Elsevier
Background
ClinicalKey AI is a clinical information tool designed to help clinicians quickly retrieve clinical answers to medical questions at the point of care. In late 2025, the product team at Elsevier planned to release a major update to ClinicalKey AI for general availability. Shortly before launch, the engineering team replaced the entire AI backend that powered response generation, accelerating the release window.
While this new backend gave the team more control and flexibility, it also introduced significant uncertainty. In healthcare, uncertainty is a risk. Even small gaps in response quality, clarity, or safety can undermine trust, slow adoption, or introduce clinical harm.
My Role
I was brought in on a contractual basis (6 months) as the sole UX Researcher responsible for evaluating and reducing this risk before launch (Dec, 2025).
I designed and led a longitudinal research study and risk mitigation workflow that helped the team understand real clinician behavior, uncover hidden failure modes, and operationalize a repeatable process for improving AI response quality.
Results
The result my work significantly impacted the response quality of the AI-powered SaaS and team efficiency.
The model: achieved 100% remediation of all high-risk query classes.
Accuracy: The model reached a 95% accuracy grade as measured by our Clinical SMEs, who benchmarked the AI responses against Elsevier’s proprietary library of medical journals.
Operational Efficiency: The engineering and product team moved from 'guessing' what clinicians wanted to a repeatable 'Clinical Quality Engine' that reduced reliance on one-off fixes through the creation of a system behavior taxonomy.
On-Time GA: Delivered actionable insights that converted to active development and product JIRA tickets within the 3-month accelerated window to meet the General Availability milestone.
