Written By Amy Sheide, Chief Data Officer, Ontada, and Wes Keller, Head of Product, Ontada

Speed in real‑world evidence (RWE) has become easier to demonstrate and harder to defend. Advances in AI have accelerated the extraction of insight from clinical data but have also raised the stakes. When evidence is reused across higher‑consequence decisions, speed no longer fails during analysis, but earlier, when teams are asked to explain where data came from, how patients were defined and whether results can be trusted beyond a single study. That trust is ultimately shaped by clear study design and well‑defined data sources.

In other words, AI has not made speed safer, but weak data riskier. AI can speed analysis, but it cannot replace clear definitions or validation, and that tension reflects a broader inflection point in oncology real-world research. While AI has made it technically possible to move faster than ever, regulators, payers and internal stakeholders are asking for evidence to support more decisions, sooner and with greater transparency. Review cycles lengthen when foundational questions resurface late, and in that environment, speed becomes fragile unless the underlying data is built to hold up under reuse and scrutiny.

What determines whether evidence holds up is whether teams can trace where data comes from and how cohorts are defined. What limits progress is not processing power, but whether the data foundation is clinically grounded, longitudinally coherent and clear enough to withstand explanation. Without a trusted data foundation, AI accelerates uncertainty as efficiently as it accelerates analysis. When the data foundation is strong, speed becomes something teams can rely on rather than defend. Even then, each use still needs to be evaluated for the specific research question.

The hidden failure point in “fast” evidence

In practice, delays rarely occur during analysis. They show up earlier, when teams try to determine whether data reflects a usable patient journey or whether results will hold up beyond a single study. We are experiencing a shift from how fast can we analyze to how can we trust what we are seeing —  a shift that often hinges on whether definitions and assumptions are clear upfront.

Trust, in turn, depends on basic clinical continuity. Large patient counts matter less than the ability to follow diagnosis timing, treatment sequencing and outcomes with confidence. When those elements are incomplete or unclear, cohorts fail under real study criteria. Time is lost revisiting definitions, provenance and clinical context that should have been resolved before analysis ever begins.

Those gaps become more consequential as evidence is reused. Data assembled to answer a narrow question rarely carries forward without friction. Each new use introduces additional scrutiny, reopening assumptions and slowing momentum. AI intensifies this dynamic by moving results downstream beyond what the underlying data can support, amplifying uncertainty with the output.

Under those conditions, throughput stops being the limiting factor. Trust becomes the constraint on speed. When disease state, treatment intent and outcomes can be followed longitudinally, questions about fit and reliability are addressed earlier. Review conversations move forward instead of looping back. AI helps surface insight more efficiently, but context and judgment determine whether that insight can be defended.

Over time, these differences compound. As new questions arise, teams return to the same data, and shifting assumptions erode confidence and make speed fragile. Stable foundations change that dynamic by anchoring work in shared definitions and data structures. When those elements are consistent, speed becomes more predictable, repeatable and easier to stand behind.

What a usable, clinical‑first foundation looks like

Speed holds up when the data can answer the next question, not just the first one. A usable foundation supports that by providing enough clinical context and longitudinal continuity for teams to trust cohorts before analysis begins, reducing the need to revisit assumptions as evidence moves into review or is applied to new decisions. Key assumptions and definitions are visible upfront, so teams can understand how the data was constructed before relying on it.

That foundation starts with clinical context
Disease state, treatment intent, sequencing and outcomes need to be established with enough coherence to support real study criteria. Without that clarity, early effort is spent testing whether a cohort will hold up rather than advancing the analysis itself.

Claims play a supporting role
When linked deliberately, claims data can strengthen continuity and fill gaps that may not be explicitly documented in structured clinical data. On their own, claims lack the detail needed to interpret patient journeys with confidence, but when paired with a clinically grounded core, they add longitudinal context that complements clinical records. Making those linkages usable depends on transparent design choices, including how records are matched, the time windows applied and how missing data are handled.

Depth matters more than raw scale
Usability depends less on how many patients are included and more on whether individual journeys can be followed over time. Longitudinal, multimodal records allow teams to examine change, sequencing and outcomes across the care continuum. That depth supports reuse as questions evolve.

Transparency determines how quickly teams can move
A usable foundation makes it clear what is included, how elements are defined and where limitations exist before analysis begins. Cohorts remain stable once real study criteria are applied, reducing rework and extended review cycles. That same transparency extends to how data are de‑identified and governed for research use, helping teams understand constraints before results are applied.

The takeaway: Speed starts upstream

Pressure to move faster in RWE is not new. What has changed is where speed is actually determined. It now lives upstream, in whether the underlying data can be reused, explained and defended as scrutiny increases. In effect, that scrutiny centers on whether the data are relevant, complete and clearly defined.

As evidence is applied across more studies and decisions, teams are asked to stand behind their assumptions for longer. Results that move quickly but cannot be reused or clearly explained introduce risk rather than advantage. In that context, speed becomes less about throughput and more about trust.

AI sharpens that distinction. It can surface insight faster, but it does not resolve gaps in clinical context or longitudinal continuity. When the underlying foundation is weak, AI accelerates uncertainty; when it is stable, results are more predictable and defensible. Even then, a strong foundation reduces rework without replacing study design or validation.

That lens underlies efforts such as ON.Journey, Ontada’s clinically grounded oncology dataset built to support reuse and defensibility as evidence moves across research and HEOR decisions. It reflects the type of foundation needed when results are expected to hold up across multiple uses, not just a single study.

The inflection point is not about adopting new tools but about building data that remains coherent as questions evolve. When teams trust the foundation early, work advances with fewer interruptions and evidence can be reused without being rebuilt. The result is speed that is more predictable and easier to stand behind.