Good Doctor — Clinical Workbench Redesign

01Impact first

The old workbench worked. It just made doctors fight the tool instead of focusing on the patient: twelve clicks and six-plus minutes for a single prescription, patient details vanishing mid-scroll, drug searches that only revealed a stock-out at the final preview.

The redesign attacked the friction, not the features. Patient context now stays pinned through the whole consult, diagnosis is clickable instead of keyboard-only, and drug search knows what the nearby pharmacy has in stock before the doctor commits. In validation, 60% of doctors completed a full prescription in under four minutes, down from six-plus, and satisfaction moved from 3.4 to 4.6 out of 5.

"Feels more like how I actually work in a real clinic." — Internal doctor, validation session

02The problem

85% of doctors needed at least six minutes to send one prescription. The workbench asked them to do their most careful work in their least forgiving conditions: a live queue, a ticking timer, and a tool that fought back at every step.

Diagnosis demanded the keyboard. ICD-X codes couldn't be clicked, mixed English and Bahasa, and weren't WHO-standard, so doctors kept a Google tab open to look them up. Drug search returned every medicine from every nearby pharmacy even when only two matched, and stock-outs surfaced only at the preview step. Patient details disappeared the moment the chat scrolled, and there was no queue alert, no read receipt, and no way to fix a mistyped message.

"I open at least four windows and still feel unsure if I missed something." — Internal doctor, discovery interview

03What I led

I ran a six-phase sprint process with clinicians as co-designers: interviews with internal doctors, log analysis from 34+ users, and a full journey map from opening the clinic to sending prescriptions. The highest-impact decisions:

· Sticky patient context (photo, complaint, chat) that stays visible during scroll
· Clickable ICD-X auto-complete to replace keyboard-only diagnosis input
· Drug search filtered by live pharmacy inventory, saving 2-3 minutes per session
· Allergy fields defaulting to "None", removing four clicks from every consultation
· Real-time queue alerts and read receipts, borrowed from messaging patterns doctors already trust

04How we understood the work

We opened with one How-Might-We: how might we help internal doctors deliver efficient, high-volume consultations backed by a reliable health-supply channel? That split into three fronts: deliver efficiently, serve many patients at once, and make it easy to find in-stock drugs near the patient.

The method was stakeholder and doctor interviews, the 34-doctor survey, session and log review, and a full experience journey map across every step of a consult. Mapping the emotional journey exposed the real problem: the tool turned "prepared" doctors into "overwhelmed" and "frustrated" ones, most sharply at diagnosis and online prescription. Rather than invent personas, I anchored on the doctors' real KPI pressure. They're measured on serving more patients than the system's own cap, so every extra click compounds across an entire shift.

Synthesized insights (grid)
· "ICD-X translator isn't a WHO standard." Doctors hunted codes on Google because the in-app translation wasn't standardized.
· "The queue isn't clearly visible." Nothing signaled urgency, so waiting patients were missed.
· "The drug search is very confusing." Every pharmacy's stock appeared at once, even when only two options fit.

05What I led

I framed the HMW with stakeholders, designed and ran the survey and interviews, built the experience journey map, and drove a six-phase design sprint end to end, from wireframes to a low-fidelity prototype tested with real internal doctors.

The hard part wasn't drawing screens. It was reconciling what doctors wanted- speed- with what patient safety required: complete and correct prescriptions- inside a tool a busy clinical operation depended on every minute of the day.

Six-phase sprint (stepper)
· Understand: HMW, UX review, interviews
· Define: Journey map, personas
· Sketch: Wireframe flows
· Decide: Assumptions, sprint questions
· Prototype: Low-fi build
· Validate: Test with doctors

06Key design decisions

· Freeze the patient context. Name, age, vitals, allergies, and history stay pinned in the header so nothing disappears when the chat scrolls.
· Make diagnosis clickable. Standardized, WHO-aligned ICD-X with mouse-driven auto-complete, plus a knowledge-base entry that converts straight into a diagnosis, retiring the Google tab.
· Filter drugs by real stock. Enter up to three ingredients and see only the nearby pharmacies that have all three. Stock-outs surface before the prescription, not after.
· Default the safe path. Allergy and contraindication fields default to "none," cutting four clicks from every prescription for the majority of patients who have neither.
· Make the queue impossible to miss. A visible queue count with a sound alert, plus read receipts and editable messages, so doctors know who's waiting and what's been seen.
· One screen, no pop-ups. Prescribing happens inline instead of behind an oversized modal that hid incoming patient chats.

Figure captions (for your screens):
· Diagnosis, clickable. A standardized ICD-X picker doctors can drive with the mouse.
· Stock before commitment. Three ingredients in; only the pharmacies that carry them show.
· Patient context, pinned. Vitals, history, and allergies stay in the header through the entire consult.

07Reflection

"Healthcare tools don't need flashy UI. They need to be a quiet, reliable space for decisions that affect patient safety."

The biggest wins weren't new features; they were removed friction and cleaner language. Uniforming the ICD-X wording and defaulting fields to "none" did more for speed than any visual polish, and collapsing four open windows into one workflow changed how a whole shift felt.

Treating doctors as co-designers, not just interview subjects, was what made it land. They knew exactly which twelve clicks to kill and why. If I ran it again, I'd instrument the workbench from day one, so every "faster" claim had a number behind it before rollout, not just after.