The Reliant AI team recently attended ISPOR 2025 in Montreal. The event highlighted many emerging healthcare economics and outcomes research areas, including a new research poster from Reliant. The most significant takeaways were opportunities and challenges for professionals excited to integrate AI into their workflows and replace menial tasks with room to focus on analysis and rigorous critical thought.

If you’re exploring Gen AI for systematic literature reviews (SLRs), data extraction, or broader evidence synthesis workflows, you must know that innovation is accelerating, but not without growing pains. Below, we break down what stood out to our team, what’s working, and where we go from here.

The State of Gen AI in Literature Review Tools 

The ISPOR crowds were most excited about using generative AI tools to accelerate literature reviews. Yet, we realized there are still some perception issues around generative AI. Generative AI is still perceived to be in the "Wild West" phase, with lots of promise but still waiting for that final coat of perceptionary polish. 

Many vendors are riding into town with tools claiming to streamline PICOS reviews. Most are leaning on similar tech under the hood (LLM APIs like ChatGPT). Still, few are doing meaningful differentiation. 

At Reliant, our flagship product, Tabular, has the differentiation needed to be your trusted partner in evidence synthesis, epidemiology research, and commercial drug strategy. 

Why Most Tools Sound and Feel Similar

At Reliant, we’ve seen firsthand how off-the-shelf LLMs fail when tasked with complex, domain-specific evidence synthesis. That’s why we’ve built Tabular on private, fine-tuned models—no open APIs, data leakage, or generic outputs.

The organizations who stood out at ISPOR? They’re doing the same: customizing models with domain data, not just plugging into a large language model and hoping for the best.

Fine-Tuning Is the Future 

Companies who have invested in domain-specific fine-tuning of LLMs stood out. While most tools adapt off-the-shelf LLMs, those performing model customization are better positioned to handle the complexity of HEOR use cases. The conference underscored the importance of tailored AI solutions rather than one-size-fits-all approaches.

Key Metrics and Observations at ISPOR 

‍

‍Accuracy Is Getting There—But…

Demonstrations of AI capabilities at ISPOR indicated accuracy rates ranging from 70% to 90%, contingent upon the complexity of the assigned tasks. While promising—and certainly surpassing legacy machine learning systems, AI tools aren’t ready for plug-and-play yet. These limitations become evident when systems are vetted for nuanced questions, multi-step prompts, and the crucial traceability of information and insights.

Furthermore, the persistent issue of AI hallucinations, where models generate confident yet incorrect responses, represents a significant unacknowledged business risk. This isn't merely about transparency; it's about establishing a clear, auditable lineage for every output an AI generates. The absence of such traceability profoundly magnifies the business risk when LLMs hallucinate. Without it, pinpointing why a wrong answer was given, what data influenced it, or which internal process led to the error becomes nearly impossible.

While solutions such as retrieval-augmented generation (RAG) offer some mitigation, they do not provide a complete remedy. The fundamental requirements for progress include a more robust data infrastructure and enhanced mechanisms for grounded output tracing.

One thing’s for sure: at Reliant, our Tabular offering currently makes 5% fewer errors than human-led analysis and ~50% fewer than generic AI tooling analysis. Tabular also provides traceable citations for each reference in our outputs, and our customer engagement team is happy to work with clients to understand why, what, and how information is being surfaced when generated using our product. 

A Sophistication Gap

One pattern that stood out at ISPOR was that AI literacy is all over the map. Some teams spoke fluently about agent workflows and fine-tuned pipelines, while others were still asking, “Wait, is this ChatGPT?”

This disparity highlights a critical point—technical sophistication isn’t just a bonus. It’s a prerequisite for effective AI adoption.

The data backs it up: just as reading, writing, and math defined previous generations of workplace competency, AI literacy is quickly emerging as the following essential skill set. This new fluency is becoming a true differentiator in biotech and pharma, especially among consulting, research, and strategy professionals.

Being AI-literate isn’t about becoming a prompt engineer. It’s about knowing enough to vet the right tools, challenge assumptions, and drive smarter decisions in a rapidly evolving space. To close the AI literacy gap, vendors (and organizations) must continually invest in onboarding, education, and expectation-setting to ensure the sciences and those who fuel it continue to flourish.

Why Tabular Isn’t Just Another Literature Review Tool

Tabular isn’t just built for evidence reviews—it’s designed to flex with your needs as they grow more complex. For example, Tabular isn’t limited to parsing manuscripts. It can process a wide range of tabular data, including clinical trial outputs, observational study tables, and regulatory documents, without requiring template-based formatting or hours of clean up. 

What makes Tabular powerful is the control it gives the user, which is user experience designed with commercial research in mind: for example, natural language header recognition, column-level language extraction, and the ability to apply custom prompts or filters to surface exactly what you’re looking for. Think: thousands of rows of clinical results distilled into the few endpoints you care about, or rapidly searching across MOA libraries without endless toggling between tabs. 

Whether comparing efficacy across oncology trials, scanning outputs for promising molecule candidates, or extracting adverse event rates from regulatory spreadsheets, Tabular is built to deliver accurate, editable, and traceable insights.

Final Thoughts: Early Adoption For Strategic Advantage

Gen AI is reshaping the HEOR landscape, and ISPOR 2025 proved we’re only at the beginning. The tools are improving, and the use cases are expanding. But to truly move the needle, teams need secure, accurate, and deeply tailored solutions.

If you’re betting on Gen AI to unlock new value in your research workflows, choose partners that don’t just follow trends—they shape them.

Curious About What Tabular Can Do For Your Team?

Request a personalized demo using the form below and see how we’re helping biotech and pharma teams extract more insight, faster, and with less noise.

‍