Lokavant’s Spectrum turns clinical-trial planning into a live simulation

The clinical-trial intelligence company Lokavant has debuted Spectrum, an AI-powered tool that crunches weeks of feasibility work down to minutes. Pfizer, an early adopter, now refreshes enrollment and timeline projections almost in real time. “Instead of it being five weeks to five minutes, essentially it’s a simulation every five minutes,” said Jonathan Crowther, Pfizer’s head of predictive analytics and an early Spectrum adopter.

That speed is a stark upgrade over the cobbled-together spreadsheets and static slide decks common in trial planning, Crowther added. “You have all these different spreadsheets, generally dusty spreadsheets, that people are trying to bring together … It’s almost like gambling with your clinical trial by letting people do that.”

By contrast, teams “can actually do a lot of [scenario modeling] live in front of the users,” running simulations mid-meeting to test tweaks to recruitment or site strategy. “You can have two tabs open, and you can show old and new,” Crowther said. That is, sponsors can turn feasibility planning into a dynamic exercise and slashing the hours once spent pulling numbers for study teams.

Inside the engine

Spectrum employs a mix of data analysis techniques. There are Monte Carlo simulations, Bayesian learning models and causal-AI algorithms plus generative AI that parses free-text eligibility criteria. Those techniques work in tandem to enable Spectrum to forecast trial timelines with more than 80% confidence. Under the hood sits a proprietary cache of some 2,000 site-level studies, spanning 14,000 investigators and 12,000 health centers, layered onto summary data from roughly 500,000 public trials. The platform ingests live site-activation feeds and enrollment logs, then re-forecasts on the fly whenever a sponsor tweaks countries, pauses recruitment or amends a protocol.

That agility is increasingly necessary as patient cohorts splinter. “We’re not looking for diabetes type 2 patients anymore … We’re looking for diabetes type 2 patients, comorbid chronic kidney disease, Metformin-naive,” said CEO Rohit Nambisan. “We’re narrowing and specializing in that niche space that ultimately is going to drive the timeline for the study.”

There is also the challenging of keeping up with changes in managing a specific disease over time. Crowther points to an oncology example. “Take breast cancer and think about where we were 5–10 years ago to where we are now in our understanding of the disease and the modalities. That’s changed,” he said. “So even the patient cohorts have changed, and how we analyze that. Then you throw in targeted and personalized therapies on top of that.”

As trials fragment into ever-finer strata, back-of-the-envelope spreadsheets struggle to keep up. Lokavant is betting that continuous, explainable forecasts, updated “every five minutes,” as Crowther puts it, will become the new baseline for planning study timelines and budgets.

Why transparency matters

Lokavant leans on Bayesian approaches partly because they’re easier to explain. “The advantage to Bayesian learning models is they’re built off prior knowledge,” Nambisan said. “Even if we convince the clinical operator, they have to explain it to their governance committee … The more explainable, the better.”

Causal AI layers in counterfactual math—“if X happens, what’s Y?”—to move beyond correlation. “Using causal AI … we can say, ‘This is why we’re projecting three months longer than your experience would tell you,’” Crowther said. The generative-AI module, meanwhile, turns semi-structured inclusion criteria into vector embeddings that can be compared against decades of historical trials, an automated gut-check many sponsors still try to do manually.

Human in the loop, not out of it

Spectrum’s forecasts come with editable assumptions, ensuring the user remains the ultimate decision-maker. “If a customer has more experience in one country, they can override any assumption,” Nambisan noted. This control is crucial, he added, because even AI-driven insights “have to be explained to a governance committee.” Coupled with efforts toward transparency via Bayesian and causal models (which help clarify the ‘why’ behind projections), this malleability aims to empower users, not overwhelm them. Or, as Crowther put it: “We don’t want to replace five weeks of data prep with five months of deciphering model output.”

Spectrum delivers its most significant value in complex Phase 2 and Phase 3 clinical trials. These later-stage studies, often sprawling across dozens of countries and involving highly specific patient groups, present the biggest challenges for traditional feasibility planning, magnifying the risks of inaccurate enrollment forecasts and unforeseen activation delays. “Phase 1 single-site healthy volunteer? Probably not. Phase 1/2 with patient populations? Then the glove fits,” Nambisan explained.  Early results underscore the potential: in pilot projects, sponsors reported a 70% drop in non-conformance events, a 70-fold jump in forecasting accuracy, and measurable cuts to timelines and budgets, according to Lokavant’s internal analysis.

The triad that makes it work

Technology alone doesn’t fix trial logistics, Crowther cautioned. Success requires a collaborative structure: “It’s that triad—Lokavant, embedded feasibility analytics [like Crowther’s team], and the study team—that really helps make this work.” The internal analytics group plays a supporting role, he explained, in translating complex outputs and “demystifying what’s happening with these models” for operational teams unfamiliar with vector databases or causal inference.

Crowther describes a sort of organizational mindset for senior leaders and study teams away from static assumptions and toward proactive, predictive planning. At Pfizer, Crowther said, Spectrum now serves as “GPS for feasibility,” moving discussions away from potentially misleading single-point dates. Instead, it feeds continuous risk ranges: “best case, worst case, and plan somewhere in between.” The result is data to help manage expectations more realistically and dynamically with leadership from the start.

More data, deeper links

Lokavant plans to bulk up its proprietary data set well past 2,000 studies and is exploring integrations that chain Spectrum’s forecasts to downstream budgeting tools. This granular, site-level performance data, painstakingly gathered during the company’s initial focus on the trial conduct phase, is the “asset that will continue to grow,” Nambisan said, providing ever-fresher insights into real-world operations. The company is also exploring integrations that chain Spectrum’s forecasts to downstream systems like budgeting tools, aiming for a more connected planning ecosystem. Underlying these plans is a commitment to improving the models themselves, driven by market needs, whether applying generative AI to eligibility criteria or causal AI for optimization. Nambisan, a longtime clinical-operations veteran, frames the ultimate technical goal simply: “To optimize something, you need to know not just what correlates, but what causes the other event to happen.”

Clinical trials will always involve risk: Lokavant’s pitch is to swap blind bets for live odds that refresh every five minutes.

---

Filed Under: machine learning and AI
Tagged With: Lokavant
 

---