Primary source: SuperDial via GlobeNewswire, November 17, 2026, 8:30 a.m. ET, San Francisco. SiliconANGLE, Mike Wheatley, updated 8:00 a.m. EST the same day, is the interview.
On November 17, 2026, at 8:30 a.m. ET, SuperDial said its AI agents have completed more than 10 million calls to insurers, and that the next product judges success by the resolved request rather than the completed call. The company calls it Agent-to-Agent RCM. If you buy voice automation for eligibility, claim status, or prior authorization, the phone is no longer the product SuperDial wants you to count. The call is the last channel, and only the minutes on a successful request are in the release.
Company-reported. Among successful requests where the required information is retrieved, SuperDial says its agents average approximately four minutes per call attempt against payer voice automation, and seventeen minutes or more when they reach a human representative. The release does not say what share of requests succeed, or what share never become a call.
SuperDial, based in San Francisco, says its agents handle payer communication for revenue cycle teams. The November 17 wire is the volume milestone and a change in what the company says it is optimizing. The subhead on the wire is "The bot wars are here." The operating model under it is a channel order.
| What the November 17 release states | What it does not state |
|---|---|
| More than 10 million AI-powered calls to insurers have been completed. | A customer count, a payer list, a date range on the 10 million beyond the milestone itself, or a resolution rate. |
| Agent-to-Agent RCM scores the request, not the call. If an electronic transaction can answer it, SuperDial uses that. If the information is on a payer portal or a faxback, the agents retrieve it there. A phone call happens when it is unavoidable, carries forward what is already known, and asks only for what is missing. | The share of requests resolved with no call. The names of the electronic transactions. A price. |
| On successful requests where the required information is retrieved, about four minutes per call attempt with payer voice automation, against seventeen minutes or more with a human representative. Those figures are SuperDial's. | A result for the requests that fail, or for the attempts that never retrieve the information. The four minutes are not an average of every call. |
| The long-term ambition is an AI-native network in which agents exchange trusted information directly. The wire also says the next ambition is "the next billion payer requests" through Agent-to-Agent resolution. | A network that exists today, a timeline, or a definition of a payer request. Both lines are the company's ambition. |
"The future of healthcare administration cannot be two bots arguing with each other faster," said Sam Schwager, co-founder and CEO of SuperDial, in the November 17 release. "If an agent already has the information, that context should travel with the request. If a digital channel can answer it, there should be no call. The goal is fewer exchanges, not faster ones."
SiliconANGLE's Mike Wheatley, in a story updated at 8:00 a.m. EST on November 17, writes that Schwager calls the bot-versus-bot calls "bot wars." The sentence Wheatley quotes, which is not in the wire, is the definition SuperDial wants on the scorecard. "The difference between this and the bot wars is the definition of done," Schwager said. "Two agents were optimizing for their own side of the exchange, but now those two agents are optimizing for a usable answer — that is, the information the provider needs to move the account forward, delivered in response to a request the payer can actually answer."
Wheatley describes the first check as EDI feeds and APIs, then the payer's web portals, and a voice call only after those are exhausted. That sequence matches the wire's electronic transaction, then portal or faxback, then phone. The wire's About section says the platform works across phone, payer portals, EDI, APIs, fax, and documents, and that it can return structured results to integrated systems and work queues. The workflows named in that About section are eligibility and benefits verification, prior authorization, claims follow-up, credentialing, enrollment, and provider data management. The About section also says SuperDial maintains controls designed to support HIPAA compliance and has completed a SOC 2 Type II examination. Those are company statements about the platform. They are not a result of the 10 million calls.
The wire cites a 2021 McKinsey estimate that simplifying healthcare administration could save up to $265 billion a year. That figure is McKinsey's, as SuperDial cites it. It is not a SuperDial savings claim, and it is not a number this article applies to the 10 million calls.
It is not a published resolution rate. The four-minute figure is company-reported, and only for successful requests where the information was retrieved, per call attempt, against payer voice automation. SiliconANGLE's rendering drops the "successful requests" qualifier and says "about four minutes" against "17 minutes or more." This article uses the wire. The release does not name a customer count, a price, the payers on the 10 million calls, or the share of requests that never reached a phone. SiliconANGLE describes a prior financing and a separate payer agreement. Neither is in the November 17 company release, and neither is restated here.
This is RevCycleAI's read for the director who still staffs a payer-call team. The wire says providers are deploying agents to chase claim and benefit information, and payers are deploying automation to answer them. When both sides do that, the call can become two bots keeping an exchange going. Schwager's "bot wars" label, in the SiliconANGLE interview, is that loop. The metric that matters in a renewal is requests resolved, and which channel resolved them. A vendor that reports 10 million calls has proved it can stay on the line. It has not, in this release, proved how often the line was the wrong tool.
The electronic transaction the wire puts first is the one your clearinghouse already sends. Eligibility and benefits are a 270 out and a 271 back. Claim status is a 276 out and a 277 back. Prior authorization has its own electronic transactions, and a portal when the plan does not answer them. The November 17 release does not name those transactions. It says that if an electronic transaction can supply the answer, SuperDial uses it, and that the phone is for what is still missing. SiliconANGLE is the piece that says the first lookup is EDI and APIs. Hold the vendor to a channel log: 271 or 277 received, portal or faxback used, call placed, information retrieved or not. The four minutes belong only in the last of those rows, and only when the request succeeded.
The About section is the scope of the request, not a performance table. Eligibility, claim follow-up, prior authorization, credentialing, enrollment, and provider data are different workqueues. A four-minute claim-status call and a four-minute credentialing call are not the same labor. The release does not split the 10 million, and it does not split the minutes. Buy the log by workflow. Do not buy the blended minute.
The "next billion payer requests" and the AI-native network are the ambition in the wire. They are not a capacity figure. Until SuperDial publishes the share resolved without a call, the payers in the 10 million, and a price, the operating fact is the channel order plus a company-reported minute count on the calls that already worked.
SuperDial's wire puts more than 10 million AI calls to insurers on the board, then says the score is the resolved request, with the phone last. The only duration in the release is company-reported: about four minutes per attempt against payer voice automation, and seventeen minutes or more with a person, among successful requests where the information was retrieved. Customer count, resolution rate, the no-call share, price, and the payer list were not disclosed.
If you are renewing a payer-call vendor, ask for resolutions by channel, not calls completed. The electronic path comes first, then the portal or the faxback, then a call that already knows what the 271 or the 277 did not return. SuperDial described that order on November 17 and did not publish how often each step finishes the request. The four-minute number is real as a company statistic, with the success qualifier attached, and it is not a resolution rate.
What the vendor counted, and which channel actually closed the request.
Sources: SuperDial via GlobeNewswire, November 17, 2026, 08:30 ET, datelined San Francisco (page opened). More than 10 million AI-powered calls to insurers; Agent-to-Agent RCM scored by the resolved request; electronic transaction, then payer portal or faxback, then phone with known context; approximately four minutes per call attempt with payer voice automation versus seventeen minutes or more with a human representative, among successful requests where the required information is retrieved; Sam Schwager's quotation is verbatim from that release, where he is identified as co-founder and CEO. The 2021 McKinsey estimate of up to $265 billion is the release's citation, not a SuperDial result. The "next billion payer requests" and the AI-native network are the release's statement of ambition. The About section's channel list, workflow list, HIPAA-controls sentence, and SOC 2 Type II sentence are that section's. SiliconANGLE, Mike Wheatley, updated 8:00 a.m. EST, November 17, 2026 (page opened). Schwager's "bot wars" label and the "definition of done" quotation, including the following sentence, are verbatim from that story. Wheatley's description of EDI and APIs, then portals, then voice, is attributed to SiliconANGLE. SiliconANGLE also describes a prior financing and a payer agreement. Neither appears in the November 17 company release, and neither is restated here. No customer count, resolution rate, no-call share, price, or payer list was disclosed, and none is estimated. The 270/271 and 276/277 names are RevCycleAI's read of what an "electronic transaction" is in a provider revenue cycle. The release does not use those transaction numbers. The RevCycleAI analysis section is our read.