Biopharmaceutical AI: Data-Driven Failures and Successes

The Failure of AI Projects in Biopharmaceuticals: A Data Problem

In the biopharmaceutical sector, the implementation of artificial intelligence (AI) faces a major obstacle: data reliability. Indeed, nine out of ten projects fail primarily due to inadequate data. To reverse this trend, a globally harmonized data foundation is essential to convert skeptics into true ambassadors of AI.

For several years, biopharmaceutical companies have been accumulating vast amounts of data. With the rise of AI, this data must be transformed into a strategic asset. However, despite the enthusiasm for AI, 95% of companies in the sector are pursuing initiatives in this area, yet 89% of them fail to industrialize more than half of their pilot projects. This lack of success is often attributed to insufficient data maturity, leading 67% of executives to abandon their projects.

Data Quality: A Major Barrier to AI Industrialization

Data quality is crucial for any AI initiative. However, 73% of executives believe that poor data quality is the main barrier to AI industrialization. Most companies attempt to harmonize siloed data sources, an approach that becomes unsustainable in the face of continuously increasing data volumes. The data scientists we spoke with spend up to 80% of their time preparing and cleaning data to make it usable.

For AI to move from the stage of aborted pilot projects to creating value, a paradigm shift is necessary. This involves transitioning from fragmented data management to a globally harmonized foundation, supported by proactive governance.

Skepticism Towards AI: A Barrier to Adoption

The sector's reliance on imperfect and fragmented data has led to growing skepticism. Indeed, 96% of executives believe their data is not ready for AI. This skepticism manifests on the ground, where 72% of companies consider using AI to synthesize updates on healthcare professionals, but adoption remains limited.

Field teams often reject AI recommendations because they do not trust the underlying data. For example, when a "next-best-action" model suggests an action based on a three-month-old affiliation change, the sales representative does not just ignore the suggestion; they lose trust in the entire platform.

Erika Husing, a business analyst in commercial operations at GSK, sums up the situation well: "If we don't trust the data, how can we draw conclusions from it? It is essential that we move from the current skepticism towards data to a genuine promotion of data."

The Cost of Manual Data Governance

Another major challenge lies in the considerable resources devoted to manual data management. Companies spend countless hours manually mapping local specialties and healthcare professional typologies according to global standards, which represents a significant administrative burden with each market opening.

In a leading biopharmaceutical company, data scientists ask field teams to review customer segmentation data every six months, diverting them from their core mission. Even more frustrating, this company estimates that only 10% of its data is clean enough to be used, and only 1% is actually utilized. Faced with the growing volumes of data, companies can no longer afford to address these issues at a local level.

Towards a Globally Harmonized Data Foundation

Data management is inherently complex, as local subsidiaries maintain their data differently to comply with regional regulations. This creates a fragmented system that makes transnational analysis and AI particularly challenging.

Before implementing a global data model, Bayer AG faced inconsistent definitions and a lack of a unified customer view. "Our global data landscape was fragmented — different countries relied on different sources," explains Stefan Schmidt, head of digital capabilities at Bayer. "To have a complete view, we needed a unified customer reference."

For Bayer, this centralized foundation provided a single source of truth and strengthened trust in the insights generated by AI. Field teams are now less likely to question the system and more willing to use its recommendations.

Maintaining High-Integrity Data Through Agentic Curation

For decades, the sector has relied on manual governance to maintain data quality. Today, a new opportunity arises: elevating the quality of millions of records through a combination of human expertise and agentic curation.

AI agents take on repetitive tasks, such as cross-referencing sources and detecting duplicates, by examining 100% of records daily. A human data steward then validates the results. Operating continuously, these agents capture changes immediately, often before they even appear in public records. This real-time accuracy allows for relevant recommendations and avoids the common pitfall of informing teams about events they are already aware of.

By shifting the burden of curation to autonomous agents, we move from a reactive model — which breeds skepticism — to a proactive model. Agentic curation combined with human governance provides the verified and reliable data necessary to industrialize AI.

You cannot industrialize what you do not trust, and you cannot trust what you have not harmonized. A consistent global data foundation allows biopharmaceutical companies to focus on leveraging data rather than cleaning it. This refocusing will transform the biggest skeptics of AI into convinced ambassadors.