Customer service teams spent 2025 deciding whether AI agents could handle real customers. In 2026, most of them stopped asking. Salesforce’s State of Service: AI Agents Edition, published in May 2026, found that 66% of customer service organizations now use agentic AI, up from 39% a year earlier. Salesforce sells AI agents, and the survey is its own.
The harder question now sits with the people who run those teams. When an AI agent closes a chat or ends a call, did the customer actually get what they came for? And who checked?
Orvera AI, a San Francisco company founded in 2024, is an AI agent company for customer experience teams. It builds, deploys and runs AI agents across voice, chat and digital channels. It also gives the human agents on the same team Agent Assist during live conversations, AI quality management after them and Voice of Customer analysis across all of them. Orvera AI’s bet is that the quality check has to cover both kinds of agent, and that it has to look past the transcript.
Most conversations were never checked to begin with
Quality assurance in customer service has long run on samples. A supervisor pulls a handful of calls each week, scores them against a form and coaches from there. In a January 2026 survey of 500 full-time support agents by Solidroad, which sells AI coaching software, 81% of agents said the majority of their conversations are never reviewed for quality. More than a third, 37.4%, said fewer than 10% of their interactions get any QA attention at all.
Solidroad’s survey covered human agents only. AI agents now add their own conversations to the same review queue. The Salesforce report also points to a trust gap. Sixty-five percent of service professionals said their customers fully trust AI, while the Metrigy Consumer CX Index, cited in the report, puts the share of consumers who trust AI with their service needs at 44%.
Checking the CRM along with the transcript
Orvera AI scores every conversation on the same scorecard, whoever handled it. Its AI quality management product reads the customer’s CRM and case system alongside the conversation.
“Quality in a contact center means what gets done for the customer as well as what gets said,” Ali Merchant, CEO of Orvera AI, said in an October 21 statement. “The refund gets processed, the case gets updated and the callback gets booked. Orvera’s Auto QA reads the CRM and case system alongside the conversation, so it scores what the agent did as well as what they said, for human and AI agents alike, against each customer’s own scorecard.”
Each customer brings its own scorecard, weights and compliance checks, and every score links back to the evidence in the transcript. Scores for AI agents and human agents land in the same view, so a team lead can compare the two directly.
Three kinds of service work where volume spikes
Orvera AI is concentrating on three kinds of customer service work. In e-commerce, order volume keeps rising. In its August 2026 release, the U.S. Census Bureau estimated seasonally adjusted retail e-commerce sales at $340.2 billion for the second quarter of 2026, up 12.2% from a year earlier. Each order can turn into a question about delivery, returns or a refund.
In travel, a single disruption can hit every channel at once. When the Federal Aviation Administration ordered flights cut at 40 U.S. airports in November 2025, ramping to 10%, American Airlines planned to cut 220 flights a day, KTVU reported, drawing on the Associated Press. Each canceled seat became a passenger looking for a new itinerary.
In insurance, the first report of a claim still runs heavily through people. Among homeowners who filed a claim, J.D. Power’s 2026 U.S. Property Claims Satisfaction Study found that 38% reported their first notice of loss digitally. Orvera AI published one auto insurance deployment in September. At a U.S. auto insurer, its voice and chat agents fully handled 68% of loss reports, and every interaction went through the same quality review for required disclosures, empathy and completeness of the loss facts. Conversations involving injuries, complex coverage questions or a distressed customer go to a human adjuster, with the conversation and suggested next steps already on screen.
What service leaders will be measured on
Zendesk, which also sells customer service software, found in its CX Trends 2026 report that 85% of CX leaders say customers will drop brands that can’t resolve issues on first contact. As AI agents take a larger share of those first contacts, service leaders will need a way to tell which conversations ended with the problem solved, whether a person or an AI was on the other end.
