A key technical obstacle in the field of AI voice customer service—delay and unnatural conversation experience—has finally been overcome. A telephone AI system that can switch models in real time and respond as quickly as humans has been developed by the joint efforts of three parties: Phonely, an artificial intelligence phone support company; Maitai, a reasoning optimization platform; and Groq, a chip manufacturer. This pushes voice AI into the "human-like new era."
According to disclosures, through this collaboration, Phonely's response speed increased by more than 70%, and the accuracy rate of AI conversations rose from 81.5% to 99.2%, surpassing GPT-4o's benchmark of 94.7%. This achievement comes from Groq's "zero-latency LoRA hot-swapping" technology, which enables seamless switching between multiple fine-tuned models without retraining and without increasing latency through Maitai's dynamic orchestration platform.
Groq's chips are specifically designed for language inference, with its LPU (Language Processing Unit), streaming architecture, and high-speed on-chip memory enabling models to respond within milliseconds. The "first response token time" dropped from 661ms to 176ms, and the overall completion time decreased from 1,446ms to 339ms, making AI phone calls almost indistinguishable from real humans.
Will Bodewes, CEO of Phonely, stated: "More than 70% of users cannot distinguish whether it is AI or humans in our system. Delay was once the most obvious sign, but now it is nearly gone."
The results have been impressive. Just this month, Phonely helped one client replace 350 human agents. Thanks to the fast deployment capability, companies can complete migration and go live within a day without modifying APIs and see performance improvements within a week.
This cooperation not only improves customer service experiences but also symbolizes a shift in enterprise AI trends: moving away from pursuing "large and comprehensive" general large models toward faster, more accurate, and more flexible multi-model fine-tuning systems. Christian DalSanto, founder of Maitai, said: "We believe the future of enterprise AI will be dominated by fine systems composed of specialized models, rather than isolated single giants."
For enterprises, this is not just an improvement in performance but also a disruption in cost structure. No need for scheduling, training, and managing large-scale customer service teams; AI customer service becomes a better solution. And the low-latency, highly controllable architecture provided by Groq and Maitai clears obstacles for large-scale AI deployment in industries such as insurance, law, and automobiles.
This breakthrough proves that the "uncanny valley" in technology is not insurmountable. The cooperation between Phonely, Groq, and Maitai not only provides a new paradigm for voice AI but also signals an impending dramatic transformation in traditional customer support industries. The era of voice AI has arrived earlier than expected.
