Hiya Health
Reimagining patient-provider communication in 2030.
Overview
Client: Hiya (Completed in partnership with Hiya through a graduate-level capstone project.)
Timeline: January 2025 - August 2025
Team: 2 Designers,1 Researcher (me)
My Role: As lead researcher, I designed research plans and conducted interviews with participants, recruited participants, led synthesis, and tested design concepts with participants. I was also involved in the design of AI voice prototypes.
"By 2030, phone calls between humans and AI will be more frequent,
how can we increase trust in these AI systems?" - Hiya
Hiya, who leads in voice security, protects millions through advanced spam detection, call authentication, and data-driven insights that make phone communication safer and more trustworthy.
By 2030, AI will be involved in how patients receive care from small outpatient healthcare clinics, and trust in these AI systems is crucial. Hiya challenged our team to create a vision for 2030, where humans interact with AI systems, and trust is at the forefront.
Solution
Hiya Health is an AI-powered receptionist that answers phone calls, triages requests, and also supports frontline staff members by surfacing relevant context in order to strengthen patient-staff communication.
.png)
Desk Research
Hiya asked us to focus on Small and Medium Sized Businesses (SMBs), so I conducted desk research, including a SWOT analysis to understand how to move forward within the SMB space. What I found was:
​
85.7% of SMBs operate in service-driven industries, where trust, responsiveness, and efficient communication are critical for success. Service SMBs need AI-powered automation to streamline outreach, engagement, and operations, improving response speed by 5X and conversion rates by 20-30% (Carta & McKinsey, 2024).
​
I then spoke with two product managers at Hiya, who both shared that healthcare is an industry that can greatly benefit from AI systems.
"The [healthcare] receptionist was picking up the phone and the script was literally the same...I feel AI agents are better for these scenarios." -Product Manager 1
Based on this desk research, our team decided to target small outpatient healthcare practices, as they met the SMB requirement, but also presented unique challenges around AI integration and sensitive patient information that we were eager to explore and address.
Generative Research
Method: 60 minute semi-structured interviews with two groups:
-
Small Outpatient Healthcare Providers (N = 5)
-
Considers themselves an “early adopter” of technology.
-
Owns or manages a small outpatient healthcare practice (less than 30 employees) and make decisions regarding communication.
-
-
Small Outpatient Healthcare Patients (N = 5)
-
Reports having communicated with a small healthcare practice in the last 6 months.
-
Reports often needing to communicate with small healthcare practices.
-
​​
I chose to speak to both providers and patients to help uncover practical needs and emotional drivers that shape the overall communication experience.
Through semi-structured interviews, I wanted to explore:
-
How small outpatient practices manage patient communication, including key tasks, tools, and pain points
​
-
How patients experience communication with small outpatient healthcare practices
​​
-
What makes AI-driven communication feel helpful, trustworthy, or worth engaging with for both groups
Analysis
Following the interviews, I led meetings with my team where we created affinity maps, analyzing patterns, pain points, and emerging themes across provider and patient data sets to uncover needs and distinct challenges for each group.
Where Patients and Providers Overlap:
Both providers and patients see value in AI for administrative efficiency
​
Providers welcome AI for scheduling, inquiries, and professional communication.
“If AI could analyze the schedule and help rearrange it in ways ...more profitable and efficient." - Provider 1
​
​
Patients embrace AI for faster scheduling and simple health questions.
​
"If I can schedule appointments quicker because of AI, that's just way more beneficial for me." - Patient 5
However...
Both groups agree: empathy requires human interaction.​
​
Providers emphasize the irreplaceable role of human judgment and empathy in patient communication.
​"[For rescheduling] As a human, I can hear what's happening with [patients] and maybe create a solution that an AI wouldn't." - Provider 4​
​
Patients want real human connection in sensitive moments.
​
“The AI would be like, ‘I'm really sorry that this has been frustrating.’ And I'd be like, ‘you don't know, you're an AI, you're not real’” - Patient 2
For patients, breakdowns in communication erode trust.
Patients struggle to receive consistent, empathetic communication from their providers, which causes them to lose confidence and increases the likelihood of switching providers. Patients expect timely, compassionate communication.
“She [the receptionist] was not very nice. I almost called back and canceled [the appointment] just because of the communication with her." - Patient 1
"I'm moving on [because the provider never responded]. We've never talked and you're already ignoring me, why would I see you as a doctor?" - Patient 2
“What makes the trust is...the detailed explanations when customers have questions. It's the professionalism in the way they speak...” - Patient 3
For providers, a critical driver of business success for small outpatient providers is the volume of successful appointments. But providers are currently facing operational challenges.
What’s Driving These Challenges:
-
Limited time and resources to manage scheduling efficiently
-
Missing or inconsistent pre-visit checks during the booking process
-
Gaps in follow-up workflows that rely heavily on staff coordination
-
Fragmented communication across multiple, unintegrated platforms
“There’s too many moving pieces… it can be very confusing.” - Provider 2
“It’s [communication] been kind of a challenge because I can’t be everywhere.” - Provider 1
Tension Matrix
To better understand where communication breaks down between patients and providers, I created a tension matrix comparing patient expectations, provider behaviors, and the resulting friction.
Key Themes Uncovered:
-
Inconsistent communication erodes trust and continuity
-
Scheduling barriers delay care and increase drop-off
-
Disconnected tools disrupt care and strain staff
Using the tension matrix, we created a list of problem areas to present to Hiya.
Presenting to Hiya and Workshop Facilitation
I facilitated our research presentation to Hiya stakeholders that included our research findings and the tension matrix. Following the presentation, we held a prioritization workshop with the list of problem areas and asked each person in the room to rank each problem area based on reach and impact. We then prioritized the top 2 problem areas to focus on for the design phase.
Our team then facilitated a second workshop with Hiya, where we asked Hiya stakeholders to brainstorm ideal and avoidable outcomes, create guiding values, and discuss where they believe Hiya was uniquely positioned to lead.
Because our project space was more visionary and focused on five years into the future, this was our opportunity to align and think openly about future possibilities before our group entered the design phase.
We facilitated this workshop through FigJam, and walked through a series of brainstorming exercises.
Impact
​
My research generated evidence based insights on AI acceptance that were shared with a key enterprise stakeholder, influencing how stakeholders at all levels think about AI driven experiences.
Evaluative Research
While the designers on my team created prototypes of the staff-facing management system for Hiya Health, I worked on creating prompts for an AI agent using the Vapi platform that would allow me to create the experience of a patient speaking to an AI operated system over the phone.
​
I then began concept tests with both groups.
Provider Concept Test
Participants: (N = 4)
-
Owns or manages a small outpatient healthcare practice (less than 30 employees) and make decisions regarding communication, particularly those who manage or participate in patient communication workflows.
Method: 60 minute concept walkthrough of two versions of the interface for managing calls and supporting staff during live conversations. Staff walked through the same patient scenario as if they were on a live phone call. Half of the participants saw Version A first, followed by Version B, while the other half saw Version B first, followed by Version A.
​
The goal was to evaluate the effectiveness of Hiya Health in reducing provider workload, clarifying next steps, and building trust in AI-generated information to support calls with patients.
.png)
Version A
.png)
Version B
Findings:
-
Experienced staff preferred AI as a supportive tool (tips, checkpoints) over rigid scripts, however scripts were seen as helpful for less experienced staff.
-
Staff valued AI support for surfacing and completing action items and prioritizing patients by urgency, but wanted greater flexibility in how outputs and dashboards could be customized.
-
There were barriers to trusting AI generated sentiment scores that reflected the patient’s emotions over the phone.
​
Recommendations:
-
Refine AI guidance by focusing more on displaying tips and checkpoints as opposed to AI generated scripts.
-
Improve AI outputs by creating editable transcripts, export options, and control to override AI-scheduled appointments when necessary.
-
To address trust barriers with AI sentiment scores, consider being more transparent about how the sentiment score is being calculated.
Patient Concept Test
Participants: (N = 4)
-
Patients of small outpatient healthcare practices who have recently communicated with a small outpatient healthcare provider.
Method: 30 minute scenario-based evaluation, where participants were asked to role-play a patient making an urgent phone call, first interacting with the AI assistant and then being routed to a human staff member (played by a member of our team).
​
The goal was to evaluate if the AI’s tone, empathy, and sentiment handling build trust and help patients navigate their needs.
This is the AI agent that I created, and an example of the patient scenario that I tested.
Findings:
-
Patients appreciated when AI was upfront about its identity, provided the option to connect to a human at any time, and had a human sounding voice and reassuring tone that contributed to a positive experience.
-
Patients valued not having to repeat themselves throughout the entire process, which significantly increased trust. Personalization also increased trust, and liked that AI knew their name, though some requested a way to confirm their identity with the AI.
-
Some patients described discussing their emotions with an AI as awkward, while others thought it was a thoughtful addition.
​
“That was the most amazing AI robot, whatever you want to call it, I've ever spoken with in my life. I felt like I was talking to a human.” - Patient 3
​​​​
Recommendations:
-
Incorporate a security check where AI confirms patient identity (likely by asking for patient date of birth) before accessing patient data.
-
Consider making emotional check ins with AI optional, allowing patients to skip if they find the emotional support unhelpful.
-
Future Consideration: (Not a recommendation as it was not mentioned during patient testing) Given the early stage of the prototype, refining the AI’s talking speed could help conversations feel more natural in future iterations.
Reflection
What I learned:
I learned so much as I led the group through end-to-end research. I strengthened my leadership skills and learned to be comfortable making decisions around reseacrh planning, participant selection, and how to move forward when things felt uncomfortable or didn’t go as expected. I also learned so much about visionary work and how impactful it can be.
​
What I would do differently :
Pilot our brainstorming workshop beforehand. We had to make a few decisions on the spot during the session that I believe might’ve been addressed if we had tested it ahead of time.
If I had more time, I would dedicate more effort to refining the prompting and iteration of the AI voice prototype. I initially provided a strict script for the agent to follow to maintain consistency and control, but with additional time, I would experiment with giving the agent more flexibility to respond dynamically while still aligning with our intended tone and goals.
Future Research: I would also want to conduct additional evaluative studies to further test and validate insights beyond the initial four participants.