Support AI tally, November 2026: seven deployments, 25% to 95%, no prices

• 1. Seven support deployments reported that AI handles between 25% and 95% of contacts, and each used a different denominator.
• 2. None of the seven published what the AI costs; the one company with dollar figures, Airbnb, saved $26 million on outside support in a quarter while its whole support line rose 8% to $393 million.
• 3. One of the seven gave an accuracy figure, ENGIE's 83%, with no definition; none gave an error rate.
• 4. None reported fewer support staff of its own; Airbnb's support payroll rose $21 million in the quarter.
• 5. Three of the seven gave a before-and-after number; the other four gave results with no baseline.

The seven: three deployments this site covered on 12 and 13 November 2026 ([Airbnb](https://modelsatwork.news/article/airbnb-ai-assistant-40-percent-support-issues-cost-per-booking-17-percent-10-q), [American Family Care](https://modelsatwork.news/article/american-family-care-flip-voice-agent-call-to-visit-11-to-34-percent-abandonment-40-to-14), [MUSC Health](https://modelsatwork.news/article/musc-health-emily-ai-phone-agent-25-percent-patient-calls-92-lines-soundhound)), two it recorded as Signals (Via Cristais, Live Nation) and two Salesforce customer pages read for this piece (ENGIE and Engine, both undated). Four of the seven are published by the seller: Salesforce for ENGIE, Engine and Live Nation, Google Cloud for Via Cristais. Two are company statements relayed by trade press. One, Airbnb, rests on filings with the US Securities and Exchange Commission. This is a count of what companies chose to publish, and companies publish wins more readily than failures: no rollback is among the seven. It cannot show a typical result.

1. From 25% to 95%, on seven denominators

• MUSC Health: 25% of patient calls, as of June 2026, on a voice agent that answers 92 phone lines.
• Airbnb: about 40% of all support issues in the third quarter of 2026, up from a third the quarter before; its [shareholder letter](https://www.sec.gov/Archives/edgar/data/1559720/000119312526449727/d437988dex991.htm) says the figure covers every issue, "not just those that begin in the AI assistant".
• Engine, a business travel platform: 50% of incoming chat cases, [Salesforce says](https://www.salesforce.com/customer-stories/engine/agentic-service/). Its client services team handles more than 800,000 requests a year across all channels.
• ENGIE: 71% of cases, from more than 1,500 conversations a day, [Salesforce says](https://www.salesforce.com/customer-stories/engie/). The agent sits in ENGIE's Belgian customer app; financial transactions and contract changes go to a rep.
• Via Cristais, a Brazilian highway operator: about 80% of contacts in the first month across phone, website chat and WhatsApp, [Google Cloud says](https://cloud.google.com/customers/via-cristais). Accidents and vehicles stopped in dangerous places go straight to a person.
• Live Nation: 95% of the questions its venue agent handles are answered without a hand-off, [Salesforce said on 16 September 2026](https://www.salesforce.com/news/press-releases/2026/09/16/live-nation-agentforce-expansion/). The questions are of the kind it lists: parking, door times, accessible seating.
• American Family Care gave no share handled. It reported the share of calls ending in a booked visit, up from 11% to 34%.

Wren's read: the percentage rises as the job narrows, from every issue a travel marketplace receives to door times at a concert venue. The list is not a ranking.

2. No price in any of the seven

None of the seven gives a licence, usage or build cost. List prices exist: Zendesk's pricing page shows $1.50 to $2.00 per automated resolution, Intercom's shows $0.99 per outcome for its Fin agent, and Salesforce's shows $0.10 per Agentforce action, all read on 14 November 2026. What ENGIE, Engine or Live Nation pays Salesforce is not disclosed. Airbnb's [quarterly report](https://www.sec.gov/Archives/edgar/data/1559720/000155972026000032/abnb-20260930.htm) is the only document with money in it. Professional and third-party service costs fell $26 million in the third quarter, which Airbnb attributes to AI lowering agent contact volume. Payroll in the same line rose $21 million, customer relations costs, which include make-good payouts, rose $33 million, and the line went from $365 million to $393 million.

3. One accuracy figure, no error rates

Salesforce's ENGIE page says the agent delivers consistent messaging "with 83% accuracy". It does not say who checked, on how many conversations, or what the customer received the other 17% of the time. The other six give no accuracy figure. Three of the agents do work where a wrong action costs money: AFC's answers billing and self-pay price questions, MUSC Health's takes bill payments and prescription refill requests, and Engine's cancels and changes bookings. None publishes a wrong-action rate. Engine tracks "topic success rates" on a dashboard, Salesforce says, and publishes none of them.

4. No one reports fewer staff of their own

Airbnb's saving sits in a cost category that its filing says mainly pays outside support partners, while its own payroll rose on higher average headcount. ENGIE's Salesforce lead, Robbie Dugardijn, says: "The goal isn't to reduce our workforce, but to give them more meaningful work." AFC's chief executive says less time on the phone means more time with patients. None of the seven states support headcount before and after, how staff were retrained, or who reviews the AI's answers.

Wren's read: a vendor page that is silent on headcount is not evidence that jobs stayed. The one filing in the set found the saving in an outsourcing bill.

5. Three of seven have a baseline

AFC says call abandonment fell from 40% to 14%, with no measurement period. Airbnb's share rose from a third to about 40% in one quarter. Via Cristais says waits went from 60 seconds to none. Engine reports chat satisfaction up 16% and average handle time, the minutes a case takes, down 15%, with no starting values. On handle time the sources pull apart: the vendor Cresta [told CX Today](https://cxtoday.com/handle-time-rose-and-nobody-complained-heres-why) that when AI takes the simple contacts, the calls left for people run longer. Engine's page does not say whether its 15% includes cases the agent closed alone.

What to watch

Airbnb's next quarterly filing will show whether the 40% and the support expense line start to move together. Zendesk has announced plans billed by outcome with [no rate published]({A(ZD)}). The next edition of this tally will add any deployment that discloses a price or an error rate, and any that is switched off.

What to do if you run support

• Write the denominator next to any resolved-by-AI percentage a vendor shows you: every contact, one channel, or only the conversations the assistant took. Ask how an abandoned conversation is counted.
• Before a pilot, record your abandonment rate, share of contacts resolved without a person and cost per contact for the last full quarter, with dates. Four of these seven published no before-and-after figure.
• Ask finance to split support spend into outsourcer fees, in-house payroll and refunds or goodwill payouts, as Airbnb's filing does, so a saving in one line is not reported as a saving in the total.
• For any task that moves money or changes a booking, get the vendor's wrong-action rate and the price at your monthly volume in writing.