Roche pRED · 4-year enterprise engagement · Field research in Germany, Japan, and the US
The product
IRIS is Roche’s internal research platform for their pRED division – Pharmaceutical Research and Early Development. An integrated content management and image analysis system, it brings together digitized histopathology slides, genomics data, and clinical imaging into a single environment, supporting the tumor analysis workflows that underpin drug development decisions. The platform is used by pathologists, scientists, and lab technicians across Roche’s facilities in Germany, the United States, and Japan.
The engagement
In October 2016, Impossible was brought in by Roche Diagnostics to redesign IRIS. I joined as one of two senior designers. Over four years, as the scope grew, I took the lead on design direction, client communication, and work planning, eventually bringing two additional designers into the team.
The problem
When we started, IRIS was a viewer. Technically functional, but narrow, a place to look at images, not a place to work. The actual research workflow was distributed across half a dozen systems: projects tracked in spreadsheets, communications handled by email and Jira, data visualization done in Spotfire, quality control in an HTML document buried in a network folder. No single workflow could be completed inside IRIS alone.
The people using it — pathologists, scientists, lab technicians — had learned to live with it, but they didn’t trust it. They didn’t trust that the information was correct. They didn’t trust that their quality checks had actually saved. They worked around it constantly.
The brief was to redesign the platform. But the research made clear that the real challenge was larger: to turn IRIS from a viewer into a system, one that could absorb the fragmented workflows around it and give researchers a single, trustworthy place to work.
The research
Before any design work, we went to the labs.
The first trip was to Penzberg, Germany, Roche’s main research facility, in early 2017. We spent several days running one-on-one interviews, observing workflows, sitting with pathologists as they worked, and running a kick-off workshop with the broader team. We talked to ten people: pathologists, scientists, a lab director, lead technicians. We also conducted a couple of research trips to Tokyo, and continued field research in the US throughout the engagement.
What emerged from those conversations was a detailed map of four distinct workflow types: clinical studies, preclinical studies, exploratory studies, and assay and algorithm development. Each with its own rhythm, its own stakeholders, and its own set of frustrations. The clinical workflow alone involved five different user roles, multiple handoffs, and dependencies on systems that didn’t talk to each other.
One thing I learned early: the senior people in the room are rarely the ones who understand the daily friction. A lab director describes the workflow as it’s supposed to work. A lead technician shows you how it actually works. You have to spend time with both, and know which one you’re listening to for what.
The design
IRIS Enterprise 1.0 was a ground-up redesign of the platform, built around the idea that researchers should be able to complete their entire workflow without leaving the system.
The architecture was reorganized around studies and patients rather than images, the actual units of research work. A study overview gave teams a live view of project status, participant lists, image counts, annotation progress, and QC state. A patient detail view brought together oncology journey, genomic alterations, diagnostic history, and imaging in a single scrollable record.
The image viewer was redesigned for the full annotation workflow: better tool placement, clearer feedback on saves and QC states, keyboard shortcuts to reduce mode-switching, and an approach to analysis result overlays that made them legible rather than obscuring the tissue underneath.
A new Data Explorer added biomarker visualizations: boxplots, heatmaps, cohort views, that had previously required researchers to export data to Spotfire and build their own charts. We also built a dashboard with a task manager, activity feed, and project timeline: the kind of operational visibility that had been living in spreadsheets and email.
Content management, the part of IRIS that handled the intake and organization of images and studies, was rebuilt from a patchwork of manual processes into a structured workflow with clear states, version control for scanned images, and contextual information accessible at every stage.
Throughout, we tested and iterated with the Penzberg team. The research wasn’t a one-time visit, it was an ongoing conversation that shaped every major design decision.
Growing the team
The engagement expanded as the scope did. I started as one of two designers. Over time, as the project grew in complexity and ambition, I brought two additional designers into the team and took the lead on design direction, client communication, and work planning.
Managing a design team on a multi-year enterprise project is a different kind of challenge than doing the design work itself. The hardest part isn’t the individual decision, it’s maintaining coherence across a large system over a long time, with a development team implementing in parallel and a client whose understanding of the product deepens as they use it. You spend a lot of time making sure everyone is solving the same problem.
The outcome
The engagement ran from 2016 to 2020. IRIS Enterprise 1.0 shipped, and the platform continued to be developed and used at Roche after the engagement ended. We didn’t hand off a prototype, we handed off a working system that the Roche team kept building on.
For a research platform used in drug development at one of the world’s largest pharmaceutical companies, that continuity is the outcome. A six-month SOW that Roche renewed repeatedly over four years is itself a measure of confidence in the work. The measure isn’t a metric, it’s that the work became infrastructure.
What it taught me
Designing for highly specialized professional users, like pathologists, research scientists, lab technicians in a regulated pharmaceutical environment, requires a level of domain investment that most design projects don’t. You can’t fake your way through an interview with a pathologist. You have to understand enough about what they do to ask the right questions, recognize which answers are significant, and translate them into design decisions that hold up under scrutiny.
The other lesson, learned over four years, is about trust. Not just user trust in the product, though that mattered enormously here, but the trust between a design team and a client that allows the work to evolve honestly over time. The platform IRIS became at the end of the engagement looked nothing like what was scoped in 2016. That’s not scope creep. That’s what happens when the research is good and the relationship is strong enough to follow it.