Robot chores

Can AI deliver the productivity gains pharma needs?

From robotic screenings to DNA editing treatments, technology can do remarkable things in healthcare.

But despite a surge in powerful scientific tools over recent decades, pharmaceutical R&D productivity has continued to decline.

The industry is approaching one of the largest patent cliffs in its history, with many blockbuster medicines set to lose exclusivity over the next decade.

The top 20 drugs account for nearly 30% of global pharmaceutical sales, highlighting the industry’s reliance on a relatively small number of high-earning products.

As patents expire, it will create a revenue gap worth billions.

At the same time, bringing a new drug to market remains expensive and uncertain. Around 90% of drug candidates fail during development, meaning years of research, complex regulatory requirements and costly clinical trials can still come to nothing.

Pharmaceutical companies are increasingly turning to AI to close the productivity gap. In theory, its ability to analyse vast and growing biological data sets should be showing positive results.

So why are many organisations still struggling to see measurable R&D gains despite heavy investment?

Today, 81% of pharmaceutical organisations deploy AI across their R&D programmes. This is unsurprising given the technology’s success in streamlining repetitive tasks, reducing human error and accelerating time to insight.

Most investment has focused on tools designed to accelerate drug discovery and improve target identification. Current sequencing technologies can generate enormous volumes of genomic data, creating new opportunities for AI-driven analysis at population scale.

In isolation, these models perform well. They can process enormous data sets, identify associations and generate hypotheses at scale.

But across the industry, measurable gains in productivity remain limited. Only 11% of organisations are achieving notable AI ROI.

The challenge lies not in the models themselves but in the integration, translation and measurement needed for their results to make real-world impact. AI is often deployed in silos, applied to specific stages of the R&D pipeline rather than embedded across the entire ecosystem. Data may sit on one platform, bioinformatics pipelines on another and AI models somewhere else entirely.

Moving between these systems is rarely seamless. Data often needs to be reformatted, reprocessed or manually transferred before it can be used, introducing friction and delays at every stage.

This fragmentation becomes even more pronounced when working across multiple data modalities such as genomics, proteomics, imaging and clinical data.
Each modality comes with its own formats, standards and quality levels, making integration technically complex and operationally time-consuming.

Even when models perform well, they often fall short in delivering the insights R&D teams actually need. AI can reveal correlations, but it may not provide the biological context or causal explanation behind what is driving disease.

Without a clear view of mechanism, translating model outputs into viable drug targets remains a major gap.

Ultimately, these challenges make it difficult to transform AI insights into decisions. They stall before they can influence downstream development, limiting the overall impact of AI on the R&D process.