Avatar and notification research at the Fred Hutch Cancer Center
Helping Users Feel Invested in Their Quitting Tobacco Journey
The 30-second version
Fred Hutch needed to know whether young adults would trust a virtual quitting guide named Alex, so I designed and ran think-aloud interviews to test that trust and see whether notifications would motivate people or just get muted. What we found reshaped the design brief: people want avatars that look like themselves, not a mascot, and a robotic voice undercuts everything else the app is trying to do.
Overview
Fred Hutch builds digital interventions for cancer prevention and health behavior change in underserved communities, including LiFT. The LiFT app pairs young adults with Alex, a virtual guide who checks in and sends encouragement throughout a quit attempt. The premise is simple: quitting tobacco is hard to do alone, and a guide that feels genuinely present, rather than a generic reminder app, should keep people engaged longer, which in turn should improve quit rates.
LiFT had already been tested with LGBTQ+ populations and was preparing for a larger clinical trial with young adults who use tobacco or nicotine. That trial was designed to evaluate the program's content and efficacy, not the surrounding UX. Two things fell outside its scope entirely:
Alex's design had been shaped by earlier generative research, but that research had never been verified, and now that the app was expanding to a broader population of young adults, it had never been tested at all. Nobody had confirmed that young adults, specifically this wider group, would actually trust or connect with the avatar as built.
Notifications had never had generative research at all. The team didn't have a clear picture of what timing, tone, or content this population actually wanted, and the clinical trial wasn't going to surface that either.
If we didn't answer these questions before the trial launched, nobody would. This research existed specifically to fill the gap that the trial itself wouldn't touch.
Would people connect with Alex?
How could we craft notifications to motivate them instead of becoming background noise?
My role
I led early research scoping, pushing the team toward the two areas with the clearest path to actionable findings before the trial launched: avatar customization and notifications. I proposed and scrapped early ideas for diary studies and codesign sessions, since long-term engagement data and notification-content testing could be captured later through other ongoing studies. Interviews were the fastest way to surface findings that could still shape the trial before it started.
From there I:
Drafted the interview guide
Created the think-aloud stimulus materials
Co-facilitated six interviews
Documented observations
Synthesized themes into product recommendations
What helped the most was creating follow-up stimulus slides after the customization task. They gave participants something concrete to react to, helping us move beyond first impressions into deeper conversations about identity, trust, and motivation. With everything laid out side by side, participants were able to give me critical, specific feedback.
I also drafted rough research plans and slides to help the team ideate on what data would be most useful and most impactful in the short term to inform the clinical trial.
I drafted rough research study plans and slides to help the team ideate on what data would be the most useful and most impactful in the short term to help inform the clinical trials.
Research focus
Main Research Question: How does the design and customization of a virtual guide, alongside notification messaging, influence young adults' motivation, engagement, and trust in the LiFT app?
Sub-questions:
How does personalizing a virtual guide influence motivation and engagement?
How do users perceive and navigate the avatar customization experience?
What emotional responses does the guide evoke?
What notification timing, content, and delivery methods do users prefer?
How much control do users want over notifications and personalization features?
Method
Why Interviews
I pushed the team toward interviews over diary studies or codesign. Long-term engagement data could be captured in other ongoing research, so interviews were the fastest way to get findings that could still shape the trial before it launched.
I created stimulus slides for after the think-aloud customization to elicit more thinking about the customization process and get more targeted feedback.
Structure
Each session followed four parts:
Their quitting journey
Customize Alex while thinking aloud
Discuss stimulus concepts
Explore notification preferences
Starting with their own quitting story, before asking them to build anything, set the tone for the rest of the session. I created stimulus slides for after the think-aloud customization task, specifically to elicit deeper reactions and get more targeted feedback on the customization experience.
Getting Honest Answers About a Hard Topic
Six people talking about their own quit attempts with two strangers on a video call is not a low-stakes conversation. Here's how I helped participants feel safe enough to be honest about it:
I made clear upfront that we weren't part of the team that built the app; we'd been brought in specifically to get honest opinions on it. Their care team wouldn't see or be affected by anything they said, and since we hadn't designed the app ourselves, critical feedback wouldn't hurt our feelings either.
I kept the interview language judgment-free throughout, especially around relapse and failed quit attempts.
I structured sessions so participants led with their own quitting story before I asked them to build or react to anything.
That ordering mattered. Once people had already talked about a hard, personal thing without being judged, they trusted me enough to say what actually felt fake or preachy about Alex.
Why Think-Aloud
People are surprisingly bad at remembering why they clicked something five minutes ago. Having participants customize Alex while thinking out loud let us catch those tiny moments of hesitation and delight in the moment they happened. Then, using the stimulus slides I created, we dug even deeper into what made Alex feel supportive or not.
Findings
People weren't customizing a mascot. They were building themselves. Several participants described the finished avatar as talking to itself. Customization wasn't decoration; it was a mirror that made the quitting journey feel personally owned.
Inclusive, but not deep enough. The customization options had something for everybody on paper, but people still asked for darker skin tones, skin texture like freckles, natural hair textures, braids and afros, and more feminine clothing.
Voice is a trust decision, not a style choice. A voice that sounded robotic or unenthused quietly undercut all other content in the app.
Notifications should show up when they're needed most. Control over timing and frequency mattered more than the message itself. Participants wanted notifications to hit at the times they were most likely to smoke.
A login issue surfaced that had nothing to do with our research questions, but everything to do with the trial's success. While observing participants navigate the app during sessions, I caught a technical issue in the login flow that would have blocked or confused users trying to get into the app on their own. Left unfixed, this could have caused major onboarding failures once the clinical trial launched, undermining recruitment and retention before the study even began.
Final research report slides showing illustrations to help give faces to participants and allow for better tracking of data while reading the report, with severity ratings of the suggested changes based on pain points
Impact
Our findings shaped the trial before it launched:
Expanded avatar customization, based on the representation gaps we found
Reconsidered voice implementation, since a robotic tone was undercutting trust
Shifted notifications to user-controlled scheduling, instead of a fixed cadence
We also caught a login issue that would have gone unnoticed any other way. A participant hesitated at the login screen for reasons that had nothing to do with our research questions. We only caught it because think-aloud put people inside the real app, not a survey or a description of it. Had this surfaced mid-trial instead, it would have hit during unsupervised onboarding, with no researcher present to explain the drop-off.
We flagged it early enough to account for before the trial launched.
Reflection
The biggest lessons:
Impactful research happens before sessions even start. Understanding prior findings, stakeholder goals, and technical constraints mattered as much as the interviews themselves. Scoping was iterative; every idea got stress-tested against whether it was actionable, and the study was stronger for cutting the ones that weren't.
Nobody complained that the app wasn't inclusive. They just quietly noticed what was missing. That's a much harder thing to catch, and a much more useful thing to know.
Presenting the final research report to the Fred Hutch team