n today’s fast-paced pharmaceutical industry, generative AI (Gen AI) is rapidly emerging as a transformative tool, much like Excel revolutionized business operations in its early days. Organizations can build a dynamic, evolving portfolio of AI capabilities for the drug development lifecycle. This article introduces a Gen AI capabilities maturity model to guide companies through the stages of AI implementation from simple foundational tasks to highly complex, data-driven workflows. We emphasize the importance of treating AI use cases as living products, evolving with business needs and technological advancements, while also addressing common challenges, such as managing expectations and leveraging existing data structures for AI optimization. This article provides a robust framework and real-world paradigms to help pharmaceutical organizations unlock Gen AI’s potential—an indispensable read for those aiming to lead in digital innovation.

Gen AI as a Foundational Capability: The Excel Analogy

When Excel first entered the business landscape, it was seen as a powerful tool for spreadsheet calculations. Over time, it became much more than that and evolved into a platform where customized solutions were built almost on the fly based on business need. The evolution of spreadsheet management tools was marked by incremental improvements, culminating in Excel’s emergence as a disruptive application that transformed collaborative resources for enterprise-level business users.

Fast-forward to today, and the same mindset can be applied to Gen AI: Rather than thinking of its justification and path as a finite database of predefined use cases, life science companies should view it as a dynamic portfolio of evolving capabilities that can be adapted to solve emerging challenges and use cases.

This thinking encourages organizations to build a set of maturing capabilities and develop a flexible approach to Gen AI applications, allowing teams to spin up new use cases as business needs evolve, just as Excel spreadsheets are used in countless ways, both in the past and today. In the life sciences and healthcare domains, Gen AI applications could range from streamlining regulatory submissions to enhancing predictive models for drug efficacy, resting upon the same foundational models and components.

Gen AI as a Living, Evolving Technology

The rapid advancements in Gen AI over the last few years are indicative of the fast-moving nature of the field. Two years ago, a Large Language Model (LLM) with an ~2,000-word (Chat GPT-3) context window was considered revolutionary. Today, models such as Gemini 1.5 Pro by Google boast context windows of ~2 million words. This capability improvement has dramatically changed the scope and complexity of potential use cases, including how business users conceptualize the “art of the possible” with Gen AI. For example, early-stage use cases for drug development might have focused on document retrieval or answering domain-specific queries. Today, the size of context windows and advancements in model fine-tuning, as well as RAG (Retrieval-Augmented Generation), enable far more mature and complex use case possibilities, such as cross-referenced data analysis across extensive drug trial reports, compliance documentation, and even research papers spanning information compiled over many years.

This progressive and complex environment begs for a pharmaceutical-focused Gen AI maturity model and roadmap framework in which to plot out how all these moving parts (changing technology, changing business needs, and changing capabilities) blend together into an outcomes-based result. A domain-focused Gen AI maturity model will assist in creating strategic alignment and prioritize the necessary building blocks, plus answer “build or buy” questions.

Gen AI Capabilities Maturity Model for Drug Development

We propose the Gen AI capabilities maturity model illustrated in Table 1. This model, starting from simple and graduating to complex, including risks, allows stakeholders to digest and position their evolving needs (Gen AI use cases) within an organizational and technological maturity roadmap. Much like product management methodologies, where products evolve through multiple versions and maturity, the recommendation is that organizations should treat their use cases as products, evolving through stages of sophistication both in linear and circular manners. The key failure points for AI projects identified in RAND’s research The Root Causes of Failure for Artificial Intelligence Projects center on misunderstanding or miscommunicating the problem to be solved, prioritizing cutting-edge technology over solving real user problems, lacking the necessary infrastructure to manage and deploy models, and attempting to apply AI to problems beyond its capabilities. The benefits of having a domain-focused Gen AI maturity model within a product approach paradigm, as in Table 1, addresses some of these failure points by providing a structured framework for aligning AI initiatives with business goals, ensuring clear problem definition, understanding risks, prioritizing user-centric solutions, and systematically building the necessary infrastructure and expertise to support scalable, impactful AI applications.

Mapping out Your Outcomes-Based Use Cases and Roadmap

Use cases are not universal. They are not turnkey. They are a function of needs, business environment, priorities, and the organizational context of technology. There may be broad domain ideas for use cases, but their implementation will vary from organization to organization. For example, “Health Assessment Questionnaires,” or HAQs, is a use case often discussed in the pharmaceutical industry. The data sources (along with its systems and pipelines) and unstructured content from Company A will vary to Company B. In addition, “how” an application is conceptualized may differ from Company A to Company B. Company A may conceptualize the use of Gen AI as a recommendation engine listing out tasks, sources, relevant stakeholders, and a recommended structured response as the output, while Company B may lean on Gen AI to draft content for a specialist to review or even aim to replace the human in the loop. Company C may tackle this use case using Gen AI with a Knowledge Graph as a quality check for the Subject Matter Expert (SME) human response. All three of these approaches to the “same” use case will require different regulatory and risk assessments.

Many organizations already have a cache of theoretical use cases, and the recommended next step is to bucket them into the maturity model. Completing this step will generate four important follow-up questions:

1. What is the trend across all the use cases? Where are they landing in terms of common capabilities?
2. How can the more advanced use cases be broken into bite-size deliverable chunks along the maturity model? What business value can be achieved with incremental builds as per a “product” approach?
3. What are the current (AS IS) states of Gen AI reliability versus expectations of the use case requirements? One recommendation is to focus on use cases with good Gen AI reliability as the lower hanging fruits, or to consider how advanced use cases can be partitioned into current reliable Gen AI capabilities.
4. What are the foundational capabilities that must be built in, in order to drive the majority of use cases? A simple Pareto Principle approach suggests that 80% of use cases may fall within one or two levels in the maturity model as a starting point.

This clustering of use cases along the maturity model will begin to paint a picture and will harmonize value creation, risk levels and tolerance, business needs, and IT efforts. With this categorization of use cases, their complexity, and the infrastructure and skills needed, etc., the organization can begin to develop a strategic risk-based Gen AI roadmap encompassing a technology roadmap, as well as change management and personnel needs.

Innovating New Use Cases

Once this “landing zone” has been established, one empowering question to ask business stakeholders can be: “Given these capabilities, what further use cases can we map to business value and user needs?”

For example, RAG (Retrieval-Augmented Generation), especially advanced RAG, is one hot topic spinning out many use cases. Once business users understand the capabilities, further use case innovation can become low-hanging fruit after assembling the appropriate technology capabilities.

Managing Expectations: Short-Term versus Long-Term Gains

As with any transformative technology, there is often a tendency to overestimate short-term gains while underestimating its long-term potential.

The temptation early is to try to “boil the ocean” with heavy-lift implementations. This usually results in some sort of paralysis. It’s important to scope low-risk use cases that can get into production. The maturity model outlined is a good tool for this enablement. For example, several companies are simply maximizing Level 1 in the maturity model: Simply getting a private instance of a Large Language Model in its own native interface, and facing it to end users, and realizing productivity gains. Moderna’s use of Open AI is an example.

The true power of Gen AI lies in its long-term, transformative capabilities. Over time, as organizations mature in their use of AI, they will unlock productivity and compliance gains that far exceed initial expectations.

Managing Expectations: The Added Value of Gen AI

Generative AI (Gen AI) can potentially be employed to perform or assist with any digital task. However, this does not mean we should rely on it to do everything. Why is that?

First, Gen AI comes with substantial costs. Just because these costs (such as GPU usage, energy consumption, and operational expenses) are not immediately visible or felt doesn’t mean they don’t exist.

Second, Gen AI providers often make enticing claims that generate high expectations, especially in professional environments where significant returns on investment are anticipated. Yet, Gen AI still presents limitations (hallucinations, biases, lack of precision, overly broad responses, and limited instruction length), which are frequently downplayed in light of these promises.

Third, Gen AI can have subtle but harmful effects, potentially diminishing human capabilities. How much assistance is beneficial, and at what point does it become a replacement rather than an enhancement of human input?

Finally, the concept of added value is highly subjective and context-dependent. As the context changes, so does the definition and assessment of value. Therefore, when considering Gen AI as a solution, a thorough and systematic evaluation of its added value should be carried out on a case-by-case basis. (This Business Value Monitoring article provides a useful framework toolkit.)

The Myth of “My Data Isn’t Gen AI Ready”

The pharmaceutical industry is a very privileged domain in the sense that it is regulated and there is already an operational expectation of quality content, such as standard operating procedures (SOPs), structured reports, operational GxP (Good Practice) data, regulatory content, real-world evidence, etc. A big chunk of these data and unstructured content is robust and structured enough to leverage Gen AI.

Rather than focus on which parts of the data are not Gen AI ready, ask the inverse question: “What systems and data structures do we already have that can leverage Gen AI for improved business value?” GxP-controlled data is already at a mature level of quality; the challenge may be that they are in different sources spread across multiple systems.

Conclusion: Gen AI as the Future of Drug Development

Generative AI is not just a trend but a foundational capability that will continue to shape the pharmaceutical industry. By adopting a dynamic, capability-focused approach, organizations can unlock unprecedented levels of productivity, compliance, and innovation in the drug development lifecycle.

Pharmaceutical companies must evolve with Gen AI, treating their use cases as living products on a maturity journey. By applying product roadmap principles and continuously iterating on use cases, businesses will ensure that they are well-positioned to capitalize on the long-term potential of Gen AI. Frameworks such as the one introduced by this article provide Gen AI users with guardrails ensuring value generation, or at least, pitfall avoidance.

As companies harness the power of Gen AI, they will transform how drugs are developed, tested, and brought to market—creating a more efficient, compliant, and innovative pharmaceutical industry for the future.