Voya's AI buildout is an operating story, and its certification is the moat

Nearly Voya's entire global workforce completed an AI literacy certification in the second quarter, while 98% of employees were already using AI-powered tools such as Microsoft Copilot and an AI-enabled support system was handling 7 million customer interactions a year—almost 3 million of them calls closed through AI-assisted self-service. Voya serves more than 18 million customers, and the push covers its retirement and investment management operations.

The rollout reaches customer service, internal productivity, Voya's own employee-benefits administration, and cybersecurity threat detection, with certification layered on so employees can demonstrate practical use. Rajat Kalia, Voya's chief technology officer for workplace technology, described AI as "more than a technological initiative" and "a cultural shift" in equipping people, and said the tools go where they produce results—faster answers, better decision-making, more employee time for higher-value work. Inside retirement operations the near-term job is unglamorous: sorting millions of inbound emails and routing each request to the right desk.

Voya's summer already included a recordkeeping mandate for the District of Columbia's public retirement plans, which this publication covered in August, and enterprise AI is now among the things sponsors weigh when they score a search.

Notice what the announcement does not measure: certification rates and tool adoption are workforce statistics, saying nothing about whether a participant's emailed request was handled correctly or how long a self-service session took. The 96% positive feedback Voya reports is the figure a sales team will lead with, unaccompanied by any detail on how it was measured or over what period.

The certification matters more than the chatbot because tooling of this kind can be licensed, while teaching nearly an entire workforce to use it well—and being able to say so in an RFP response—takes quarters of internal work. AI is moving from pilot to production in DC consulting paperwork, and the split will come down to the quality of the source trail a firm can produce, not which model it bought. Voya's release is silent on that point: it does not say which participant-facing decisions the virtual agent may make unsupervised, or what record exists when a request is routed by a model and a plan auditor asks. The number worth watching next is a cost-to-serve figure alongside the satisfaction score—that is the one a sponsor can take into a fee negotiation.