Leading discovery for a consumer AI phone agent to build trust and global reach
.png)
Switch was founded in 2015 as a call recording and transcription app, now with over a million users. The next big step was ambitious: what if the app could make calls for you?
I joined as the sole designer to lead product discovery, user research, and design for this new AI outbound calling experience. My challenge: earn user trust and make sure the product could work globally.
We began with a broad hypothesis: If people hate making and receiving calls, they’ll happily let AI do it for them.
Early discovery told a different story:
- Trust gap: ~70% of interviewees said they wouldn’t trust AI for sensitive calls.
- Perceived value: People were only open to trying AI if it was free, solved low-risk tasks, or solved an important enough problem for them to pay for it.
- Global barriers: Early interviews highlighted that without international numbers, the product would not be usable or desirable for many groups.
So the real challenge became: How do we design an AI caller people actually trust, and identify where it creates enough value to drive adoption?
I led the design of a transparent and global-first AI calling experience, built on two pillars:
- Trust by design: live transcripts, mid-call chat corrections, and clear post-call summaries so users knew exactly what the AI was doing.
- Global accessibility: new flows to let users add a Twilio numbers covering 180+ countries OR connect their existing number via call forwarding for seamless setup.
These decisions addressed the two biggest adoption blockers: trust and reach.
Product discovery & niching down
Early GTM smoke tests showed weak traction with a broad “AI phone calls to manage tasks” pitch. I led an internal strategy session where I mapped out 10–15 possible niches. Given founder-market fit and strong signals around travel and translation, we decided to explore the digital nomad, travel, and immigrant space further. In ~12 interviews with people from these groups, two adoption barriers came up again and again:
- Lack of trust in AI making calls on their behalf.
- The need for strategic and intentional features for global adoption.
Designing for trust
- Clear AI signals: On mobile, calls opened directly into a live transcript view. On the new web app, I created a split screen: chat on the left, transcript sliding in on the right once a call started. That flow felt natural and consistent across devices.
- Exploring natural AI voices: Our team tested multiple TTS providers (ElevenLabs, Orpheus, Sesame). We ended up using Sesame, as it sounded natural enough that people actually believed the AI could hold a real conversation. We gave options to either pick from a list of voices or clone their own; for cloning, users were consistently delighted at the AI's accuracy in copying intonation, tone, and accent.
Designing for a global audience
- Global numbers with Twilio + eKYC: users could quickly add a new number compliant with regulations in 180+ countries.
- Keep your own number: users could connect their existing number via call forwarding, lowering friction for onboarding.
Prototyping fast with Cursor
I used a mix of Cursor + Shadcn component library + Figma MCP for rapid prototyping. In close collaboration with web and mobile engineers, I helped ship the frontend for several features.
Below is an example of the full workflow for a feature called 'Web Search.' I first mocked up the UX/UI and user flow in Figma, then shared the file with Cursor. Lastly, I connected it to the Figma MCP plugin and continued to chat with Cursor until the code aligned with the actual UI as closely as possible.
The beta experience delivered a transparent, global-ready AI caller: users could set up numbers in 180+ countries or keep their own number via call forwarding, while transcripts and summaries made the AI’s actions visible.
- Led GTM and received 150+ signups for private beta within a few days of launch.
- Validated that translation and insurance claims were high-potential drivers for future adoption.
- Shaped the team’s strategy: adoption would depend on trust + global reach, not just a realistic AI voice.
- Trust is crucial: Learned that trust isn’t earned by accuracy alone; it’s earned by visibility of what the AI is doing (e.g. showing the AI's thought process via chat)
- Global-first matters: Without international support, we’d lose core user groups.
- Continuous discovery:
- Feedback highlighted niche groups (immigrants, digital nomads, people handling complex admin) as early adopters. This informs us to continue refining product positioning and messaging.
- Early testing showed higher willingness to delegate low-stakes calls (e.g. restaurant booking, service appointments) over higher risk ones (e.g. booking a plane ticket).
- People aren't interested in fancy AI features if it doesn't solve a problem.
- Design is strategy: As the only designer, I wasn’t just designing UI — I helped drive product direction through research, GTM validation, and adoption insights.
