R adoption at scale: How GSK moved 3,000 biostatisticians to open source

Q: What does it take to migrate biostatisticians from SAS to R in a pharma environment?

Simply provisioning platforms and granting access to open-source packages is not enough to change daily habits. Practitioners frequently stall because they lack complete, end-to-end templates showing how to execute their exact, regulated pipeline steps in a new language. Adoption scales when you replace generic software instruction with role-specific, step-by-step example workflows.

Q: How do you build confidence and trust in statistical programming teams migrating from legacy, proprietary systems?

Confidence is built through immediate, low-friction user support rather than rigid ticketing channels. By establishing a dedicated team with a strict no-refusal policy, GSK ensured that programmers never felt stranded when a script failed. Direct mentoring on daily code bugs provides the support users need to transition away from systems they have relied on for decades.

Q: How do you design an open source program in pharma that does not create permanent support dependency?

The program must be designed with an explicit end date, shifting the burden of knowledge from a central support group to the wider community. GSK achieved this by making training integral. Users received advanced training in exchange for acting as local subject matter experts in their respective therapeutic areas. Pairing this network with community forums and weekly technical sharing sessions allows the organization to grow in a self-sustaining manner.

Enterprise R adoption in life sciences requires more than package access. GSK's STAR program moved 3,000 biostatisticians from SAS to R by building an infrastructure that prioritized real-time support and role-specific workflows over generic training software.

Ben Arancibia

Director of Data Science

GSK

What you'll take away from this session

Open source adoption fails at the daily workflow layer rather than the tooling layer

Users stall when they don’t know the specific commands to write to produce standard trial deliverables. Success requires role-specific, end-to-end workflows that map directly to their pipeline tasks.

A strict no-refusal support policy is what converts low-confidence users in regulated environments

GSK established a dedicated "Starburst" team that was tasked with answering every pipeline question without exception. Resolving basic execution errors rapidly is the fastest way to build trust.

Documenting individual user problems publicly builds institutional knowledge faster than structured courses

Instead of keeping solutions siloed in private threads, the support team published one microblog post per answered question every day. Answering an issue once allows hundreds of other programmers with the same blocker to self-serve.

Tiger teams solve deep statistical methodology gaps that generic instruction can’t touch

GSK deployed small, cross-disciplinary specialist teams to solve high-friction bottlenecks. They used this framework to build R-to-SAS methodology mappings, streamline code environment sharing with external contractors, and create debugging checklists.

Self-sustaining open source adoption requires four things

Self-sustaining pharma open source adoption requires closing skills gaps, deepening proficiency, continuous learning, and open source culture. GSK built community forums, expert networks, and mentorship programs designed to meet those criteria after the STAR program ends.

GSK has a mandate already in motion: transition 3,000 biostatisticians and statistical programmers to open source and hit a senior leadership commitment of 70% open-source code across R&D. While the organization had experimented with open source since 2017 and validated foundational GxP tools, scaling that momentum across thousands of users introduced friction. Programmers frequently didn’t know how to translate their deeply ingrained routines into an unfamiliar, multi-package R ecosystem.

Ben Arancibia, Director of Data Science within the Statistical Data Science Innovation Hub, details how GSK confronted this challenge. The organization realized that true adoption isn't about tooling. It’s a workflow problem that requires a high-touch, human response. They launched a three-month experiment centered on a high-touch support group called "Starburst," operating under a strict mandate to never say no to any user request, regardless of its complexity. The team bypassed standard ticketing queues, connecting directly with users via email to mentor them through error screens and build code fluency in real time.

To capture this localized knowledge, the Starburst team published a daily microblog post with every solution, creating an active internal repository that quickly reached over 800 unique viewers. As patterns emerged from these queries, GSK began building internal data products, such as a standardized library of tables, figures, and listings (TFLs). Ben details the scaling arc that followed:

Expanding from the central squad to a network of local subject matter experts
Establishing a community forum that operates like an internal Stack Overflow
Deploying dedicated tiger teams to map complex SAS methodologies directly into R scripts.

This architecture successfully shifted GSK’s programming practice into a self-sustaining open-source culture before the formal program reached its conclusion.

FAQ

What does it take to migrate biostatisticians from SAS to R in a pharma environment?

How do you build confidence and trust in statistical programming teams migrating from legacy, proprietary systems?

How do you design an open source program in pharma that does not create permanent support dependency?

Transform the work that matters most

See how Domino helps the world’s most regulated enterprises build, scale, and govern AI-powered applications.

Watch demo Talk to an expert