66% of customer service teams already run an AI agent — but Salesforce finally separates "deflection" from "resolution"
Salesforce's 2026 State of Service research measures 66% adoption, up from 39% a year earlier, with 70% of teams seeing measurable value within 60 days. Gartner forecasts agentic AI will autonomously resolve 80% of common requests by 2029. But independent analysts flag a gap: a ticket "closed without a human" (deflection) isn't the same as a customer actually helped (resolution) — and the second number is far harder to earn than the first.
Adoption doubled its pace in a year according to Salesforce: two out of three customer service teams run at least one AI agent in 2026. Gartner puts the horizon at 2029 for 80% autonomous resolution of common requests, with an associated 30% cut in operating costs. These are real, dated, mutually consistent numbers — the shift is real.
What these numbers don't say is the difference between a "deflected" ticket (closed without a human stepping in) and one actually resolved from the customer's point of view. A 90% deflection rate achieved on a highly structured ticket mix (password reset, hours, order status) doesn't carry over to a mixed-intent portfolio — and a customer who gives up midway is often still counted as "deflected" even though they got nothing. That's exactly the gap AppH's pilot contract measures differently: a target of 40% real resolution without a human at day 90, not a ticket-closure rate.
For AppH
- AppH sets a resolution target (customer helped, case closed) rather than a deflection one (ticket closed) — it's the pilot's contractual number, measured and reported as-is, even when it comes in under 80%.
- When the agent can't resolve something, it hands off to a human at the branch instead of faking a fix — the case stays open, and it's never counted as a win.
Against / the honest limit
- AppH doesn't yet have the ticket volume to publish a resolution curve as robust as the studies cited here — the 40%-at-day-90 figure is a contractual target for the ABCLIV pilot, not a market average.
- The deflection/resolution gap documented by these analyses comes from third-party studies of the English-speaking customer service market (SaaS, e-commerce); the data for business domiciliation in France still needs to be built, branch by branch.
What strikes us about these two numbers (66% adoption, 80% resolution promised for 2029) is that neither tells an owner what actually matters: how many of THEIR customers got real help. We'd rather announce a more modest target and hit it than repeat a deflection figure that flatters a dashboard while saying nothing about the customer left without an answer.
Reviewed by a human at AppH