In a customer story published by OpenAI on October 27, 2026, voice AI platform Retell AI said it now serves more than 4,000 businesses, with monthly call volume growing from 30 million in early 2026 to more than 100 million by September. According to the writeup, OpenAI models generate responses for 80% of calls on Retell, and the platform has surpassed 1 trillion tokens consumed across those models. Retell's platform lets contact-center teams define an agent's business context, goals, and rules; during a call its harness runs audio models and manages turn-taking, while OpenAI models generate responses and request tool calls that Retell executes through integrations. When human judgment is needed, the agent briefs a representative before connecting the caller. The piece cites two customers: electronics company Anker reported an 80.4% case resolution rate and a Net Promoter Score of 63 for Tier-1 post-sales support, and Pine Park Health said it recovered medical-assistant time equivalent to 2.1 full-time roles, with a 55.7% booking rate when its scheduling agent reached a patient. Retell also described its vCX benchmark, in which it says GPT-6 Astra ranked first for instruction following at 89.7% on October 13, 2026. These figures come from Retell and OpenAI; they have not been independently verified.
- Retell serves 4,000+ businesses; monthly calls grew from 30M (early 2026) to 100M+ by September
- OpenAI models generate responses for 80% of Retell calls; customer resolution rates reported above 80%
- Anker reported 80.4% case resolution and NPS of 63; Pine Park Health recovered time equal to 2.1 FTEs
- Figures are self-reported by Retell and OpenAI and not independently verified
What it means for you
This is a vendor case study: OpenAI showcasing a customer that uses its models. The useful part isn't the headline numbers — those are self-reported — it's the pattern it describes. Voice agents handling routine, repetitive calls (order tracking, warranty checks, appointment reminders) are now working in real contact centers, and the design keeps a person in the loop for anything requiring judgment. That's the sensible shape. The claimed resolution rates only hold up when the agent is pointed at narrow, well-defined tasks with clear rules.
Try this
Pull a week of your inbound calls or support tickets and sort them into two piles: 'routine and repetitive' (same questions, clear answers) versus 'needs a human to think.' That map tells you whether a voice agent would actually help — and it's worth doing before you talk to any vendor, because an unclear process automated is just confusion at scale.
Who should care
Owners and managers running phone-based customer support, scheduling, or order inquiries with high, repetitive call volume — especially if daytime peaks swamp your staff.
Skip this if
Your call volume is low, your calls are mostly complex or one-off, or you don't handle customer phone traffic at all. This is also safe to skip if you're not actively shopping for a contact-center tool — it's a marketing writeup, not a decision point.
Sources: OpenAI — read the original