Over 50 percent of customer experience leaders have prioritized investments in real-time agent-assist technology, according to a 2025 CMP Research study.
In doing so, they’re unlocking many powerful new use cases to improve agent experiences.
For instance, some applications can auto-draft responses across digital channels, so agents simply have to review, edit, and send them off to customers.
Additionally, when an agent is on the phone, the tech might proactively feed them relevant knowledge articles to accelerate a path to resolution.
Further use cases include real-time transcriptions, auto-summarizations, and automated coaching suggestions. Yet, these are just some of the many possibilities.
As contact center providers stretch the limits of generative and agentic AI innovation, many more will likely emerge.
However, while that’s an exciting possibility for many market evangelists, agent-assist deployments still, in many cases, underwhelm.
Indeed, some contact centers have uncovered the hidden downsides of the tech, derailing their deployments and infuriating agents, the very people the tool should "assist".
The Downsides of Contact Center Agent Assist
No one doubts the potential of agent-assist software. However, a 2025 paper from a select group of Chinese universities spotlighted how the tech is frustrating customer service reps.
The study, "Customer Service Representative’s Perception of the AI Assistant in an Organization’s Call Center", found shortcomings in even the simplest use cases.
Take real-time transcription, for instance. Researchers found that often the tech would freeze after transcribing the first ten minutes of a call. They also found frequent transcription errors when customers would switch between languages, accents, and dialects.
Moreover, reps reported the AI’s inability to interpret phone numbers and addresses as a common concern, resulting in frequent manual corrections.
The paper summarized that repeated errors like this have placed three new types of burden on contact center agents. These are:
- A Learning Burden – When agent-assist tech fails to perform tasks, like accurately transcribing phone numbers and addresses, agents have to learn workarounds, adding to their cognitive load.
- A Compliance Burden – Often, AI-generated outputs won’t align with internal policies, as most large language models aren’t domain-specific. As such, agents can spend quite some time making sure outputs are compliant.
- A Psychological Burden – Think of use cases like next-best action. If the suggested action is superfluous, that can irritate agents during busy, high-pressure moments. It can also distract them from listening closely to customers.
While this is just a single study, with a small sample size, others have hinted at the hidden downsides of agent-assist technology.
More Than Just One Study…
While few other studies have honed in on contact center agent-assist technology specifically, some have highlighted agent resistance to new tools.
For instance, in 2023, a paper by the research firm Gartner found that 45 percent of contact center agents avoid adopting new tech.
Contact centers can allay such issues with a comprehensive agent buy-in strategy. However, the underwhelming performance of agent-assist tooling is a significant factor.