B2B ML Solutions: UXR for a multi-organization health platform and clinical discovery

In March 2021, Highmark Health — one of the largest integrated delivery networks in the US, serving 5.6+ million members — announced a strategic collaboration with Verily and Google Cloud to build the Living Health model. The goal was to eliminate fragmentation in healthcare by delivering proactive, personalized care through a digitally-enabled platform combining Verily's technology, Google Cloud's infrastructure, and Highmark's clinical expertise.

The initial focus areas were complex chronic conditions: congestive heart failure and COPD. These aren't simple consumer products — they involve multi-user systems spanning patients, clinicians, care teams, and administrative staff, all of whom have different mental models, workflows, and definitions of "a good outcome."

I joined as a UX Researcher at Verily to deliver the foundational UXR that would inform the ML data architecture — specifically, helping the team understand when and how clinical interventions happen, and what data would be needed to automate or augment those decisions at scale.

The Living Health platform was being built on a core premise: that if you can identify the right intervention windows — the moments in a patient's health journey where a timely, personalized action would have the most impact — you can meaningfully improve outcomes and reduce costs.

But identifying those windows requires understanding the clinical and behavioral reality of chronic disease management. ML models are only as good as the data fed into them, and the data strategy depends on knowing: what decisions do clinicians actually make, when do they make them, what information do they need, and where does the current system fail them?

This was B2B research with a technically sophisticated audience. The stakeholders weren't end consumers managing their own health — they were clinicians, care coordinators, and health system administrators making high-stakes decisions under time pressure, with fragmented data and limited context. Understanding their workflows, mental models, and pain points was the essential precondition for building a platform that would actually be used.

Working within the terms of a multi-million dollar Statement of Work, I planned and delivered three foundational UXR studies — managing participant recruitment, vendor communications, and research operations end-to-end.

The most impactful output was the identification of key intervention windows — specific moments in the care journey where a timely, contextually appropriate action had an outsized probability of improving outcomes. These weren't obvious from the medical literature alone; they emerged from understanding the lived experience of both clinicians and patients navigating a fragmented system.

The clinical decision trees documented how clinicians actually reason when managing chronic conditions — what information they look for, in what order, and what absence of information causes them to default to lower-quality decisions. These trees became the behavioral scaffolding for how the ML model would structure its predictions and recommendations.

Critically, the research also surfaced opportunities for cost and time savings in clinical workflows — places where the platform could reduce administrative burden and free clinicians to focus on the decisions that actually require human judgment.

This project fundamentally shaped how I think about research for ML products. The temptation in ML product development is to ask "what data do we have?" and build models from there. The better question — the one that produces ML systems that actually get used — is "what decisions do humans need to make, what information would change those decisions, and what does the absence of that information cost?"

Research on clinical populations also requires a different kind of care. The stakes of a poorly-designed intervention in chronic disease management aren't just friction — they're real clinical risk. That weight clarified something important: UX research in regulated health contexts isn't just about usability. It's about understanding the full ecosystem of decision-making that the product is entering, and ensuring the technology augments rather than disrupts clinical judgment.

Delivering against a multi-million dollar Statement of Work as a sole researcher also required a level of research operations rigor — participant management, vendor coordination, quality control, timeline management — that I've carried into every project since.