In high-volume healthcare contact facilities, each affected person dialog carries each scientific and operational significance, making correct real-time transcription needed for automated workflows. Correct, prompt transcription allows clever automation with out sacrificing readability or care, in order that groups can automate digital medical report (EMR) report matching, streamline workflows, and remove guide information entry. By eradicating routine course of steps, employees can keep absolutely targeted on affected person conversations, enhancing each the expertise and the result. As healthcare programs search to steadiness effectivity with empathy, real-time transcription has grow to be a functionality for delivering responsive, high-quality care at scale.
Switchboard, MD is a physician-led AI and information science firm with a mission to prioritize the human connection in medication. Its service improves affected person engagement and outcomes, whereas decreasing inefficiency and burnout. By designing and deploying clinically related options, Switchboard, MD helps suppliers and operators collaborate extra successfully to ship nice experiences for each sufferers and employees. One in every of its key options is streamlining the contact heart utilizing AI voice automation, real-time medical report matching, and steered subsequent steps, which has led to vital reductions in queue occasions and name abandonment charges.
With greater than 20,000 calls dealt with every month, Switchboard, MD helps healthcare suppliers in delivering well timed, customized communication at scale. Its AI platform is already serving to scale back name queue occasions, enhance affected person engagement, and streamline contact heart operations for clinics and well being programs. Clients utilizing Switchboard have seen outcomes comparable to:
- 75% discount in queue occasions
- 59% discount in name abandonment charge
Regardless of these early successes, Switchboard confronted a important problem: their present transcription strategy couldn’t scale economically whereas sustaining the accuracy required for scientific workflows. Value and phrase error charge (WER) weren’t simply operational metrics—they have been important enablers for scaling automation and increasing Switchboard’s impression throughout extra affected person interactions.
On this put up, we study the particular challenges Switchboard, MD confronted with scaling transcription accuracy and cost-effectiveness in scientific environments, their analysis course of for choosing the correct transcription resolution, and the technical structure they applied utilizing Amazon Join and Amazon Kinesis Video Streams. This put up particulars the spectacular outcomes achieved and demonstrates how they have been ready to make use of this basis to automate EMR matching and provides healthcare employees extra time to concentrate on affected person care. Lastly, we’ll have a look at the broader implications for healthcare AI automation and the way different organizations can implement comparable options utilizing Amazon Bedrock.
Selecting an correct, scalable, and cost-effective transcription mannequin for contact heart automation
Switchboard, MD wanted a transcription resolution that delivered excessive accuracy at a sustainable price. In scientific settings, transcription accuracy is important as a result of errors can compromise EMR report matching, have an effect on really helpful remedy plans, and disrupt automated workflows. On the similar time, scaling assist for hundreds of calls every week meant that inference prices couldn’t be ignored.
Switchboard initially explored a number of paths, together with evaluating open supply fashions comparable to Open AI’s Whisper mannequin hosted regionally. However these choices introduced tradeoffs—both in efficiency, price, or integration complexity.
After testing, the workforce decided that Amazon Nova Sonic supplied the correct mixture of transcription high quality and effectivity wanted to assist their healthcare use case. The mannequin carried out reliably throughout dwell caller audio, even in noisy or variable situations. It delivered:
- 80–90% decrease transcription prices
- A phrase error charge of 4% on Switchboard’s proprietary analysis dataset
- Low-latency output that aligned with their want for real-time processing
Equally essential, Nova Sonic built-in easily into Switchboard’s present structure, minimizing engineering elevate and accelerating deployment. With this basis, the workforce lowered guide transcription steps and scaled correct, real-time automation throughout hundreds of affected person interactions.
“Our imaginative and prescient is to revive the human connection in medication by eradicating administrative obstacles that get in the best way of significant interplay. Nova Sonic gave us the pace and accuracy we wanted to transcribe calls in actual time—so our clients can concentrate on what actually issues: the affected person dialog. By decreasing our transcription prices by 80–90%, it’s additionally made real-time automation sustainable at scale.”
– Dr. Blake Anderson, Founder, CEO, and CTO, Switchboard, MD
Structure and implementation
Switchboard’s structure makes use of Amazon Connect with seize dwell audio from each sufferers and representatives. Switchboard processes audio streams by means of Amazon Kinesis Video Streams , which handles the real-time media conversion earlier than routing the information to containerized AWS Lambda features. Switchboard’s Lambda features set up bidirectional streaming connections with Amazon Nova Sonic utilizing BedrockRuntimeClient’s InvokeModelWithBidirectionalStream API. This novel structure creates separate transcription streams for every dialog participant, which Switchboard recombines to create the whole transcription report. All the processing pipeline runs in a serverless atmosphere, offering scalable operation designed to deal with hundreds of concurrent calls whereas utilizing Nova Sonic’s real-time speech-to-text capabilities for instant transcription processing.
Nova Sonic integration: Actual-time speech processing
Harnessing Amazon Nova Sonic’s superior audio streaming and processing, Switchboard developed and constructed the aptitude of separating and recombining audio system’ streams and transcripts. This makes Amazon Nova Sonic significantly efficient for Switchboard’s healthcare functions, the place correct transcription and speaker identification are essential.
Amazon Nova Sonic affords configurable settings that may be optimized for various healthcare use circumstances, with the pliability to prioritize both transcription or speech era based mostly on particular wants. A key cost-optimization function is the power to regulate speech output tokens – organizations can set decrease token values when primarily targeted on transcription, leading to vital price financial savings whereas sustaining excessive accuracy. This versatility and value flexibility makes Amazon Nova Sonic a useful instrument for healthcare organizations like Switchboard trying to implement voice-enabled options.
Why serverless: Strategic benefits for healthcare innovation
Switchboard’s selection of a serverless structure utilizing Amazon Join, Amazon Kinesis Video Streams, and containerized Lambda features represents a strategic determination that maximizes operational effectivity whereas minimizing infrastructure overhead. The serverless strategy eliminates the necessity to provision, handle, and monitor underlying infrastructure, in order that Switchboard’s engineering workforce can concentrate on growing scientific automation options fairly than server administration. This structure gives built-in fault tolerance and excessive availability for important healthcare communications with out requiring intensive configuration from Switchboard’s workforce.
Switchboard’s event-driven structure, proven within the following determine, allows the system to scale from dealing with dozens to hundreds of concurrent calls, with AWS routinely managing capability provisioning behind the scenes. The pay-as-you-go billing mannequin helps Switchboard pay just for compute assets used throughout name processing, optimizing prices whereas eliminating the chance of over-provisioning servers that might sit idle throughout low-volume durations.
Conclusion
Switchboard, MD’s implementation of Amazon Nova Sonic demonstrates how the correct transcription know-how can remodel healthcare operations. By reaching 80–90% price reductions whereas sustaining clinical-grade accuracy, they’ve created a sustainable basis for scaling AI-powered affected person interactions throughout the healthcare trade.
By constructing on Amazon Bedrock, Switchboard now has the pliability to broaden automation throughout extra use circumstances and supplier networks. Their success exemplifies how healthcare innovators can mix accuracy, pace, and effectivity to rework how care groups join with sufferers—one dialog at a time.
Get began with Amazon Nova on the Amazon Bedrock console. Study extra about Amazon Nova fashions on the Amazon Nova product web page.
Concerning the authors
Tanner Jones is a Technical Account Supervisor in AWS Enterprise Help, the place he helps clients navigate and optimize their manufacturing functions on AWS. He focuses on serving to clients develop functions that incorporate AI brokers, with a selected concentrate on constructing protected multi-agent programs.
Anuj Jauhari is a Sr. Product Advertising Supervisor at AWS, the place he helps clients innovate and drive enterprise impression with generative AI options constructed on Amazon Nova fashions.
Jonathan Woods is a Options Architect at AWS based mostly in Nashville at present working with SMB clients. He has a ardour for speaking AWS know-how to companies in a related approach making it straightforward for patrons to innovate. Outdoors of labor, he tries maintaining together with his three youngsters.
Nauman Zulfiqar is a senior account supervisor based mostly in New York working with SMB purchasers. He loves constructing and sustaining sturdy buyer relationships, understanding their enterprise challenges and serving because the buyer’s major enterprise advocate inside AWS.
