DoctorAnywhere
Provider choice relied on experienced agents remembering information spread across several tools. I designed an explainable recommendation layer that brought five matching signals into the existing booking flow while leaving the final choice with the agent.
A short project brief before the detailed story: the setting, my responsibility, the people involved and the outcome we were working towards.
Specialty, location, availability, language and coverage were reviewed in different places. Under time pressure, that made decisions slower and more dependent on individual memory. A black-box ranking or a separate tool would have introduced new risks.
Decision support should improve confidence without taking accountability away from the person making the choice.
Behavioural, operational and qualitative evidence did different jobs. Together, they helped define what the team needed to solve before choosing a direction.
Experienced agents carried provider knowledge in memory, making speed and consistency dependent on who handled the case.
The interface needed to externalise useful judgement without pretending every case had one automatic answer.
Specialty, location, availability, language and coverage were reviewed across separate information sources.
The highest-value product move was to unify the comparison at the moment of choice.
Agents were already operating inside a case flow under time pressure; switching to a standalone recommendation tool would add friction.
Decision support needed to sit inside the existing case and lead directly into booking.
The blueprint connects the customer journey, product behaviour and operational responsibilities across the same sequence.
The patient’s referral, preferences and constraints define the request.
The request is clarified before provider options are presented.
The agent confirms specialty, urgency, coverage and preference context.
Only relevant providers should remain in consideration.
Unsuitable options are filtered from the comparison.
Rules apply specialty, coverage, location, language and availability signals.
Trade-offs become easier to discuss and understand.
A ranked view exposes the most relevant options and why they fit.
Signal-level rationale supports, rather than replaces, agent judgement.
A suitable provider is selected with clear reasoning.
The agent can review, override and explain the recommendation.
The final choice and exception reasoning remain accountable.
The selected option progresses into appointment coordination.
The recommendation connects directly to booking actions.
The existing case and ticket workflow continues without a separate hand-off.
Scroll horizontally on smaller screens to follow the complete service.
The artefacts mattered when they helped the team choose. These are the questions, evidence and decisions behind the final direction.
Patient preferences, clinical context and operational exceptions could not be reduced to one deterministic answer.
Rank and explain suitable options while keeping the final decision with the agent.
A score alone would be difficult to trust, explain or override responsibly.
Expose the matching signals behind each recommendation at the comparison point.
Agents already had a complex case workflow and could not absorb another destination or tool.
Embed recommendations inside the case flow with a direct continuation into booking.
The operating service and the interface formed one system. Each lens addressed a different part of the same problem.
Shaping the ecosystem, hand-offs and operating model.
Turning service decisions into clear, usable product behaviour.
The work began with the decision agents were making: which provider is suitable for this patient, under these constraints? That clarified which signals mattered and where guidance could help.
Specialty fit, location, availability, language and coverage were brought into a ranked view so agents could compare options without switching tools.
Each option showed the signals behind the match. Agents could assess the rationale, handle exceptions and remain responsible for the final choice.
Recommendations sat beside the case context and continued directly into booking, avoiding another destination for agents to manage.
I used low-fidelity flows to test the order of questions and decision points before visual design. The high-fidelity experience then translated that logic into a branded, guided journey from patient intake to a confident specialist selection.
The first sketches modelled the concierge as a progressive conversation. Each response unlocked the next relevant question, keeping the interaction focused while showing how specialty, hospital and preferred dates would shape the recommendation.
The final direction combined conversational intake with focused task surfaces. Patients could upload a referral, choose up to two preferred time windows and compare recommended specialists without leaving the concierge journey.
These constraints affected the direction, the order of work and the compromises made with the wider team.
A recommendation is only as dependable as its availability, coverage and provider information.
The interface showed the signals being used and required agent review rather than presenting more certainty than the data supported.
A score without rationale would be hard to trust, explain or override responsibly.
Every recommendation showed why it matched, allowing the agent to compare options and use judgement where the rules were incomplete.
Extra screens or another tool could consume the time saved by ranking providers.
The feature reused case context and connected the selected provider directly to the next booking action.
This case presents validated design direction and workflow evidence. I do not claim a post-launch business metric until production measurement is available.
A memory-dependent decision became a guided comparison.
The rationale for each recommendation remained visible.
Agents could compare provider options with less cognitive load.
The capability extended the concierge platform without creating a separate tool.
I designed the recommendation as accountable decision support: useful ranking, visible rationale and human control in the same interaction.
In a sensitive service, explainability is not supporting copy. It is part of the interaction and the basis for trust and exception handling.
I would measure time to match, overrides, re-referrals, booking completion and the reasons agents reject a recommendation.