The claim versus the survey
Contact centre AI vendors have spent the last two years selling the same promise: agent-assist tools take the busywork off human agents so they can focus on harder conversations. A survey published by Verint in April 2026 puts a number on how much of that promise has actually reached the agent's desk. It is not flattering.
Verint surveyed 1,000 contact centre agents at companies running at least 300 agents, across five industries, fielded between November 18 and December 9, 2025. The headline finding: 31% of agents say they are likely to leave their job within six months. Among agents aged 18 to 34, that figure rises to 46%. Among agents 45 and older, it is 8%.
What is actually eating agents' time
The survey did not just ask agents if they were unhappy. It asked what specific tasks were consuming their calls, and the answers describe exactly the work AI agent-assist tools are marketed to remove.
According to Verint's data, 45% of calls require an agent to search for an answer mid-conversation. 54% require after-call work such as writing up a summary or updating a case record. 67% require the agent to complete a task on the customer's behalf inside a separate system. 57% require the agent to manually gather context when a case is escalated.
None of those four tasks are the empathetic, judgment-heavy conversation work that AI vendors say agents should be freed up for. They are lookup, documentation and system navigation, the exact category of task that agent-assist and after-call automation products claim to have already solved for the industry's biggest contact centres.
Why the gap exists
The discrepancy between the pitch and the survey data is not really about whether the technology works. It is about deployment. Reporting on the same Verint research, CX Today and CMSWire both describe organisations stuck in long-running AI pilots that never reach production, or that adopted AI narrowly as a headcount-reduction tool rather than building it into the agent's actual workflow. A pilot that automates ticket tagging in one queue does nothing for the agent still opening four systems to resolve an escalation.
That distinction matters for reading any vendor case study in this category. A tool can reduce average handle time in a controlled test group while agents across the rest of the floor still report spending three minutes on wrap-up work after every call, because the deployment never left the pilot.
The self-service side tells a similar story
The agent-side gap has a customer-facing mirror. A Gartner survey of 5,728 customers, conducted in December 2023 and still the most cited benchmark on the topic, found that only 14% of customer service issues are fully resolved through self-service channels, even though 73% of customers try self-service first. Even issues customers themselves describe as "very simple" resolve at only 36% in self-service. The pattern in both surveys is the same: automation absorbs the easy first step of an interaction and hands the harder remainder back to a human, who is then measured as if the whole interaction were automated.
What this means for buyers evaluating contact centre AI
Two questions separate a real deployment from a pilot that will show up in next year's version of this same survey.
First, does the tool operate inside the agent's existing workflow, or does it require the agent to switch to a separate assistant window. Verint's own breakdown of the busywork tax treats context gathering, search and after-call documentation as the specific tasks to measure, not because they are the hardest to automate but because they are the easiest to automate badly, in a way that adds a step rather than removing one.
Second, is the vendor's improvement number measured against the full agent population or a pilot cohort. A tool that cuts after-call work time by half in a 20-agent pilot group says little about whether the other 980 agents in a 1,000-seat centre are still doing that work by hand six months later.
Verint's survey does not settle whether contact centre AI works. It measures the distance between what the products promise and what agents report doing, at scale, right now. For a category where every vendor publishes its own efficiency numbers, a third-party attrition survey is one of the few data points that is hard for a vendor to spin.