AI changes the how, not the what

Sébastien, you joined BB Biotech in 2026 as its first Head AI. What did you find when you joined, and where did you see the greatest potential to further develop our existing investment process with AI?

I joined a highly experienced investment team with an approach developed over decades. Biotech was initially a new field for me. That also had an advantage: I was able to look at existing processes with a fresh perspective and quickly learn how scientific evidence is translated into an investment decision.

The philosophy, however, was crystal clear: Everything starts with the science.

The opportunity was on the technology side. The challenge was not access to information, but its fragmentation. Knowledge was distributed across people, models, documents and systems. My goal was to bring these elements together in an investment system that preserves context, continuously builds knowledge and gives the team the analytical depth it needs across the breadth required for the portfolio, without changing the fundamentals of the investment decision.

BB Biotech already had the what. What we are building is the how.

We deliberately refer to a proprietary «operating system» rather than an AI platform. What is the key difference, and why are we developing this layer ourselves?

A platform provides functionality. We are building an operating system that thinks and remembers. You can think of it as combining 30 years of accumulated data with the way our analysts understand biotechnology.

In a nod to BB Biotech, we called it «bibi». It connects things that are normally separate: news, research, catalysts, assumptions, models, decisions and outcomes. The unit of work is no longer the document, but the investment question together with the underlying evidence and reasoning.

Developing this system ourselves does not mean that we are building our own foundation model. Quite the opposite. Models are increasingly becoming interchangeable infrastructure, and we should take advantage of every improvement. What we need to own is the layer specific to us: how we map biotechnology, how evidence is linked to an investment thesis, how our analysts reason and how that reasoning evolves over time.

Clinical trial data can change an investment case very quickly. What happens in bibi when such data is released, and how does the process differ from asking a chatbot?

A chatbot starts with a question. bibi starts with the investment context.

When clinical trial data is released, we already know what matters. The system has run tens of thousands of potential outcome scenarios based on the trial design, our assumptions, the valuation and the portfolio. So even before the data is available, we have defined what a strong, weak or inconclusive outcome would mean.

Our agents draw on primary sources, compare the reported figures with the trial design and previous data, and flag anything that deviates from the expected pattern.

bibi does not try to be the first to reach a conclusion. Markets do not lack quick opinions. Our goal is to be the first to ask the right question in a structured way. What has actually changed? Which assumption has shifted? Has uncertainty increased or decreased? Is the market reacting to the headline or to something that changes the economic prospects of the drug?

At that point, the team takes over.

This last point is likely to be particularly important for many investors. If a system can analyze trials, run scenarios and challenge investment theses, where do we draw the line between technological support and the actual investment decision?

I do not think it makes sense to rigidly assign tasks to either humans or machines. That boundary will continue to shift.

A more enduring distinction is between computation and judgment. AI can analyze a trial, challenge an assumption, generate scenarios, identify inconsistencies and even make the case that we should change our view. I want it to do all of that.

But at some point, the question «What can we infer from the evidence?» becomes «What do we believe, how strong is the evidence and how much capital are we prepared to allocate to it?». That is an investment decision, and it rests with the investment team.

There is also a clear boundary when it comes to the traceability of evidence. If the system cannot show where a material statement comes from, that statement does not exist for us. Speed is useful. Speed without traceability is not.

From an investor perspective, there is another important point for me: More data and greater speed do not automatically lead to better decisions. How do we prevent AI from simply creating more noise in the biotech sector?

Noise arises when a system reads everything but does not determine what is relevant. That is why we have defined what matters to us in bibi: For every company, we want to know whether the drug works, whether it is safe and what would change our probability of success, valuation and the risk we are taking. New information is only relevant if it changes one of these assessments. Otherwise, it remains in the background.

This works because bibi understands how the individual elements are connected: A clinical outcome relates to a drug, the drug determines part of the company’s value, and the company is part of our portfolio. Without these connections, you are simply reading faster.

My benchmark for bibi is simple: If something actually changes in an investment case, the team should identify it as early as possible.

We talk a lot about efficiency and better information processing. But I am almost more interested in the long-term impact. BB Biotech has more than 30 years of investment experience. What can bibi do with that experience that was not possible before?

What excites me most is constructive disagreement. We are building bibi to push back: it should challenge a thesis, question an assumption and present all the evidence. It becomes particularly interesting when the system disagrees with an analyst who has followed a company for ten years. Sometimes the analyst is right and the system learns. Sometimes the system is right and we identify something we might otherwise have missed. In both cases, we get better. Apply that week after week and year after year to every company we analyze, and you create a continuously improving investment judgment.

So if we look three or five years ahead, where should BB Biotech’s real competitive advantage lie – in more powerful AI models or in the knowledge we build with their help?

In a few years, the models will be almost unrecognizable. That is why I do not make predictions about the technology. But our competitive advantage should never lie in the models themselves anyway. What matters is the layer around them: the context, the reasoning, the record of every decision and why we made it. In most companies, that knowledge is lost every time someone leaves. If bibi does its job, that will not happen at BB Biotech.