Twilio Publishes CX Benchmarks From 5 Enterprise AI Deployments

The numbers from five deployments

Twilio gathered the figures for its Innovator's Report, based on conversations with CIOs and CTOs at known enterprise companies. According to Twilio, most published benchmarks come from surveys of self-reported averages, which show what people say they achieve rather than what a specific deployment produced.

• Sweetwater (retail): 50,000 texts and 40,000 calls per day, 850 agents supported, and a contact center configuration deployed in under two days.
• WestCX (healthcare): 50 to 80% higher conversion on RCS than on SMS and email, 3x customer engagement, and a 20% revenue uplift from reduced no-shows.
• Eltropy (financial services): voice AI containment of up to 70%.
• Nova Gestões (financial services): reduced first response time and cost per contact, with no figure published.
• Lendi Group (lending): improved response time, engagement and conversion, with no figure published.

Two of the five companies reported only direction, not magnitude. Twilio says this is common in regulated industries, where specific operational figures are sensitive, and advises keeping it in mind when making comparisons.

What the containment figure really means

Containment rate is the share of customer conversations an AI agent handles from start to finish without passing them to a human. Twilio gives the example of a voice agent that fields 1,000 calls, of which 720 never reach a live agent: a 72% containment rate.

Eltropy's figure comes from voice deployments at community banks and credit unions, covering high-volume inquiries such as balance checks and transaction history. Voice accounts for 65% of all customer communication at those institutions. Twilio calls the result an upper band for well-scoped, transactional voice automation and says it will not transfer to a general-purpose agent.

Twilio also flags a caveat. Containment counts conversations the AI resolved and conversations where the customer gave up and stopped replying in the same way. According to Twilio, 71% of consumers will abandon a conversation entirely if an AI fails to recognize who they are, and each such abandonment improves the containment rate. The company recommends splitting containment by outcome before benchmarking against anyone.

Speed and patience

Twilio cites its Inside the Conversational AI Revolution report, in which 92% of consumers say AI typically replies in under 30 seconds. A sub-30-second response is therefore an expectation, not an achievement.

Patience is the more telling figure, according to Twilio. 39% of consumers will not wait at all for an AI agent, while 88% will wait for a human. Consumers will wait 8.5 minutes for a person against 4.3 minutes for AI. Twilio also says ineffective AI agents waste 81 seconds per interaction on failed attempts before escalating.

A satisfaction gap of 30 to 35 points

Two Twilio studies asked the same question and found similar gaps. In the 2026 Customer Insights Series, 86% of companies believed customers were satisfied with their AI-powered service, while only 51% of consumers agreed. In Inside the Conversational AI Revolution, 90% of organizations believed customers were satisfied, against 59% of consumers.

Twilio notes the samples and years differ, so it does not draw a trend line. It does say internal satisfaction estimates run 30 to 35 points above what customers report. Benchmarking your own AI CSAT against an industry average may therefore compare one optimistic number with others. The FAQ in the post adds that post-interaction surveys only reach customers who stayed.

How Twilio suggests benchmarking

Twilio says most benchmark disappointments are measurement mismatches rather than performance gaps. Its recommendations:

• Split containment by outcome, separating resolved from abandoned.
• Measure resolution on the underlying issue, across sessions and channels.
• Track repeat contact within seven days, which needs no survey and cannot be gamed, according to Twilio.
• Measure abandonment by stage, since CSAT cannot see this group.
• Time the handoff separately. 76% of consumers say the human agent they reach has little or no context.

Twilio advises picking two or three of the benchmarks, checking whether the scope resembles your own, and measuring the same way. Its point is that each figure is tied to a named company, industry and scope that can be evaluated, which makes the numbers more useful than their size alone.