# AI in Healthcare: Can Algorithms Become the Next Pharmaceutical Breakthrough?

Author: Norbert Biedrzycki
Published: 2020-12-15
Language: en
URL: https://norbertbiedrzycki.pl/en/ai-in-healthcare-can-algorithms-become-the-next-pharmaceutical/
Summary: Biology has always been the industry where money goes in for a decade before anything comes out. AI promises to change that clock. Investors should test that promise carefully.

**Biology has always been the industry where money goes in for a decade before anything comes out. AI promises to change that clock. Investors should test that promise carefully.**

If 2020 has taught the world anything, it is that the speed of scientific discovery is a matter of economic survival. Vaccines are being developed in under a year, something that would have sounded absurd twelve months ago. Against that background, it is no surprise that investors are looking at AI in healthcare with new urgency.

**Why the economics are so attractive**

The pharmaceutical industry has a well documented productivity problem. Estimates of the cost of bringing a new drug to market run to around $2.6 billion when failures are included, and the process typically takes more than ten years. Analysts even coined a name for the trend, Eroom's Law, Moore's Law spelled backwards, to describe how the number of new drugs per billion dollars of R&D has fallen for decades.

AI offers a seductive proposition: combine the scalability of software with the enormous value of a successful molecule. If algorithms can narrow the search space, predict toxicity earlier or identify better trial patients, even modest improvements in the probability of success translate into billions of dollars of value.

This year has delivered real evidence. In January, Exscientia and Sumitomo Dainippon Pharma announced that a molecule designed with AI, DSP-1181, had entered human clinical trials, after an exploratory research phase of around twelve months rather than the usual four to five years. Schrödinger, Relay Therapeutics and AbCellera completed successful IPOs. Recursion raised a large Series D. AlphaFold's breakthrough in protein structure prediction has opened a conversation about how much of early stage biology could eventually be computed rather than discovered in the lab.

**The gap between science and business**

Having spent much of my career around technology investments, I have learned to separate two questions that are easily confused: is the science real, and is the business real?

In AI drug discovery the science is increasingly real. The business is less proven. A molecule designed faster is still a molecule that must pass Phase I, Phase II and Phase III trials. Most failures in pharma happen not because the molecule was hard to find, but because it does not work well enough in humans or turns out to be unsafe. AI has so far mainly accelerated the cheapest and earliest part of the process. The most expensive part remains largely untouched.

The same applies to diagnostics and medical imaging. Algorithms that read X-rays or retinal scans with impressive accuracy exist. But a hospital does not buy accuracy. It buys a product that fits clinical workflow, is approved by regulators, is reimbursed by payers and protects the clinician from liability. Each of those is a separate hurdle, and each takes time.

**Business models to watch**

I see three emerging models among AI healthcare companies.

Platform providers sell software and services to pharmaceutical companies. They have recurring revenue but limited upside, because the value of a successful drug largely stays with the client.

Asset builders use their platform to develop their own drug candidates. They take on clinical risk but capture the full value of success. Valuations here behave more like biotech than software.

Hybrids partner with large pharma on milestone and royalty agreements, trading some upside for capital and validation.

For an investor, the key is to understand which model a company really runs, regardless of how it describes itself. A software multiple on a biotech risk profile is a dangerous combination.

**What would change my mind**

The decisive evidence will come over the next five years: do AI-designed molecules succeed in later stage trials at a higher rate than traditionally discovered ones? If the answer is yes, the economics of pharmaceutical research will be rewritten and the companies that own the best data and models will become extremely valuable. If the answer is no, AI will still be useful, but it will be a productivity tool rather than a new industry.

*Investment question: Can AI fundamentally change the economics of pharmaceutical research and create a new generation of high-value biotech companies?*
