ClinicalKey AI is the AI component of Elsevier’s ClinicalKey platform, a desktop-based research and content delivery tool used by clinicians to find and access medical information and evidence. While ClinicalKey AI already supported AI-assisted research, its primary role was still as a reference and information retrieval tool.
Our goal was to explore how ClinicalKey AI could evolve from a reference and research tool into a more active clinical decision support experience. We experimented with different AI capabilities, workflows, and product concepts to understand where the existing technology could meaningfully support physicians during clinical decision-making.
1
Initial Questions
Our team was asked to investigate three foundational questions:
What enterprise-level analytics would provide the greatest value?
What patient data is most valuable to an external CDS tool?
We entered discovery expecting to answer three strategic questions. Instead, we discovered the questions themselves weren't well defined. Different stakeholders were describing different versions of success. Product leaders emphasized AI capabilities, clinicians discussed workflow friction, and commercial teams focused on differentiation. Each perspective was valid, but they weren't pointing toward the same outcome.
Orienting to the problem space
To align stakeholders around a common goal, we facilitated a future visioning workshop centered on defining success. We synthesized the team's expectations into shared themes, transforming several competing priorities into a single strategic focus: agentic AI solutions for general practitioner physicians.
Misalignment
Each stakeholder had a competing definition of success.
Bridging the Gaps
We consolodated all our project information and began to look for patterns. We engaged with our stakeholders regularly as we sorted the data via workshops and alignment sessions.
Shared Direction
By holding brainstorming sessions and future visioning workshops we established a shared vision for the project.
2
Discovery
With a clearer understanding of the opportunity space and the physicians we were designing for, the next step was grounding our direction in real clinical workflows. I designed and ran 8 co-design sessions with generalist physicians to uncover their needs, challenges, and opportunities for AI-assisted decision support.
Rather than asking abstractly "what do you want from AI," participants worked through a structured matching exercise, pairing real clinical use cases (drawn from prior research and jobs-to-be-done work) against candidate AI capabilities. By doing this, we were able to sift through the chatter of what physicians wanted and identify the capabilities that clustered around the same underlying need.
We then took these use cases and capabilities and mapped them together to identify 4 concept area's to test
Four use cases emerged as recurring themes
What's the most likely diagnosis, and what workup confirms or rules it out?
What treatment or management option is right for this patient, right now?
What's the best next step — treat, test, escalate, observe, or change course?
How should these labs, imaging findings, or risk scores be read in context?
Four capabilities emerged as priorities
Surfacing missing, incomplete, or conflicting information and what to check next
Personalizing insights to how the clinician actually assesses and manages patients
Anticipating the next clinical question or decision point
Highlighting which pieces of patient data are actually relevant right now
3
Concept testing
Validating Clinical Perspectives
A CDS experience designed to improve decision quality, confidence, and efficiency.
Support more informed clinical decisions
Strengthen evidence-backed recommendations
Improve defensibility of decisions and reduce denials
Decrease time-to-decision and operational burden
Bridging workup gaps
A CDS experience designed to reduce unwarranted variation and support better clinical decisions
Strengthen medical necessity justification for pathway deviations
Promote safer, more consistent clinical decisions
Prevent avoidable readmissions through earlier intervention
Minimize diagnostic delays with better decision support
4
Validated Direction
Workup Recommendations
Combined workup gaps and clinical validation into a focused recommendation experience that provides actionable guidance while preserving clinician judgment
Impact
Validated Product Direction
This work transformed an ambiguous AI opportunity into a validated product direction, giving embedded teams the foundation to move into MVP development and real-world testing.
Reduced Product Uncertainty
Created shared alignment across product, clinical, and business stakeholders around the future direction of AI-powered clinical decision support.
Defined Target Users & Use Cases
We identified general physicians as the initial target users and validated the highest-impact use cases for a third-party clinical decision support tool.
