Watch
Executive summary
AHS replaced that with Krisp AI Voice Translation, building multilingual support into the existing call flow. And it did so as demand surged: regulatory changes drove roughly 90,000 additional calls from February to August 2026, about 15,000 a month. Every metric AHS tracks improved anyway.
Key outcomes in pilot program:
- 31% reduction in Average Handle Time, the driver behind the gains below
- 32-second improvement in Average Speed of Answer, and abandoned calls down from 5% to 3.08%
- CSAT up from 81% to 86%
- 96% translation accuracy scored on 100% of calls, with zero patient safety or experience incidents
- 9 in 10 multilingual calls completed in-flow, and a 500-agent rollout compressed from over a month to a single day
Together, AHS and Krisp showed that multilingual support can move from a slow dependency into an in-call capability, so agents are judged on how well they help, not on which language the caller happens to speak.
The interpreter tax
For AHS, a call is rarely about a product. It is about whether someone can access healthcare or benefits. And when a caller does not speak English as a first language, the first 30 seconds that decide the interaction are the hardest to get right.
Before Krisp, every non-English call followed a manual, multi-step process. The agent identified the language, placed the caller on hold, and started a three-way call with a third-party interpreter service before resuming the conversation through that interpreter for the rest of the call.
That process cost time and added complexity to every non-English interaction:
- Time. Hold time to connect, plus waiting for an available interpreter, extended handle times. On application processing, where a script must be read slowly and consent captured correctly, a five-minute call could stretch toward 45, with a pause every time the agent, interpreter and caller took turns.
- Added steps. Identifying the language, holding the caller, and coordinating a three-way call added handling complexity to every call.
It also broke the conversation at the moment trust forms, as Martin Cronkhite, Corporate Director of Omnichannel and Telecommunications at AHS, describes:
“It sounds like an immediate halt to the conversation. I’m no longer focused on what you have to say, I’m focused on my next step in the process, and I’ve removed the connection we were going to establish.”
AHS primarily supports callers in Spanish, Portuguese, Arabic and Haitian Creole, with Spanish, Portuguese and Arabic making up most non-English volume. Every one of those calls carried the interpreter tax.
Why AHS chose Krisp
AHS’s goal was clear: reduce Average Handle Time by removing the delay of connecting to third-party interpreters, while improving the customer experience. After evaluating alternative solutions, AHS chose Krisp for four reasons:
- Native integration and real-time performance suited to a live contact center environment
- Efficiency and scalability across the state program portfolio
- Privacy-first architecture with the option to run without storing any conversation content, alongside opt-in recording and transcript storage where the program wants it, useful for sensitive healthcare and benefits interactions
- Ease of deployment, with translation available to an agent at the click of a button
From a 20-agent pilot to a same-day rollout
AHS scoped the initial proof of concept deliberately small, roughly 10 to 20 agents in a single center, to validate performance before any wide impact. The pilot answered the question quickly: handle time fell in the areas that mattered most, and the site chose to extend Krisp to every agent. A rollout planned to take more than a month across 500 agents was completed the same day as release, with deployment as simple as enabling single sign-on.
“We introduced this to 20 agents, and they saw such a dramatic improvement that they decided to go with a big bang scenario and give it to every single agent.”
Accuracy and safety, on every call
With Krisp, translation begins when the call begins. There is no hold, no queue, and no interpreter wait. Multilingual support runs inside the existing call flow, and complex or high-risk scenarios can still be routed to a third-party interpreter service within a hybrid model.
Quality was measured on every translated call, not sampled:
96%
overall translation accuracy across meaning preservation, critical information transfer and natural conversation flow
99%
conversation flow
97%
entity accuracy
95%
intent accuracy
95%
native fluency
- In post-call agent surveys, agents rated 57% of calls fully accurate on critical information and 33% good with no impact on safety or outcomes, with 10% routed to a third-party interpreter service as part of the hybrid model
Most importantly for a healthcare provider, there were zero reported patient safety or experience incidents across high-stakes calls involving medical terminology, patient identifiers, birth dates, prescription names, addresses and insurance carrier names.
One change, a chain of results
Reducing handle time did not just speed up calls. It set off a chain that lifted every metric AHS tracks, all while the operation absorbed a sharp rise in volume.
From language skills to empathy
The clearest change agents describe is what they no longer have to do. Instead of acting as a router between the caller and a translator, they can stay with the caller.
“You can actually feel what the caller is calling about, and focus on that, and have that empathy for them. It becomes more of a conversation rather than a transaction.”
“Krisp makes non-English calls simple and fast.”
That change also reshaped how AHS thinks about hiring. When translation is built into the call, the deciding factor for a role is no longer which languages a candidate speaks, but whether they have the empathy and commitment to the mission the work demands.
“Our focus transitions to what is most important: hiring someone that can do the job, that has the empathy, and follows our mission. It’s not so much about the additional language you can speak, but more about what it’s going to drive for the customer.”
Treated as infrastructure rather than a staffing constraint, clear multilingual conversation is what lets a provider like AHS offer continuity of care at the point of service, in any supported language, without penalizing a caller for not speaking English as a first language.
What’s next
AHS intends to implement Krisp Voice Translation across all of its projects, and sees opportunities to explore additional Krisp capabilities on the same platform, including Noise Cancellation, Agent Assist and Speech Analytics.
