AI

Drugmakers put workflow friction, not model speed, at the center of AI returns

GSK, Genentech and Biogen leaders described measurable local gains while questioning whether faster tasks are shortening the path to treatment.

Author
HDM Staff
Save
Share
Four participants discuss pharmaceutical AI on the Pharma Force Stage
The Clocking AI’s Real Gains panel discusses pharmaceutical workflows and patient access on November 17. Photo credit: HLTH USA 2026

    GSK is targeting trials that run 15% faster, with a more ambitious 50% acceleration goal for priority studies, but its enterprise AI leader said much of the opportunity lies in coordinating work rather than solving a new scientific problem. The distinction sharpened a discussion in which pharmaceutical executives described real improvements inside particular processes while acknowledging that enterprise productivity and faster patient access remain harder to measure.

    Madhavi Ramakrishna, GSK’s senior vice president for enterprise AI, described encouraging oncology work involving fewer trial sites and early identification of a problem that could have caused a six-month delay. She presented those examples as signs of progress, while describing the wider acceleration percentages as company goals. The panel did not establish that AI has already reduced every GSK trial’s duration by those amounts.

    The November 17 Clocking AI’s Real Gains discussion also featured Dr. Lauren Powell, Biogen’s head of health equity and clinical innovation, and Xingchu Liu, Genentech’s chief data, digital and analytics officer. Moderator Cathy Gao, a partner at Sapphire Ventures, kept returning to two different clocks: the time required to discover a candidate and the time required to deliver treatment to a patient.

    Fewer handoffs can matter more than another application

    Ramakrishna described site coordination as an accumulation of systems, information searches and preparatory tasks. A visit may require staff to assemble material from numerous places before they can do useful work. AI’s practical contribution, in her account, is to connect those steps and reduce avoidable coordination burdens, rather than simply offer another destination for a user to open.

    “Aligning to the core problems that the company is solving for is absolutely critical,” she said. That observation applied both to internal deployment and to startups selling into pharmaceutical companies. A technically impressive capability can still add work if employees must leave their normal environment, repeat information or reconcile disconnected outputs.

    GSK’s accelerator approach pairs outside companies with an internally identified problem, Ramakrishna said. She described working with smaller and sometimes stealth vendors to co-create products, expedite partner risk assessments and address integration requirements. The objective is to give a startup a route through the organization around a defined need, rather than ask it to knock on a succession of unrelated doors.

    Powell added that vendor onboarding itself is a material constraint. Biogen’s external growth and innovation team examines opportunities through therapeutic areas, but procurement and administrative processes still take time. Her advice was to account for those processes and show how a product fits the workflow, instead of treating a successful demonstration as the end of implementation.

    Local gains need an enterprise explanation

    Liu described an early period of openness to individual AI solutions followed by a more difficult realization: their outputs did not necessarily connect. A company can accumulate useful tools and still create fragmented intelligence if each application works from different context or cannot exchange information with the others.

    “The AI insights has to be consistent,” Liu said, emphasizing the need to prepare the data layer. He distinguished relatively mature applications such as content generation and conversational interfaces from the continuing work of organizing information so that answers remain coherent when the business context is the same.

    He reported substantial gains in selected areas, including doubled speed, cost reductions exceeding 80% and improved patient reach of 30% or more. Those were company-reported examples, not results from a common evaluation presented to the audience. The panel did not supply denominators, study designs or a portfolio-wide calculation that would allow the figures to be compared directly with another company’s performance.

    “The ROI is real,” Liu said, while also acknowledging that measuring individual productivity across the enterprise remains difficult. His distinction matters: a faster content workflow is a meaningful operational result, but it does not by itself demonstrate that a medicine reaches all eligible patients sooner or that total development time has fallen.

    Trial efficiency has a reach problem

    Powell described AI applications in trial-site selection, patient identification and communication with providers about available studies. She also discussed digital twins as a potential aid in rare-disease development, where recruiting patients can be especially challenging. Her broader point was that computational speed operates within a healthcare system whose access problems remain real.

    “AI can help us move faster, but it can't necessarily cancel out some of the systemic inequities and challenges we still have across the healthcare ecosystem,” Powell said. Better selection among visible sites will not necessarily identify rural organizations that are poorly represented in the information being analyzed.

    She extended that caution to patient engagement. Disease-specific chatbots and other tools can surface symptoms and information needs, helping a company understand where education may be useful. Yet those channels reach people with the connectivity, devices and data access needed to use them. Counting digital engagement without examining who is absent can make expansion look more inclusive than it is.

    “Can we get to the hardest to reach patient faster?” Powell asked when discussing return on investment. Her proposed measure changes the evaluation from how much content an organization generates to whether its new capabilities improve access for the people most likely to be missed. Faster service to already connected populations answers only part of that question.

    Adoption requires ownership of the work

    Ramakrishna described investing in staff who understand both the business domain and AI tools. These employees can help translate a capability into a meaningful change in daily work. Training alone, however, cannot repair a deployment that introduces another disconnected system, so integration and fluency have to develop together.

    She also described an opt-out approach for tools intended to support employees: the organization asks why a tool is not being used and what is failing. That creates a feedback mechanism about adoption, rather than allowing usage statistics to stand in for an explanation of whether the product is useful.

    “Introducing these tools in everyday work is the key to unlock,” Ramakrishna said. She described an accountability framework connecting business objectives to jobs, tasks and the people performing them. The harder work is redesigning the full process so that improvements at individual steps change an outcome the company actually cares about.

    Gao’s closing question about ROI exposed the panel’s common ground. Tool access and frequency of use are easy to count, while shorter trials, better decisions and wider treatment access require more disciplined measurement. The executives’ examples suggest tangible operating gains, but their own qualifications leave the central test open: whether connected workflows and broader reach can make the second clock move as convincingly as the first.