Getting a drug-safety report to the right specialist can take hours at British drugmaker, GSK. Information arrives through emails, paper mail, faxes and data loggers, and someone must read it, assess its priority and decide where it belongs. As it moves between systems, the same details may be entered and reconciled repeatedly.
“It’s not a process problem, it’s a data problem,” Lisa-Ann Chiriseri, director of drug safety and pharmacovigilance applications at GSK, told Newsweek.
GSK wants to change how that information moves through pharmacovigilance, the collection and evaluation of data about medicines’ safety, so specialists can spend more time making decisions that require their expertise.
Chiriseri discussed the work as part of the Seeking Value video series presented by Cognizant. She was joined by Mohammad Haque, global head of intelligent enterprise platforms and large deals at Cognizant. Gabriel Snyder, Newsweek’s executive editor, enterprise, moderated the conversation.
...
Cognizant has worked with GSK on automating drug-safety workflows. Haque said improving individual tasks has limited value when the next team still has to reconstruct or reenter the same information.
“The problem has not gone away, it has just moved,” he said.
The companies are examining how a safety report progresses from its initial arrival through processing, evaluation, regulatory reporting and other responsibilities. Their goal is to preserve information as it moves through those activities while reducing the manual steps required to prepare it for review.
GSK’s established safety systems remain a source of truth for much of the company’s information, Chiriseri said, but their existing capabilities should not determine how every workflow operates.
The company wants information to move more efficiently without necessarily replacing the underlying systems. Decisions about automation begin with identifying a problem and determining whether technology can address it reliably.
Chiriseri said some tasks are repetitive and predictable, even when highly trained specialists currently perform them.
GSK uses both AI-based and conventional automation to classify incoming safety reports, establish priorities and direct information to the appropriate specialists. Structuring the data as it arrives also helps preserve the context needed for subsequent checks and reviews.
Haque sees a larger role for specialized AI agents in connecting those activities across the pharmacovigilance process.
He outlined a proposed operating model combining Cognizant’s services with an AI-based platform covering seven interconnected areas of drug-safety operations. The activities extend from receiving and processing reports to evaluating potential safety signals, preparing regulatory information and supporting inspections.
Rather than relying on one general-purpose system, the design would coordinate agents with different responsibilities. Haque estimated that approximately 40 specialized agents could eventually support activities across the full life cycle, including intake, coding, checking information and retrieving supporting evidence.
Some tasks could run in parallel, reducing the need to wait for one team to finish before another begins.
GSK and Cognizant have already pursued narrower automation initiatives, while the coordinated agent architecture remains part of the longer-term plan. Haque described it as a way to redesign how information and responsibilities move through the organization.
Human judgment remains a central consideration because drug-safety decisions carry regulatory obligations.
He distinguished activities governed by clearly defined rules from those requiring interpretation and domain knowledge. Conventional automation can handle many predictable tasks, while AI may help prepare information that requires a more complex assessment.
The proposed model includes two forms of oversight. In one, agents prepare work, but a qualified professional must act before a regulated decision proceeds. In the other, agents carry out defined workflows while employees set thresholds, monitor performance and intervene when necessary.
“We’re not asking for blind trust from our regulators and quality and compliance teams,” Chiriseri said.
GSK is building traceability into its applications from the outset. Audit trails make it possible to reconstruct how information was handled and decisions were reached, while validation and ongoing monitoring help keep systems within their intended limits.
She sees regulation as a framework for introducing new capabilities responsibly. Establishing appropriate safeguards during development, rather than waiting until a system is ready for deployment, can help companies pursue innovation while meeting their obligations.
The companies are also reconsidering how they assess the value of their technology investments.
Haque said conventional automation measures, such as the percentage of activities handled by software, do not capture the full benefit of changes to drug-safety operations. A task may be completed faster without substantially improving how quickly a safety concern is investigated or resolved.
“We start measuring the outcomes,” he said. “And the big one is time.”
He described targets that include reducing case processing time and the effort required to prepare aggregate safety reports by 50 percent, along with shortening the period between identifying a potential safety signal and reaching a decision by 40 percent.
Cognizant is also targeting greater than 90 percent consistency in quality-control activities.
Chiriseri said GSK is moving beyond pilots toward production use, with early results from its work with Cognizant giving her confidence in the approach.
Another measure of success, Chiriseri said, is how employees adapt to the technology.
She acknowledged concerns about AI eliminating jobs but said GSK wants workers to become proficient with the technology and participate directly in improving their own processes.
The effort includes what businesses often call citizen development, in which employees outside traditional software engineering teams help create applications or automated workflows.
She sees the value in involving those employees because they understand where existing systems create unnecessary work.
“These are the people who do the job every day,” she said. “They know the pain point.”
Chiriseri also emphasized reusing existing information and technologies rather than continually rebuilding the same capabilities. Connecting data across systems can support new workflows without forcing every application to be replaced.
Haque offered similar advice to other companies operating under extensive regulatory oversight. He urged them to map their business processes and data before introducing AI agents.
“Let the agents do the predictable work, make every decision visible and keep people accountable by design,” he said.