Life sciences companies seeking to harness the potential of artificial intelligence (AI) face a critical decision: Do they build in-house or buy an existing solution designed specifically for healthcare?

Choosing to build in-house, in turn, offers two routes: leverage a general large language model and attempt to customize it for established workflows or build a proprietary solution from the ground up. Numerous companies have pursued the former and built atop a foundational model to achieve a head start. Unfortunately, due to data challenges and a lack of context engineering expertise, most report that this approach fails to deliver actionable insights. Building from scratch, however, is even more daunting; It entails both significantly more time and more complex challenges, not to mention an even greater across-the-board investment.

The alternative is to purchase a fit-for-purpose solution. This approach requires in-depth investigation to ensure that the vendor has the highly specialized talent, resources, and infrastructure in place. Ultimately, choosing this route boils down to trust—in the vendor and the AI itself.

Notably, and what many companies may fail to consider, is how their approach to implementing AI also determines how much of its potential can be realized, i.e., will it be limited or unleashed? Because of the way biopharma companies are structured, AI is often launched in siloed systems, one domain at a time or even on a use-case-by-use-case basis. This piecemeal approach can limit its impact. Conversely, developing a comprehensive AI strategy at the outset establishes a new framework for insight generation across the R&D-to-commercialization continuum.

Unleashing AI’s potential

Only a miniscule percentage of promising molecules ever reach blockbuster status. Most meet their demise in the “valley of death”—that period between discovery and commercial success that encompasses Phase I, II, and III clinical trials, launch, and therapy uptake. It’s a reality that begs the question, “How can we leverage AI’s ability to bridge this gap and usher more life-changing therapies to more patients?”

When pursuing a comprehensive AI strategy, the opportunities are limitless. Had today’s AI been available just 10 years ago:

• Biopharma companies exploring the use of monoclonal antibodies for the treatment of Alzheimer's disease could have leveraged AI for patient stratification. AI can analyze patient data to find subtle subtypes and could have identified the small subset of patients for whom amyloid clearing is the right approach, thereby informing Phase III trial design.
• It could have been used in a predictive toxicology assessment for torcetrapib, which showed high efficacy but was pulled from the market due to safety concerns. AI models trained on vast chemical libraries can predict a molecule's likelihood of causing off-target effects at the earliest stages of drug design.
• A more accurate understanding of the patient journey for those diagnosed with lipoprotein lipase deficiency would have been possible, and, in turn, more realistic market-sizing for the gene therapy treatment Glybera. AI can quickly analyze immense volumes of real-world data across sources to ensure a more accurate understanding of a therapy’s market potential before committing to late-stage development.

Consider the long game

There are three additional considerations when deciding whether to build or buy.

1. Talent. Biopharma companies typically lack the specialized expertise needed for sophisticated AI engineering. This requires a deep understanding of healthcare data and how to integrate disparate sources; healthcare domain knowledge and context engineering capabilities; and mastery of the compliance and regulatory landscapes. And that's just the beginning.

Equally crucial is domain-specific expertise. For example, success in R&D requires rare "bilingual" experts—fluent in both computational science and deep, nuanced biology or chemistry—who can grasp complex disease mechanisms and properly structure and interpret “multi-omics” data (genomics, proteomics, transcriptomics). These individuals are among the most sought-after professionals in the world.

In the commercial domain, experts must understand how to interpret fragmented provider networks, complex payer landscapes, intricate patient journeys, HIPAA regulations, state privacy laws, and industry-specific marketing restrictions while maintaining transparency for regulatory scrutiny. Recruiting, training, and retaining this expertise in an intensely competitive environment is extremely challenging.

2. Speed. For companies choosing to build AI, especially from the ground up, the time frame from pilot to initial deployment of a single AI agent for a defined, limited-scope workflow can take anywhere from nine to 24 months. Building a healthcare-specific AI engine requires two to four years before achieving meaningful utility. This significant time investment is often underestimated and, ironically, delays organizations from reaping what is arguably AI’s most critical advantage: speed, i.e., the ability to accelerate and scale insight generation, data analyses, and decision-making.

3. Focus. Building and maintaining AI creates an ongoing operational burden that is far removed from drug development workflows. Ultimately, biopharma leaders must consider not only the initial development timeline but also the opportunity cost of diverting resources and attention from their primary mission of bringing therapies to patients.

What’s at stake

From early scientific exploration to commercial execution, the build-vs.-buy decision shapes how quickly and reliably teams can turn data into insights. Developing AI capabilities in-house can offer flexibility and control, but it also demands specialized talent, a mature data infrastructure, and sustained investment. Without these foundations, organizations may face slow development cycles, inconsistent model performance, and challenges in validating outputs. This, in turn, may impact the ability to make the right data- and insight-driven decisions (e.g., prioritization of targets, assessment of safety signals, identification of patient populations, and launch planning).

Even small inefficiencies can affect timelines, resource allocation, and the ability to respond to market or scientific signals. Conversely, fit-for-purpose external solutions can accelerate adoption and delivery of value, though they come with their own considerations around integration, customization, and governance.

Ultimately, what’s at stake is the organization’s broader ability to deploy AI effectively and sustainably. The question is how best to balance speed, capability, and risk so that AI strengthens scientific progress and commercial performance without overextending internal capacity.

A fast-moving train

Just one year ago, we couldn’t have imagined AI that is a thought partner, co-pilot, orchestrator, accelerator, and so much more. And who knows what it will be able to do a year from today? That’s yet another reality to contemplate when considering whether to build or buy.

Building in-house requires biopharma to recruit, continuously develop, and retain expertise; divert focus and resources to an entirely different domain; accept a slower path to implementation that may limit AI’s potential; and assume significant risk. The other option is to partner with organizations that allocate 100% of their time and focus to healthcare-specific AI analytics while teams continue doing what they do best: develop and deliver breakthrough therapies that improve patient outcomes, reduce the burden of disease, and save lives.

Rathi Suresh, PhD, is Vice President, Analytics, at Komodo Health and a member of Pharm Exec’s Editorial Advisory Board