Beyond the Large Language Model Hype: Creating Sustainable Patient Value in an Era of Continuous Generative AI Evolution

Every day, we wake up to news stories about the launch of a new or improved version of a large language model (LLM) that claims to have disrupted the market. Globally, 19 different commercial and open-source LLM providers launched approximately 58 significant models in the past 12-18 months, with the last quarter of 2024 seeing the release of roughly 18 of these models. Furthermore, commercial providers are making strides toward real disruption of the LLM landscape by introducing 36 models over these 12-18 months, accounting for 62% of all releases and clearly outnumbering the 22 open-source models.

While this looks fascinating, and some AI evangelists refer to this as the beginning of the artificial general intelligence or AGI era, the current scenario and the rush to launch newer models pose the following key questions.

• How can businesses determine which models are just hype and which will actually deliver what they promise?
• How can a company go beyond conventional metrics like an LLM’s number of users or how many messages it delivers to determine if its qualitative performance meets established business needs?
• In terms of cost-value analysis, how should companies decide on the right partner for their generative AI (GenAI) needs and ensure they're investing in solutions that address real business challenges rather than chasing the latest technology?
• How should pharmaceutical companies establish AI governance frameworks for observability and behavior to comply with regulatory obligations?

With recent advances in artificial intelligence, including the sudden shift of trends toward techniques like GenAI, businesses are looking for ways to go beyond conventional BI or standard automation to add value for patients.

The pharma industry is no exception to this case, and companies are tapping into diverse functions to adopt GenAI techniques that generate value for patients. One McKinsey study¹ predicted that AI or allied technologies like GenAI could generate $60-110 billion in annual economic value for the pharmaceutical industry. For pharma companies to make the most of this opportunity, they need to do more than simply adopt the latest available model from the so-called best provider – they need to make strategic partnerships with the right partners and identify value chains across the industry. Another report, published by EY (formerly known as Ernst & Young) and Microsoft at BioAsia 2025², outlines five key strategic facets for successful AI implementation in pharma and predicts the pharmaceutical AI market will reach $16.49 billion by 2034. The report underlines the importance of enterprise-wide AI integration rather than siloed, one-off implementations.

Many life science organizations still rely heavily on complex manual processes. Sensitive data, regulatory obligations, and complicated stakeholder relationships add further complexity. However, newer models can offer powerful drug discovery and patient insights solutions. By analyzing free-text data (e.g., social media posts or scientific papers), models can help teams identify symptom patterns, emerging treatment expectations, and quality-of-life considerations—all of which feed a more patient-focused drug development process.

As LLMs evolve and mature, companies are starting to create multi-LLM agents that use GenAI and reasoning to solve complex business problems. Using multiple models with unique characteristics to uncover medical insights employs GenAI and reasoning to dive deep into customer interactions and improve the patient experience and overall care. With added support for vision, speech, and video capabilities, companies can build multi-modal systems that handle diverse data sets beyond free text to tackle complex business questions.

While businesses are experimenting with multi-modality, there is a strong buzz around using agentic AI³ as an add-on feature to a multi-modal approach that executes autonomous tasks based on the knowledge and insights generated. This approach allows businesses to have a unified system or product that gathers expertise and insights across different sub-functions within a company and allows them to make decisions and act accordingly. The medical insights team can augment this multi-modality with agentic AI to create cases in CRM systems, deliver relevant content or articles for HCPs on the fly, and request a study, etc., based on situations and context.

Choosing the Right Large Language Model for Maximum Patient Value Realization

What should pharma companies focus on?

Pharma companies should carefully navigate the LLM marketing hype by focusing on the following key aspects:

• Focus on value generation over hype: While the simmering war between model providers is fierce, businesses must encourage rapid prototyping to capture learnings for improved value generation.

• Measuring quality over quantity: Most pharma companies measure the ROI for an implementation based on user adoption, but this does not provide a transparent picture of the problem's complexity or the qualitative insights generated by using GenAI. In collaboration with the right partners and experts, businesses should measure ROI based on the complexity of the problem, integration feasibility across diverse datasets, performance benchmarking, etc., to accurately measure return on investment when implementing such use cases.

• Blending observability with the right tools: With advancements in the model provider space, regulatory authorities are releasing guidelines based on the specific region or data type to ensure that data privacy and compliance are paramount. Pharma companies should integrate a layer of “LLM observability⁵,” allowing business SMEs to make informed decisions based on observations, evaluations, and the metrics logged for each response.

• Prioritizing the key business functions for early integration: Most healthcare and life sciences companies tend to solve business problems in silos rather than building a more efficient and unified ecosystem where information and data flow freely throughout the company. In such cases, current LLMs lack the ability to consolidate and consider the end-to-end context of the entire problem space or data diversity. It is essential to note that not all problems can be solved with just a few iterations of prompt engineering.

Key takeaways and things to remember

As LLM providers continue their race to produce the best algorithms, they introduce more and more models at a frantic pace. Healthcare and life science businesses have an abundance of opportunities to explore and can now tap into the benefits of a multi-modal approach to solving complex business challenges. As the industry continues to collaborate with the most suitable partners for their GenAI integrations, regional and global governance on the use of these models in the pharma context emerges as a critical pillar for emphasizing responsible and ethical usage.

LLM observability⁵ plays a pivotal role in governance to build a foundation for transparency and accountability. Observability is typically defined as the ability to monitor, track, and audit model behavior. With the right tools for implementing LLM observability, pharma businesses can proactively flag hallucinations, inaccuracies, and vulnerabilities in model responses to ensure the usage aligns with regulatory standards and ethical considerations.

Some recent publications explore new avenues of using GenAI techniques to augment the gaps and challenges with a human-in-the-loop approach for improving performance. An interesting AI-in-the-loop feedback approach⁶ provides a promising avenue for iterative refinements to GenAI use cases. Human feedback loops are augmented with model responses, allowing businesses to refine responses and ensure continuous evolution and learning.

To conclude, the key success factor for pharma businesses is to explore the full potential of LLMs and rely on the most effective collaborations. Companies must identify opportunities to collaborate with AI service providers, regulatory bodies, and SMEs to co-innovate or develop use cases tailor-made to customer-centric needs. This collaboration framework can accelerate innovation, bridge knowledge gaps, and ensure that models are deployed responsibly to generate meaningful insights.

Pharma companies must focus on identifying and prioritizing the best use cases, embedding a governance layer so LLMs operate securely, building feedback loops, and fostering collaboration rather than getting carried away with the hype of LLMs. Choosing the right LLMs to solve key business problems is essential to successful GenAI use cases and will define the way forward for the pharma industry.

References

1. McKinsey and Company. Generative AI in the pharmaceutical industry: Moving from hype to reality. Accessed April 30, 2025. https://www.mckinsey.com/industries/life-sciences/our-insights/generative-ai-in-the-pharmaceutical-industry-moving-from-hype-to-reality
2. EY-Parthenon and Microsoft. Scaling AI in pharma: EY-Parthenon and Microsoft report charts path to enterprise adoption at BioAsia 2025. Accessed April 30, 2025. https://www.ey.com/en_in/newsroom/2025/02/scaling-ai-in-pharma-ey-parthenon-microsoft-report-charts-path-to-enterprise-adoption-at-bio-asia-2025
3. Finn T, Downie A, Agentic AI vs. generative AI. Accessed April 30, 2025. https://www.ibm.com/think/topics/agentic-ai-vs-generative-ai
4. MarketsandMarkets. Artificial Intelligence in Drug Discovery Market: Growth, Size, Share and Trends. Accessed April 30, 2025. https://www.marketsandmarkets.com/Market-Reports/ai-in-drug-discovery-market-151193446.html?gclid=Cj0KCQjwtamlBhD3ARIsAARoaExprtC63yj-WsAJqur2BY9ySUOOtg5JbCsJ5kbe_JL7QSMY8vlKfScaAvKsEALw_wcB
5. The Blue AI. Observability and Monitoring of LLMs. Accessed April 30, 2025. https://theblue.ai/blog/llm-observability-en/
6. Lee H, Phatale S, Mansoor H, et al. RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback. 2024, https://arxiv.org/abs/2309.00267