Contact centre buyers are being handed a number and told to trust it. "80% resolution." "60% deflection." "AI CSAT that beats human agents." The figures come from vendor decks, and the definitions behind them are rarely the same from one vendor to the next. Before that number goes into a business case, it is worth knowing what it is actually measuring.
Deflection and resolution are not the same metric
Deflection counts a conversation that a human agent never touched. Resolution counts a problem the customer confirms is solved. A chatbot can deflect a high share of contacts by ending the conversation, closing the ticket, or routing the customer to a help article, without the underlying issue being fixed.
Gartner treats these as separate measures. Its widely cited forecast is itself a resolution number, not a deflection one. By 2029, agentic AI is predicted to autonomously resolve 80% of common customer service issues without human intervention, up from a much lower share in 2024, with an associated 30% cut in operating costs (Gartner, 2025). That is a five-year projection, not a claim about what is deployed today. A vendor citing "80% resolution" in 2026 is either quoting that forecast out of context or reporting a number built on a different definition.
The spending is real, even where the resolution proof is thin
What is not ambiguous is that enterprises are buying. Five9 reported second quarter 2026 revenue of $312.4 million, with AI revenue up 78% year over year to roughly $39 million and an annualized run rate above $150 million, now about 15% of subscription revenue versus 9% a year earlier. The quarter included a roughly $100 million total contract value deal with a Fortune 100 financial services customer (Five9, 2026).
NICE reported AI and self-service annual recurring revenue of $362 million for the same quarter, up 52% year over year and now 15% of cloud revenue, plus its largest CXone and Cognigy contract to date: an eight-figure annual deal with the UK's HMRC, delivered through Capgemini (NICE, 2026).
Those figures describe contracts signed and revenue booked. They say nothing about whether the AI deployed under those contracts is resolving customer problems at the rate the sales conversation implied. Revenue growth measures adoption. It is not a proxy for outcome quality, and treating it as one is the most common mistake in this category.
What to check before trusting a resolution number
A defensible resolution figure should hold up against three questions.
First, who confirmed the resolution. A number based on the system closing the ticket is weaker than one based on the customer answering a follow-up question, or the issue not recurring within a defined window.
Second, what counts as in scope. A high resolution rate on password resets and order status lookups says little about billing disputes or account changes. Ask for the rate segmented by contact type, not the blended average.
Third, over what time frame the number was measured, and whether it comes from a pilot cohort or the full deployed volume. A resolution rate from a 90-day pilot on a curated set of intents will not hold once the system meets the full range of real contact volume.
Where the industry is heading
Forrester's Customer Service Solutions Wave for the first quarter of 2026 named Salesforce, Microsoft, Pegasystems and ServiceNow as leaders. Zendesk and Freshworks were rated strong performers. The report frames the shift as a change in operating model, not a platform swap. AI is increasingly the first layer of contact, not an add-on to it (Forrester, 2026, as reported by CX Today).
That framing raises the stakes on definition. If AI handles first contact by default, the resolution definition used for that layer decides whether the rest of the operation's numbers mean anything.
What commonly goes wrong
Buyers compare resolution rates across vendors without confirming the two are measuring the same thing. A 65% figure with a strict, customer-confirmed definition is a stronger result than an 80% figure that counts anything that did not reach a human. Ask for the definition before the number. If a vendor cannot state precisely what counts as resolved, the figure is a marketing number, not an operating metric.
Sources
Gartner. "Gartner Predicts Agentic AI Will Autonomously Resolve 80% of Common Customer Service Issues Without Human Intervention by 2029." 2025. https://www.gartner.com/en/newsroom/press-releases/2025-03-05-gartner-predicts-agentic-ai-will-autonomously-resolve-80-percent-of-common-customer-service-issues-without-human-intervention-by-20290
Five9, Inc. Second Quarter 2026 Earnings Release (Form 8-K exhibit). 2026. https://www.sec.gov/Archives/edgar/data/0001288847/000128884726000121/a063026ex991earningsrelease.htm
NICE Ltd. Second Quarter 2026 Earnings Release (Form 6-K exhibit). 2026. https://www.sec.gov/Archives/edgar/data/0001003935/000117891326003846/exhibit_99-1.htm
NICE (NICE) Q2 2026 Earnings Call Transcript. The Motley Fool. 2026. https://www.fool.com/earnings/call-transcripts/2026/08/12/nice-nice-q2-2026-earnings-call-transcript/
CX Today. "The Forrester Wave Says AI Will Run Customer Service, CX Leaders Need a New Operating Model." 2026 (covering Forrester Wave: Customer Service Solutions, Q1 2026). https://www.cxtoday.com/ai-automation-in-cx/the-forrester-wave-says-ai-will-run-customer-service-cx-leaders-need-a-new-operating-model/