Bring information together.
Reduce the need to assemble a picture from separate worksheets and conversations.
SiteSelection / Clinical research / Product design
A recommendation.
A reason to trust it.
Helping teams move from fragmented data to an informed shortlist of clinical trial sites — with machine learning support and human review.
- Product
- Clinical trial site selection
- My scope
- User research, wireframes,
testing and high-fidelity prototypes - Collaboration
- User interviews, stakeholder presentations and design handoff
- Work shown
- Research and interactive prototype
Sketch, Anima & Zeplin
01 / The challenge
The decision was bigger
than the shortlist.
The business goal was to shorten the effort required to identify potential research sites and reduce the manual analysis involved in preparing a candidate list.
“Usually I am skeptical about the data I am handed over. I used to doublecheck them.”Interview excerpt from the project research
Speed only helps when the person making the decision can understand the evidence behind it.
The interviews described work spread across spreadsheets, repeated checking and coordination with multiple people. Users also needed clear metrics and a defensible basis for their recommendations.
Make evidence inspectable.
Give users a route from a headline prediction to the underlying history and context.
Support a reasoned choice.
Keep people involved in adjusting the shortlist and reviewing it before proceeding.
02 / Research into direction
Design around the decision,
not a generic persona.
I interviewed managers about the difficult parts of the existing process and the support they wanted from a new tool. I used those conversations, business objectives and technical constraints to shape the product direction.
Multiple worksheets
The design brings search, site profiles and comparative data into a connected environment.
Repeated data checking
Predictions and history are separate views, with site and investigator profiles available for deeper inspection.
A need for clear metrics
Recruitment, activation and historical performance become visible reference points for assessing the candidate sites.
A need to justify the choice
The shortlist can be adjusted and passed to a further human review stage.
Set the goal
Describe the study context, recruitment target and timeframe.
Explore sites
Review the proposed set or search independently.
Inspect evidence
Compare candidates and explore their historical data.
Submit for review
Pass the selected list to the next assessment stage.
03 / Design decisions
Keep the recommendation
open to scrutiny.
The interface connects a guided starting point with an adjustable shortlist and progressively deeper information.
01
Start with intent
Two routes into the same decision.
The starting screen offers machine learning guidance alongside independent site selection. Study parameters are expressed as a short sentence, making the request the centre of the interaction.
The balance: The sentence-based form offers an approachable entry point. Clear field labels, validation and a visible submit action would be priorities in a further refinement.
02
Retain control
A suggested set, not a closed answer.
The results workspace connects the proposed site list with aggregate recruitment information. The concept lets users add sites manually and inspect how the metrics change, then send the list for further review.
Human review remains part of the flow
The results screen includes an action to approve the list and send it to CSMs for a further check. The recommendation supports preparation of the decision; it does not remove the review stage.
The balance: A compact comparison view is important alongside these cards: a large set of sites should not require users to remember values across multiple screens.
03
Inspect the evidence
Move from a summary to the history.
The site profile brings together institutional information, performance metrics and study-level records. Selecting a metric exposes a more detailed chart, while the investigator profile provides a related view of research experience.
The balance: Summary metrics are useful starting points. Their definitions, time periods and source coverage need to remain accessible to support interpretation.
04
Compare in context
Make reference data part of the workflow.
Internal and external benchmarks provide a broader frame for interpreting performance. Filters and a shared table structure allow the relevant studies to be inspected rather than relying on an isolated score.
The balance: A comparison only becomes useful when users can understand what is being compared. Cohort definitions, units and missing-data states deserve the same attention as the chart itself.
04 / Testing & iteration
Prototype, test,
make the reasoning clearer.
I built an interactive prototype in Sketch and Anima and tested it with users in recurring weekly sessions. Zeplin supported the connection between design and development.
My contribution
Connect research to the interface.
My scope covered interviews, wireframes, testing and high-fidelity prototyping, as well as presenting the work to major stakeholders.
Working with others
Make the direction discussable.
The prototype provided a shared reference for user feedback and stakeholder presentations. Zeplin supported design handoff and UI documentation.
Boundary of the work shown
Designing the decision experience.
This case study describes the interface and workflow. It does not claim ownership of the machine learning model or measured improvements in its accuracy.
Next / Retrospective iteration
Clarify what the
prediction actually means.
The following is a proposed next iteration based on a review of the prototype, rather than a reported change from the original testing sessions.
Current design
The results screen gives prominent numbers for sites, patients and months. It is not explicit enough about the distinction between the user’s target and the model’s estimate.
Proposed refinement
Separate “Recruitment target” from “Estimated recruitment”. Add accessible explanations of data coverage and prediction uncertainty, where supported by the model. Show how changing the selected sites affects the estimate.
Test task
Ask a participant to explain the recommendation, remove a candidate site and describe what changed in the expected recruitment outcome.
Evidence to collect
Observe whether the participant distinguishes a target from a forecast, understands the effect of their change and can identify where more evidence is needed.
Why this iteration mattersThe research surfaced a need to double-check data. Making a forecast easier to inspect would address that concern directly, rather than asking users to trust a more prominent number.
05 / Contribution & reflection
A clearer basis
for a considered choice.
The design brings study intent, candidate sites and supporting information into a connected decision workflow.
What the work demonstrates
An end-to-end design contribution spanning research, product direction, interactive prototyping and user testing. The key design idea is to pair recommendations with opportunities for inspection and human review.
Project stage & outcomes
The work presented here covers research and a tested interactive prototype. Faster selection and lower analysis effort were business objectives; production deployment and measured time or cost savings are not established in the material available for this case study.
What I would take forward
Make sources and uncertainty explicit. Increase the readability of labels and controls. Provide compact comparison alongside cards. Use consistent metric definitions and units. These are the next refinements I would prioritise before evaluating the workflow again.
A useful recommendation helps people see both why they might accept it and what they still need to check.