Retell AI uses Google Cloud and Gemini to power real-time voice agents.
Cutting voice latency to sub-600ms to build the AI contact center of the future
Redesigning contact centers from the ground up
Traditional contact center workflows require customer service representatives to manually log calls and navigate fragmented databases. These tedious and repetitive workloads drive incredibly high turnover rates, which substantially increases overhead costs related to recruiting, training, and preserving institutional knowledge. For Bing Wu, Co-founder and CEO of Retell AI, this widespread operational bottleneck presented a clear call to action: “Contact centers across verticals are based on error-prone and repetitive manual tasks. We saw an opportunity to fundamentally redesign them, much like how modern warehouses have moved from manual labor to automation.”
These teams have typically used Interactive Voice Response (IVR) solutions to help reduce burnout, but these solutions are often unreliable or frustrating for customers to navigate. Retell AI is using large language models to replace traditional tools, rather than simply patching and building on top of legacy systems. Since scaling out of Y Combinator and raising a $4.6M seed round in 2023, the company has grown to more than $100M in annualized run rate revenue on its journey to reinvent contact center communications.
The team identified major use cases across multiple industries, including healthcare, financial services, retail, and more. "Contact centers across verticals are based on error-prone and repetitive manual tasks,” says Bing Wu, Co-founder and CEO of Retell AI. “We saw an opportunity to fundamentally redesign them, much like how modern warehouses have moved from manual labor to automation.”
To make this large-scale change possible, the team had to construct a solution that could maintain sub-second voice latency across millions of interactions. This meant preserving a natural feeling to each conversation without awkward pauses or the AI talking over callers. The platform also has to manage massive fluctuations in traffic. Daily call volumes increase more than 50 times between overnight lows and daytime peaks across more than 100 million monthly voice calls.
Orchestrating real-time voice and AI quality assurance
Retell AI turned to Google Cloud to consolidate its core AI workloads with Compute Engine and Cloud Run and create a high-performance orchestration layer. By centralizing AI workloads from multiple models in a single, low-latency infrastructure, Retell AI has reduced cross-cloud latency from 24ms to less than 10ms. It also now consistently achieves sub-600-millisecond voice response times and automates 70% to 85% of call center interactions for major enterprise clients.
The company has also conducted a rigorous internal benchmarking test of more than thirty audio-capable models, reviewing hallucination rates, tool-calling accuracy, latency, and token cost. Gemini 3.5 Flash-based agents emerged as the clear winner for Retell AI's automated Quality Assurance (QA) engine. "During our bake-off, Gemini came out on top," Wu shares. "Its best-in-class audio capabilities enable our AI QA agents to analyze call nuance and turn-taking independently and objectively, giving our customers unprecedented visibility into performance."
Built on Google Cloud: Retell AI’s system architecture integrates Gemini 3.5 Flash to automate voice workflows and evaluate up to 100% of customer calls in real time.
Unlike traditional QA that manually samples a fraction of call transcripts, Retell AI's Gemini-powered agent asynchronously evaluates up to 100% of calls. It analyzes native audio nuance, turn-taking, tonality, expressiveness, and background noise to generate objective performance scores. In a single 30-day window, Gemini powered over 21,000 hours of audio evaluation across the platform.
Transforming the future of voice AI with real-time review and iteration
Google Cloud has helped generate nearly $1 million in revenue across enterprise accounts. At the performance level, the infrastructure can easily handle spikes during busy inbound seasons or outbound marketing campaigns without any disruption in service. By collaborating with the Google team, the team continues to find new ways to expand its offerings.
The Google Cloud team brings a genuine and product-focused mindset. They listen to our pain points and help us on our journey, so we can focus on democratizing voice AI and building full-coverage contact center solutions.
With that roadmap in mind, Retell AI is moving toward a closed-loop customer experience engine. This solution will use automated QA insights from Gemini in agent prompt optimization in near real-time. The team is also implementing real-time customer satisfaction monitoring during live calls and expanding features for routing, warm transfers, and CRM synchronization to truly build the AI call center of the future.