Earlier than beginning her firm, Brooke Hopkins labored on methods that helped self-driving automobiles function safely on public roads. At Waymo, her workforce ran thousands and thousands of simulated miles for each code change, since failures after launch weren’t an choice for autonomous autos. She noticed that comparable requirements would quickly be wanted for AI voice brokers, though few folks observed this development again then.
At this time, Hopkins’ startup Coval raised $28 million in a Collection A spherical led by Norwest, with Base10 Partners, Twilio Ventures, and Y Combinator additionally investing. This brings Coval’s whole funding to $31 million because it launched in 2024.
“Each firm goes to have a voice agent similar to they’ve a cell app or an internet app. However in the present day, most enterprises don’t have the infrastructure to deploy these methods with confidence,” Hopkins says.
More than $7 billion was invested in voice AI within the first quarter of 2026, and the market might develop to over $20 billion by 2031. Firms are rapidly including voice brokers to customer support, healthcare, and monetary companies.
Nonetheless, most depend on handbook testing, wherein engineers evaluation name transcripts, search for issues, and hope their fixes work. This methodology doesn’t scale properly.
Making use of Waymo’s testing strategies to telephone calls
At Waymo, Hopkins led the workforce that constructed analysis methods. They developed instruments to run simulations on distributed computer systems, which helped Waymo launch automobiles on public roads. After she left, Hopkins talked to AI agent groups and located they confronted comparable issues: too many potential inputs, no dependable technique to take a look at, and critical penalties when issues went improper.
Coval, based in 2024 via Y Combinator and primarily based in San Francisco, runs tens of thousands and thousands of simulated checks for voice brokers. The platform checks for points similar to accents, interruptions, background noise, and weird conditions earlier than actual customers are affected.
After launch, Coval continues to watch brokers and routinely sends failed calls again into testing. For instance, a monetary companies firm can simulate hundreds of callers who give conflicting data or grasp up unexpectedly earlier than the agent talks to actual prospects.
The startup says its platform can cut back handbook high quality checks by as much as 30x and pace up deployment by as much as 10x.
Coval’s major rivals are Hamming, which handles regulatory edge instances for healthcare and monetary companies, and Roark, a YC W25 graduate that has processed over 10 million minutes of calls and specialises in replaying failed conversations with up to date agent logic.
Whereas these corporations give attention to sure areas, Coval supplies a whole answer. It presents pre-deployment simulation, dwell monitoring, and human evaluation, all constructed particularly for voice audio moderately than tailored from common language mannequin instruments.
Zoom and Deepgram already use Coval
Coval works with Deepgram, Zoom, and greater than 60 different corporations. Endorsements from Deepgram and Zoom stand out as a result of each have intensive expertise with voice AI failures.
“Reliability and observability are a high precedence for us at Zoom as voice AI strikes into customer-facing manufacturing environments. Coval provides Zoom’s prospects the power to guage conversations systematically at scale, determine edge instances earlier than they affect customers, and transfer considerably sooner with confidence,” says Ram Rajagopalan, head of product for CX AI, Zoom.
Deepgram, which supplies voice AI infrastructure to different corporations, makes use of Coval to check its merchandise earlier than they launch.
“Voice brokers introduce a brand new stage of complexity in comparison with conventional software program testing. Brooke has constructed Coval right into a core a part of the fashionable enterprise analysis stack by enhancing reliability previous to scaled deployment. For any critical enterprise deployment, that is now not a nice-to-have,” provides Anoop Dawar, COO, Deepgram.
The buyers and what’s forward
Norwest, which has invested in corporations like Gong, Vuori, and Spiff, which is now a part of Salesforce, led the funding spherical. Scott Beechuk, a accomplice at Norwest, stated they invested due to Hopkins’ expertise at Waymo.
“Along with her deep expertise constructing analysis methods for autonomous applied sciences at Waymo, Brooke is uniquely positioned to guide Coval in defining how corporations deploy and scale voice brokers reliably. She helped show self-driving automobiles might work, and now she’s tackling voice AI,” notes Beechuk.
Twilio Ventures is concerned in additional than simply funding. Its enterprise prospects, who use its voice infrastructure, are a pure match for Coval’s analysis platform. Twilio needs to ensure these deployments are profitable.
“Belief is vital to scaling these experiences. Our funding in Coval displays our conviction that complete analysis and testing instruments, mixed with a powerful observability and reliability layer, are foundational to sustaining momentum in in the present day’s voice AI renaissance,” says Andy O’Dower, VP and subject CTO, Twilio
Coval plans to make use of the brand new funding to develop its gross sales and options engineering groups and enhance its product by including deeper simulations, new integrations, and higher human evaluation and monitoring. The corporate says its income has grown ten occasions 12 months over 12 months, nevertheless it has not shared its ARR or present headcount targets.
A giant query for the trade is whether or not voice AI analysis will keep impartial or change into a part of the platforms it helps. Twilio’s option to put money into Coval, moderately than construct its personal instrument, reveals that a minimum of one platform prefers to maintain analysis separate.
