will continually disrupt business models in ways that are sure to be painful for existing SaaS businesses, but through this pain, SaaS companies like mine will have an opportunity to help their customers in ways the pre-AI world simply didn’t allow for. To explain where these opportunities lie, consider this system for tracking the value that SaaS companies offer their users: three categories of value that are connected — context, intelligence, and outcomes.
Inspired by where venture capital was putting its money, the SaaS businesses getting funded have largely gone from “here’s a dashboard, figure it out” to “let me put my brain in your application to make it smarter.” In my view, this is the wrong spot to put your money as intelligence gets quickly commoditized. Profits mostly migrate to the adjacent stages. The winners of the commoditization of intelligence, I believe, will be companies that own the un-modularizable stages — the context where data is born and the outcomes that people use the intelligence to empower. Or, they will be companies that handle intelligence in a way that’s fundamentally different to the way an LLM can handle it.
My company, Serif, is middleware that sits in the existing health and longevity platforms. Our customers are the apps and platforms that already own the context layer — the wearable dashboards, the EHR portals, the longevity clinics that aggregate Apple Watch data, blood panels, sleep scores, and VO2 max tests into one place. Most of these platforms are in one of two stages:
Continuing to please these customers will mean needing to demonstrate clear non-commoditized intelligence. We currently do this through individualized Bayesian causal inference — a fundamentally different approach to intelligence than what an LLM provides. Where a foundation model draws on population-level training data to give you the average answer, our models treat each user as a sample size of one. They update continuously as new data arrives, build a posterior distribution specific to that individual’s biology and history, and simulate counterfactuals.
If this person shifted their sleep window by 45 minutes, what would happen to their resting heart rate over six weeks — and how confident should we be in that prediction given the data we actually have? The output isn’t a recommendation. It’s a confidence interval. That distinction is the entire product.
But non-commoditized intelligence alone isn’t enough. The real opportunity (and the one the Entrepreneurs Challenge helped us see clearly) is using that intelligence to close the loop all the way to outcomes. The general wellness consumer doesn’t have enough data to make our models meaningful or enough knowledge to know what questions to ask. The longevity community is different. They’re data-rich, knowledge-rich, and outcome-obsessed. They’ve already read the studies and tried the protocols. What an LLM tells them isn’t a premium — they’ve already internalized the population-level knowledge. What they need is specificity (or heterogeneity): does this intervention work for me, given my data?
We validated this directly. We’ve been working with a partner in EHR middleware — a company whose platform touches over 150,000 providers across Epic, Cerner, and athenahealth. With combined thousands of years of dense longitudinal health data — quarterly blood work, VO2 max testing, continuous wearable streams – Serif can provide a differentiated intelligence, moving closer to actual health outcome ownership, answering the question: given everything in my history, what should I actually change, and how certain should I be that it’ll work?
In a world where intelligence is becoming a commodity, the companies that survive will be the ones that can take data, run it through intelligence a language model can’t replicate for its use case (for the foreseeable future), and use the time they’re buying to move to truly owning their users’ outcomes. Serif will continue delivering transformation results to its users through non-commoditizable intelligence built on transfer learning, Bayesian methods, and moated data, while moving closer to realizing health outcome ownership.
