Drug development has made huge scientific progress over the last decade. The way we run clinical development has changed far less.
Timelines are still long. Costs are still high. And most of the risk is decided before a trial even begins—we just don’t treat it that way. Teams are making critical decisions about study design, patient populations, and site strategy without fully knowing whether those choices will hold up in the real world.
The problem isn’t a lack of data. It’s how that data gets used. In many cases, decisions are still based on static analyses and point-in-time views of performance. Issues show up late, usually once recruitment begins, when they’re harder and more expensive to fix. By that point, the options are limited.
AI is starting to change that. But if it only helps teams run trials more efficiently, it misses the bigger opportunity. The real value comes from improving the decisions that happen earlier, when there is still time to adjust course.
The Citeline–Recursion partnership reflects that shift. It brings together an AI-native biotech designed around digital decision-making with a clinical data foundation built to make those decisions viable in the real world.
At its core is a simple idea: decisions in clinical development should be informed by data from the outset, not tested against it later.
Designing trials that hold up
This is most visible in trial design.
A protocol can be scientifically sound and still fail the moment it meets reality. Eligibility criteria may be too narrow. Target patients may be harder to find than expected. Investigators may not perform as assumed. None of that becomes obvious until recruitment slows.
Using real-world data earlier changes that dynamic. Instead of moving forward on assumptions, teams can test feasibility upfront. They can see whether a cohort actually exists, whether a protocol is viable, and where patients and investigators are most likely to be found before the first patient enters a study.
That matters even more for companies that don’t have decades of institutional experience to fall back on. Recursion is a good example. It operates without the historical relationships and accumulated knowledge that larger pharma companies rely on, so getting those early decisions right is not optional. It has to be built into the model from the start.
AI-native doesn’t remove the need for data—it raises the bar for it.
What this starts to address are some long-standing issues in development: trial designs that don’t reflect real-world populations, recruitment strategies that start too late, and decision-making that depends too heavily on past experience rather than evidence.
Where AI actually differentiates
A lot of the conversation around AI still focuses on models. Who has the best one, who is furthest ahead, what capabilities are coming next.
In practice, that’s not where most of the advantage will come from.
Models are improving quickly and becoming more accessible. The harder problem is making sure they are working with the right context. That means linking clinical trial history, real-world data, and market insight in a way that reflects how development actually works.
That’s the role data platforms play in this shift. Not replacing models, but grounding them.
A large part of that context sits in unstructured data. Clinical notes, observational records, and patient-level detail have always contained useful signals, but they have been difficult to use at scale. AI changes that. It becomes possible to identify patterns earlier, including signals that would have taken much longer to surface through structured analysis alone.
Moving decisions upstream
The practical impact is a change in where decisions happen.
Instead of focusing only on execution, more attention moves to the choices made before a trial starts. Which programs to advance. How to design studies. Where to run them. Those decisions shape everything that follows.
Improving them does more than save time. It reduces the likelihood of failure and makes development more predictable.
It also raises expectations around trust. If decisions are being made earlier and with greater confidence, the underlying data and outputs need to be transparent. Teams need to understand where insights come from and be able to trace them back to source.
What this points to
The Citeline–Recursion partnership highlights a broader shift that is already underway.
AI-native companies are not just adding new tools to existing processes. They’re building development around data from the beginning. That requires a different kind of data foundation, one that connects context across the lifecycle and supports decisions as they are made.
The companies that move fastest won’t just execute better. They’ll start from better decisions.
If you want to find out more about Norstella is harnessing data and AI models to build smarter decision support tools, please get in touch.