Victoria Panacci
Back to all work

Four in ten participants finished the daily questionnaire. Nobody knew that, because nobody had counted. I counted, and then I rebuilt the app.

Role
Product Designer and Product Manager. Sole designer on a team of 12 PMs; also ran participant training. DribbleCollect*
*DribbleCollect isn’t a real company. This work comes from a real project under NDA. The client name is the only thing that changed.
Platform
Native iOS (SwiftUI), site-issued iPhones
Result
Completion 40% → 80%. Error-free entries 60% → 95%. Support tickets down by half.
Before: counted by hand across live studies. After: event tracking in the new app, first on two studies with about 200 participants, then every client that migrated.
A clinician in a white coat with a stethoscope holds a phone in both hands.
The problem

Every training session I ran ended the same way.

Where do I press? Is 10 good or bad? Did that save? Where do I go now?

These were people over 60, using a tool a study told them to use, on a phone the site handed them. Every bit of friction was a missing data point.

Before and after of the entry summary. Before: the legacy Confirm Entry Summary screen, a plain two-column list of questions and answers with small Ok and Cancel links in the corner. After: the rebuilt Your evening diary review, marked Submitted 7:30 pm and editable until 7:30 pm tomorrow, with answers grouped under How you feel and symptom rings for pain, nausea and overall.
The entry summary, before and after. The legacy screen was pitched and sold to clients exactly like this. Nothing could be read from the database automatically, so every answer was typed in by hand, and the button that completed the entry was easy to lose on the summary screen.
The number nobody had
  • There was no analytics tooling. I pulled the numbers by hand, one live study at a time.
  • The app was a web wrapper on a legacy database engineers called “the scrolls.” One new question type: 15 engineering hours plus 2 of validation.
  • Leadership had heard “the app is bad” for years. They’d never heard “the app costs 17 hours per question.” A pitch built on taste would have died in the room. This one was built on hours.
Four baseline number cards: 4 in 10 participants finished a questionnaire; 6 in 10 of those finished without an error; about 15 hours of engineering per new question type; plus 2 hours of validation on top of each one.
Every one of these numbers was counted manually. That was the first finding.
The rules

Effortless daily completion · Respect the validated instrument · Visible compliance, no nagging · Trust over speed

The colours stayed because a palette change meant a brand review with every live sponsor and retraining people who had just learned the old look. The budget went to components, where the completion problem lived. I’d make the same call.

Sponsors kept asking for in-app messaging with coordinators. I asked the coordinators. Immediate no: one more inbox on a job that already had too many. The sponsor saw a feature. The coordinators saw their evening. I sided with the coordinators.
Four decisions

  1. The pain scale

    Chose
    A vertical list of numbers.
    Rejected
    The slider I walked in expecting to pick. Smallest target on the screen, most ways to overshoot it. Finished third with 20 testers, half over 60.
    Cost
    Two weeks of live data on the two finalists, which produced a tie. Leadership broke it on taste. Once the data said both worked, taste was the only thing left.
    Now
    I’d make the same call.
    Three versions of “How much pain do you have right now?” answered with 7, Quite a lot: a slider labelled finished third, a vertical list from Worst pain 10 to No pain 1 labelled shipped, and a grid of number buttons labelled tied.

  2. The diary window

    Chose
    One reminder, a visible open/closed window, a done state that says exactly when the entry counted.
    Rejected
    Escalating reminders and a hard lock at window close. Both sponsor asks. Both would have raised compliance on paper by annoying people into quitting.
    Cost
    A harder conversation with sponsors than a “yes.”
    Now
    Test reminder timing per study instead of one default schedule.
    Two screens. Open now, and what’s next: the home screen greeting Maria with an Evening diary card and Start button, Day 23 of 84 in the SUNRISE Study, and tomorrow’s morning diary opening at 7:00 am with a reminder on. Done, with a timestamp: the evening diary marked Submitted 7:30 pm and editable until 7:30 pm tomorrow.

  3. Two systems for a year

    Chose
    Run old and new in parallel so studies closing within the year didn’t retrain mid-study.
    Rejected
    A forced cutover that made the migration chart look clean.
    Cost
    Two codebases, two training tracks, validation on both, for about a year.
    Now
    Set the migration cut-off in the contract, so the choice happens at signing.

  4. Easy Mode

    Chose
    One switch that turns on high contrast and larger text together.
    Rejected
    A settings menu of contrast, text size, reduce motion. More flexible, and nobody over 60 was going to find it.
    Cost
    A second set of component states.
    Now
    Almost nobody found the switch alone. When a coordinator or family member turned it on, they kept it. I’d designed accessibility as a setting; it turned out to be a handoff. It belongs in onboarding.
    Easy Mode switched on: a question in Easy Mode with larger type, a “Pick one” hint, bigger option rows and a “Speak your answer” button, beside the Display and reading settings card with Easy mode on and Read questions aloud off.
What changed
  • Completion: 40% → 80%
  • Error-free entries: 60% → 95%
  • New question type: about 10 fewer engineering hours
  • The app went from the slide we skipped in demos to the one we led with
The prototype is the easy part to show. The part that mattered is how cheap it became to change. Tap Start to answer Maria’s evening diary; the settings gear turns on Easy Mode.
The training room

The training sessions changed too. The same hands stopped going up. People just answered the question, which was the only thing we had ever asked them to do.

More work