GSK uses AI agents to cut drug-safety processing time by up to 50 percent

The data bottleneck in drug safety

Pharmaceutical companies face a significant hurdle in pharmacovigilance, which is the process of collecting and evaluating data regarding medicine safety. Currently, reports of side effects arrive through a disorganized mix of emails, paper mail, faxes, and data loggers. This creates a massive manual burden where staff must read every report, assess its priority, and decide where it belongs.

GSK has identified that the core issue is not just a slow process but a fundamental data problem. As information moves between different systems, the same details are often entered and reconciled repeatedly by different teams. This redundancy forces specialists to spend their time on manual data entry rather than making high-level decisions that require their specific expertise.

To solve this, GSK wants to change how information flows through the entire safety lifecycle. The goal is to preserve the context of a safety report as it moves from initial arrival through processing and regulatory reporting. By streamlining these transitions, the company hopes to allow its specialists to focus on the complex evaluations that require human judgment.

Moving from automation to agents

While many companies use conventional automation to handle predictable tasks, GSK and Cognizant are looking toward a more sophisticated model of specialized AI agents. Instead of relying on one general-purpose system, the proposed design coordinates dozens of agents with different responsibilities. These agents will eventually support activities across the full life cycle, including intake, coding, and retrieving supporting evidence.

This architecture allows for tasks to run in parallel, which removes the need for one team to finish their work before the next team can begin. By coordinating these agents, the companies can redesign how responsibilities move through the organization. This approach ensures that information remains consistent as it moves through various stages of evaluation and regulatory reporting.

The strategy distinguishes between activities governed by clearly defined rules and those requiring deep interpretation. Conventional automation can handle the predictable, rule-based tasks, while AI helps prepare the information needed for more complex assessments. This hybrid approach ensures that the technology supports the human experts rather than trying to replace their unique domain knowledge.

Maintaining human oversight and trust

Because drug-safety decisions carry heavy regulatory obligations, GSK is not seeking blind trust from its regulators or quality teams. The company is building traceability into its applications from the very beginning to ensure transparency. Audit trails will allow the company to reconstruct exactly how information was handled and how specific decisions were reached.

The proposed operating model includes two specific forms of oversight to manage these risks. In one scenario, AI agents prepare the work, but a qualified professional must take action before any regulated decision proceeds. In another scenario, agents carry out defined workflows while employees set the thresholds, monitor the performance, and intervene whenever the system needs a human touch.

GSK views regulation as a framework for introducing new capabilities responsibly rather than a barrier to innovation. By establishing safeguards during the development phase, the company aims to meet its obligations while still advancing its technology. This proactive approach to compliance helps ensure that the AI systems stay within their intended limits while providing value to the organization.

Measuring outcomes over software use

GSK and Cognizant are changing how they measure the success of their technology investments. They are moving away from simple metrics, such as the percentage of tasks handled by software, which do not always reflect actual improvements. A task might be completed faster by a machine, but that does not necessarily mean the safety concern was investigated or resolved more effectively.

Instead, the companies are focusing on specific outcomes, with time being the primary metric. They are targeting a 50 percent reduction in the effort required to prepare aggregate safety reports. Additionally, they aim to shorten the period between identifying a potential safety signal and reaching a final decision by 40 percent.

Another key success metric is achieving greater than 90 percent consistency in quality-control activities. By focusing on these high-level outcomes, GSK can better demonstrate the value of AI to its stakeholders. This shift ensures that the technology is actually improving the speed and reliability of drug safety rather than just automating for the sake of automation.

Empowering employees as developers

A significant part of the strategy involves involving the people who do the work every day in the development of new tools. GSK is utilizing a concept often called citizen development, where employees outside of traditional software engineering teams help create applications. These workers understand the specific pain points and inefficiencies of the current systems better than outside engineers might.

By involving these employees, GSK hopes to help them become proficient with the new technology and participate directly in improving their own workflows. This approach addresses concerns about AI eliminating jobs by repositioning workers as active participants in the technological transition. It ensures that the tools being built actually solve the problems that the staff encounters on a daily basis.

Finally, the company emphasizes reusing existing information and technologies rather than constantly rebuilding the same capabilities. Connecting data across different systems allows for new workflows to be created without forcing the replacement of every underlying application. This strategy prioritizes efficiency and continuity, allowing the company to scale its AI capabilities while maintaining its established sources of truth.