From Survey to Interviews:
Uncovering Care Platform Opportunities
Background
PMP, Health2Sync's clinic-facing platform, helps physicians, nurses, and dietitians track which patients are due for a follow-up, quickly check patients' recent physiological data, and ultimately improve overall care quality. I initiated this research and, with my manager's support, began planning it, aiming to understand where the existing experience could better match how healthcare providers actually work, and to uncover opportunities for new features worth building.
Given our team's limited size and engineering capacity, I took a cross-role research approach: first, a screening survey to understand different roles, clinic sizes, and usage patterns (targeted 20 responses, ultimately included 40 valid responses), then selected a diverse and deliberately skewed-toward-low-satisfaction sample for 12 in-depth field interviews, aiming to surface high-value opportunities that would benefit physicians, nurses, and dietitians alike.
I led the research design, interview execution, insight synthesis, and prioritization, working alongside one PM, and independently pulled and analyzed the GA4 data used to validate outcomes. From the interviews, I distilled 42 opportunity areas, categorized them, and ranked them by priority to guide the following product iterations, and drove a first round of low-cost design changes in the short term.
Problem exploration
Before spending our limited interview time, we ran a survey first (target was 20 responses, we got 41, and after removing 1 admin-role response, 40 were used for analysis). The survey covered usage frequency, patient load, and satisfaction across three core tasks: finding priority patients, viewing data charts, and messaging patients.
From the 40 valid responses, 3 signals stood out. Together they answered three questions: where to dig deeper, what proves that direction is worth digging into, and how to pick the right interview candidates to cover the full picture.
Signal 1：Of the 3 core moments, "finding priority patients" had the weakest satisfaction
For all 3 value dimensions, none of the 41 responses said "disagree," but some still landed on "neutral" or below "strongly agree."
For "quickly judging patient priority," 5 responses (12.2%, mostly physicians) rated it only "neutral," clearly higher than "data charts help track patient status" (7.3%) and "patient messaging improves care quality" (0%). Messaging, the area we expected the most complaints about, actually scored the highest satisfaction of the three. The area that's weakest, and easiest to overlook, is finding patients.
Signal 2: Different roles use very different ways to find patients, with no shared pattern
Every role has settled into its own fixed pattern, but no two roles are alike. Physicians mostly use "recent activity" and "search by name." Nurses mostly use "search by name" and "group filter." Dietitians mostly use "recent activity" and "group filter." Health educators mostly use "recent activity" and "group filter." This shows the platform has no single, clear path for finding patients that works across roles, which also explains why this dimension scored the lowest.
Signal 3: Different roles carry very different patient loads and work in very different ways
All four roles use the same PMP, but their day-to-day work looks completely different. Dietitians manage 27.9 patients a day on average, three times more than physicians (9.1). Each role also prioritizes patients differently: physicians are "responsible for all patients," nurses focus on "pre-assigned patients," while dietitians and health educators focus on "patients with poor control or unstable conditions." The data each role checks most often is different too: nurses and dietitians mostly check blood glucose, CGM, and diet; physicians check CGM, blood glucose, and blood pressure; health educators only check blood glucose and CGM. This is why the "finding priority patients" problem can't be understood from just one role's point of view.
All four roles use the same PMP, but their day-to-day work looks completely different. Dietitians manage 27.9 patients a day on average, three times more than physicians (9.1). Each role also prioritizes patients differently: physicians are "responsible for all patients," nurses focus on "pre-assigned patients," while dietitians and health educators focus on "patients with poor control or unstable conditions." The data each role checks most often is different too: nurses and dietitians mostly check blood glucose, CGM, and diet; physicians check CGM, blood glucose, and blood pressure; health educators only check blood glucose and CGM. This is why the "finding priority patients" problem can't be understood from just one role's point of view.
These 3 signals shaped how I picked interview candidates
Based on these signals, I set three conditions for choosing interviewees: prioritize HCPs whose satisfaction across the three value dimensions wasn't at the top (answered "neutral" or "agree" rather than "strongly agree"), make sure roles were well mixed, and cover high, medium, and low patient loads, while trying to capture diverse usage patterns across the three core moments. During the interviews, I found some clinic types had limited insight to offer, so I added more interviews as I went. In the end, the plan grew from an initial 8-9 candidates to 12 completed field interviews.
Design Strategy
From interview notes to a list of opportunities
I recorded all 12 interviews. For each finding, I wrote it on a sticky note, grouped the notes by theme on a wall, and checked the groupings with my team. This turned scattered notes from 12 transcripts into one clear list of 42 opportunities.
How I ranked them: mentions, number of clinics, and difficulty
42 opportunities is too many to fix all at once, so I scored each one by three things: how many times it was mentioned, how many clinics mentioned it, and how hard it is to build (easy, medium low, medium high, or hard). This naturally split them into three groups.
① Quick Wins (add to this round's backlog)
These are mentioned often, affect many clinics, and are easy to build, so they give the best return on effort. Priority for this round:
-
Multi-day line chart is too cluttered. Mentioned 4 times, in 4 clinics, the highest of the whole list. Pre-selected days will drop from 14 to 7, and "select all" will be removed.
-
Inconsistent icons for message replies. Mentioned 3 times, in 2 clinics. Fixed by updating icon and button styles to solve the "which one do I click" confusion.
-
"AI Analysis" is hard to find and use. Mentioned 3 times, in 2 clinics. The star icon will become an "AI" icon and move into the side menu, and related image recognition issues will be fixed too.
-
Color labels are hard to tell apart, and patient ID and edit actions are not intuitive. Each mentioned 2 times. Both are small fixes with a big payoff.
-
Filter icon meaning is unclear, and there is no quick "Quote reply" shortcut when viewing photos. Each mentioned once. Low mention count, but very easy to fix, so both are included too.
② Items that need more planning
These are mentioned often and affect many clinics, but they are hard to build, so they need full design and engineering work:
Feature discoverability is low (D 1). This is the single most mentioned item (6 times) and affects the most clinics (4) in the whole list. However, it needs a full onboarding or tour system, like a "key feature tour," which is too big for one sprint. This is a case where something is worth doing but should not be rushed. I recommend testing the direction first with a low cost alternative, instead of building the full system right away. This is also where the later "button labeling" design solution came from.
Appendix: Prioritized HMW opportunities from the interview findings
Out of the 42 opportunities, I'm walking through the four below in detail. The rest are mostly small tweaks, like swapping an icon, where there isn't much design thinking to unpack beyond a visual change. These four involved more substantial problem solving, so they're a better showcase of how I actually work through a design decision.
➊ Button display/labeling enhancements.
This addresses B 1 (faster contact and entry point recognition) and B 3 (unclear Filter icon). The messaging and AI icons will change from icon only to icon plus text label. Because of this change, the floating buttons on the patient page (AI Analysis, Notes, Chat) will also switch to a labeled style. The Filter icon on the messages page will be updated at the same time, making the filter function easier to understand.
➋ Daily Overlay Graph Enhancement
This addresses C 5 (multi day line chart too cluttered). Several HCPs reported that the 14 day overlay chart has too many colors and overlapping lines, making it hard to read. The team originally considered removing the "select all" feature, but was concerned this could make the interface look empty and even harder to discover. Instead, we chose a more conservative approach: changing the default number of days from 14 to 7, and observing the effect as a trial first.
To further support this, we added a short tutorial that guides users to move their cursor over a line to see it more clearly. This matches how users naturally scan the chart: when they want a closer look, they tend to focus on one line at a time rather than trying to read all of them at once. The highlighted line stays sharp while the rest fade into the background, and the tutorial makes sure this interaction is discovered rather than left for users to stumble on.
During this optimization process, I also identified an alternative way to present the same data. Instead of overlaying all 14 days on one chart, each day could be shown as its own small chart arranged in a grid, a format I proposed internally as "Quick CGM Review." This would let HCPs compare daily glucose patterns side by side rather than untangling overlapping lines, especially since insulin dose markers were also layered on top of the existing chart. I presented this concept to the team, and it was well received as a direction worth developing further. For now, though, we agreed to first observe how the current changes perform (the reduced default range and the hover based highlight) before committing engineering effort to build it out.
➌ Quote Reply Shortcut in Enlarged Photos
When a diet log entry only has a photo, users instinctively tap it to enlarge, but from there they have no way to start a Quote reply. This addresses C-9, where a dietitian noted that patients' photo only entries left her stuck in the enlarged view with no clear path to responding. Quote reply already existed elsewhere in the product, but wasn't available from the enlarged photo view, a gap this fix closes by adding the shortcut directly inside that view.
➍ CGM Chart Screenshot & Quick Send
This addresses C 6 (CGM screenshots and patient communication). HCPs previously had to take a screenshot separately and switch windows just to share a patient's glucose trend, breaking their workflow in the middle of a review. With this addition, they can capture the chart directly from the CGM page in a single tap and attach it straight into a message, without ever leaving the screen. To move this forward quickly, I first put together a design mockup so the idea wasn't just a concept on paper, then brought it to the team for discussion early. Once it's aligned, it's ready to be slotted into a Sprint at any time.
Early Signals
After the button labeling update went live, the AI Analysis feature saw a clear increase in usage, with events up 50% and active users up 17%. Notes, on the other hand, declined, with events down 22% and active users down 10%. This is likely related to the fact that many clinics already rely on paper records, combined with the removal of accidental clicks that came from an unclear icon before the redesign.
It's also worth being honest here that event tracking across the platform is incomplete. Adding or filling in tracking requires competing for engineering resources, and it hasn't been a high priority internally. As a result, several features, including the hover guide on the overlay chart and the one tap chart screenshot and forward, still can't be validated directly with data. We're relying on qualitative feedback the BD team gathered during clinic visits instead. Several HCPs noted that the updated overlay chart is somewhat better than before, but compared to being able to view each day's curve laid out individually, they'd rather see the one tap chart capture and forward feature launched first.
What's Next
・Several completed designs (color label editing, clickable target range, quote reply from enlarged photos) are already in the backlog and will move into upcoming Sprints soon.
・Feature discoverability (D-1) was the most frequently mentioned issue across all findings and affected the most clinics, but it requires a full onboarding mechanism and carries a higher build cost, so it can't be tackled in one pass. The next step is figuring out how to phase the work, validating the direction first before investing in the full design.