About EliseAI
About 1 in 6 US apartments use EliseAI. If you’ve called to book a tour, report a broken dishwasher or ask about a renewal, there’s a fair chance an EliseAI voice agent picked up.
EliseAI has now run more than 1 billion AI conversations across housing and healthcare. In housing, they cover the renter journey from first inquiry to maintenance, renewals and resident support. In healthcare, they schedule patients and handle intake.
At their volume of high-stakes calls, the voice on the line often decides whether a caller gets what they need or gives up.
The challenge
Whether a caller is applying for an apartment or rescheduling an appointment, EliseAI’s agent has to sound right to both of them. Getting there meant solving a few problems at once:
- Consistency at scale: a voice that sounds great in a demo may fall apart over thousands of live calls, drifting off-tone at the worst moment.
- Choosing the right voice: most vendors hand over a catalog to test, with no long-term collaboration.
- Reaching every caller: a generic voice in a single language can’t serve people who don’t speak English at all.
The solution
EliseAI runs its voice agents on Cartesia’s Sonic models, and the two teams work closely on which voices ship and how.
When Akshay joined EliseAI, picking a voice meant close partnership with Cartesia, giving constant feedback on which voices and regional accents held up on real calls. EliseAI tried a Midwestern voice in Illinois and a Southern voice in Texas, and Cartesia helped tune voices region by region.
With multilingual voices, EliseAI has been able to deploy agents that can speak a caller’s language natively, especially important for affordable housing.
We run 5M+ voice calls a month helping people through eligibility, income verification, and recertification. Multilingual means we can now do it in their own language, after hours, when they need it, because the language you speak shouldn’t make housing harder to get.
Nothing at EliseAI ships on a hunch alone. Every new voice or model clears four gates at EliseAI:
- A first check against public benchmarks
- A blind A/B test where raters each score calls on naturalness and flow
- A 1% production rollout, measured on metrics like tour bookings, patient appointments and engagement
- A manual audit of sampled production calls before the wider rollout
We love working with Cartesia because they really listen and iterate based on what we need. The future of voice involves personalization, from tone to emotion to speed, with so much more we can do to help our customers build trust through delightful voice experiences.
The results
The regional voice test moved every northstar metric EliseAI tracked.
- Tour conversions rose 5%.
- Caller engagement rose 8%.
- Callers spoke 16% more words, which gives the agent more to work with and a better shot at solving the problem on the first call.
Sonic-3.6 has also pushed naturalness further. In early testing, renters rated it 12.5% higher on voice naturalness than every previous model.
What’s next
EliseAI keeps chipping away at the hardest calls, such as a TV blaring in the background or two people talking over each other. Together with Cartesia, the team treats these edge cases as problems worth investing in and solving properly.
Voice is the interface for AGI. Most of the world still hasn’t adopted AI, and distribution is the real bottleneck, not capability. Voice will be how we bring AI to the masses.