Merck and Mastercard Implement Agentic AI: Infrastructure as the Key to Success

Pharmaceutical giant Merck has achieved significant results with agentic AI, reducing drug development cycles by 33% and accelerating marketing material preparation by 70–80%. As Sean Finnerty, Vice President of Digital Platforms, noted, the key to success was preliminary infrastructure work. Without a robust foundation—cloud platforms, data management systems, and interaction protocols—agentic AI risks becoming a source of technical debt.

Merck leverages AI in three key areas: scientific research, regulated business processes, and application modernization. For example, in marketing, the technology generates draft materials that meet 99% of compliance requirements, cutting approval times from months to days. In drug development, AI helps analyze molecular structures and accelerate hypothesis testing, enabling faster market entry for therapies.

However, implementing agentic AI comes with challenges. Finnerty acknowledged that AI may generate incorrect test scenarios or suggest non-existent features. To mitigate risks, Merck employs multi-layered verification: for instance, if the first output is generated by the Claude model, it is cross-checked by Microsoft Copilot, and the result is re-evaluated. This enhances accuracy and reduces error likelihood.

At Mastercard, agentic AI is used to optimize dispute transaction processing. The payment dispute process involves multiple stages requiring analysis of both structured data (e.g., card status) and unstructured data (e.g., customer complaints). AI agents expedite decision-making, but the company must balance efficiency with reputational risk. As Chief Data Officer Andrew Reiskind emphasized, it is crucial to predefine acceptable error levels and assess their implications.