Bolger linked the rise of voice AI to a market gap between growing customer expectations and limited AI deployment capacity among telecom providers. He said mid-tier operators were watching AI reshape their industry but often lacked the budgets, internal expertise, and deployment cycles needed to roll out such systems themselves.
Bolger also described how voice AI is being integrated into telecom infrastructure, shifting voice from a commodity into an “intelligent service layer.” He said SimplyAI integrates directly into telecom systems, enabling AI agents to operate within existing phone environments and handle tasks such as answering calls, booking appointments, and updating CRM systems “in whatever language the customer speaks.”
From Demos to Deployment
Panelists emphasized connecting AI to business systems and workflows so agents can complete tasks rather than just respond to queries. Adekunle said failures in production stem less from “wrong answers” than from “incorrect actions, broken workflows, or edge cases” that were not designed for.
Panelists also noted that while building voice AI prototypes has become relatively easy, deploying reliable systems at scale remains more difficult. “Anybody can build a voice AI in five to ten minutes,” Sujith said, adding that production-grade systems require years of experimentation.
As part of that, Sujith pointed to domain-specific tuning as one area of differentiation, saying Wec.ai focuses on a small number of verticals and uses multiple specialized agents within a single interaction. He also highlighted governance and guardrails as critical to monitoring performance and maintaining reliability in production.
Trust and Scale
As AI agents increasingly interact directly with customers, speakers highlighted the need to ensure reliability and build trust. Approaches discussed included constraining workflows to defined use cases, grounding responses in real business data, and providing visibility into system decisions.
Adekunle stressed observability, while Bolger said trust ultimately comes from results. “When a tradesperson sees that every call was answered, or every inquiry was captured, or they’ve got new bookings they could have missed, that’s what builds trust,” he said.
Bolger also emphasized that multilingual capability is built into voice AI systems rather than added later. He said localization needs to be handled natively, noting that mistranslations or incorrect responses in local markets can quickly undermine trust.
On scaling, speakers pointed to partnerships, limited initial use cases, and repeatable workflows. Bolger said his company builds alongside partners rather than in isolation, while Adekunle said BimpeAI started with simpler, lower-value use cases before moving into more complex, higher-value workflows.
Looking ahead, Sujith highlighted ongoing development in areas such as proactive agents, persistent memory, and multimodal interactions across voice and messaging channels. He estimated that “80 to 90%” of processes could be handled by AI agents in the near term, with humans focusing on more complex or high-value interactions. Adekunle expressed a similar view.
