Benefits
Overview
Roojoom delivers artificial intelligence (AI)-powered orchestration to help businesses improve their user success, customer engagement, and customer service journeys. The company recently worked alongside Amazon Web Services (AWS) to upgrade PickMyCall, an AI-voice agent designed for SMBs to optimize inbound call handling, lead capture, and customer engagement processes. Using Amazon Nova Sonic—a speech-to-speech foundation model (FM) for conversational AI—Roojoom implemented a speech-to-speech native architecture. After implementing Amazon Nova Sonic, Roojoom has achieved reduced latency, improved security, and better price performance for PickMyCall.
About Roojoom
Serving enterprises across telecommunications, insurance, finance, and software, Roojoom uses personal AI engines to help businesses achieve their desired outcomes. The company has offices in Tel Aviv, Israel; Dallas, Texas; and Paris, France.
Opportunity | Transforming every call into a business opportunity
For small and midsize businesses (SMBs), customer service can be a powerful way to stand out from the competition. It can boost customer retention and loyalty, create upsell opportunities that drive revenue, and strengthen long-term credibility. But delivering great customer service takes commitment, resources, and time—things that are often in short supply for small businesses that constantly juggle the responsibilities of daily operations. Every missed call is a missed opportunity not only to deliver great service but also to generate leads and build lasting relationships with customers.
Roojoom understood that reality and created PickMyCall to help SMBs deliver exceptional customer service even when they can’t answer the phone. The application delivers AI-powered features that are designed to meet the needs of small businesses, such as automated call and chat management, lead capturing, appointment scheduling, and smart call prioritization. It empowers SMBs to turn every call into a business opportunity—boosting customer satisfaction, driving revenue, and improving productivity.
With any speech-to-speech solution, latency can make or break the user experience. Stilted exchanges with an AI assistant don’t just cost customers time—they can quickly lead to frustration. “Even 1 second on a phone call is extremely important,” says Anat Harel, vice president of research and development at Roojoom.
Solution | Going speech-to-speech native with Amazon Nova Sonic
To meet the standards that business owners expect, PickMyCall needed an FM that could facilitate human-like conversations with low perceived latency. So Roojoom implemented Amazon Nova Sonic, which delivers near real-time, human-like voice interactions with exceptional price performance and low latency. In conversation, the FM offers smooth dialogue flow, adaptive speech responses, and outstanding speech recognition. And importantly for Roojoom, Amazon Nova Sonic supports native speech-to-speech architectures, combining speech comprehension and generation into a single model.
Earlier versions of PickMyCall operated within a traditional speech-to-text-to-speech architecture. The pipeline used a service that converted speech into text, which was routed to a large language model that generated a relevant text response. This was then sent to a third-party solution for text-to-speech synthesis, ultimately producing a natural-sounding audio reply.
Although the workflow was functional, every stage of the original architecture introduced latency, which affected conversational flow. “Trying to integrate smarter processes every step of the way was creating more and more latency,” says Or Meirov, chief technology officer at Roojoom. The PickMyCall team also encountered significant timing and synchronization issues when integrating text-to-speech, which resulted in some calls disconnecting entirely. There were also repeated instances of mispronunciation and customer interruption, negatively affecting the customer experience.
Implementing Amazon Nova Sonic helped the PickMyCall team quickly create more engaging AI voice experiences with lower latency—and with improved price performance. In less than 2 weeks, the team built a working prototype to test critical functionality before moving to a production-grade minimum viable product. “The simplicity of implementing native speech-to-speech with existing frameworks was amazingly fast,” says Meirov. “The AWS prototyping team made it all very quick to learn and implement.”
Another priority for Roojoom was security. “We have an enterprise mindset,” says Meirov. “That means quality security.” Amazon Nova Sonic is available in Amazon Bedrock, a comprehensive, secure, and flexible solution for building generative AI applications and agents. Accessing Amazon Nova Sonic through Amazon Bedrock provides Roojoom with full control over the data used to customize generative AI models. All data is encrypted in transit and at rest, and businesses can access a suite of additional security capabilities for added protection.
More broadly, PickMyCall is also using multiple AWS solutions to maximize efficiency. For example, the team is using Amazon OpenSearch Service to simplify AI-powered search, observability, and vector database operations, all without the need to manage, monitor, and maintain infrastructure. “Using Amazon OpenSearch Service helps us to focus on innovation, not maintenance,” says Harel.
Outcome | Subsecond average latency powers human-like conversation
Adopting Amazon Nova Sonic has helped Roojoom improve PickMyCall’s conversational flow and has reduced maximum latency from 7 seconds to 2 seconds. Even more impressively, average latency has been reduced from approximately 4.5 seconds to 800 ms—more than an 82 percent improvement. The team has also recorded improved minimum latency, which has fallen from 2 seconds to 300 ms. “It’s a major, major change,” says Harel.
Roojoom has also begun to implement and test Amazon Nova Understanding Models—state-of-the-art models that deliver frontier intelligence and excellent price performance—for what Meirov describes as “around-the-call” processes. “One process involves understanding what happened on the call,” says Meirov. “We need to summarize the transcription and to provide actionable tags, such as ‘This person wanted to call back,’ ‘This person wants to set a meeting,’ and so on.”
PickMyCall is already being user tested in different international markets. "We're running pilots with customers in Germany, France, and the United States," says Meirov. Behind the scenes, the team works with Amazon Nova Sonic to deliver an even better product to its customers. "Amazon Nova Sonic feels like having a conversation with a human being, not a conversation with a system," says Meirov.
Amazon Nova Sonic, feels like having a conversation with a human being, not a conversation with a system.
Or Meirov, Chief Technology Officer, Roojoom
AWS services used
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages