March 24, 2026
Human Agents Are Trained, Evaluated on Speed, and Left to Guess the Rest
Every day, support agents handle hundreds of customer conversations that will never be seen by a quality reviewer. No one will check whether the agent communicated clearly, handled the issue effectively, or left the customer feeling valued.
A new study from AI coaching platform Solidroad finds that most customer support conversations are never quality reviewed. Surveying 500 full-time customer support agents in January 2026, the State of CX report reveals that 81% of agents say the majority of their conversations are never reviewed for quality, with 37.4% reporting that fewer than 10% of their interactions receive any QA attention at all.
When agents do receive quality feedback, 79% describe it as helpful, with 41.8% calling it very helpful. The problem is that the feedback QA programmes are producing barely reaches anyone.
The most trusted delivery method is also the one that scales the worst. More than half of agents said one-to-one coaching sessions were the most helpful way to receive QA feedback, outpacing written feedback, team reviews, and scorecards. Personalised coaching works because it gives agents the context to understand not just what to improve, but why it matters in their specific interactions. Yet that same reliance on manual, individual delivery is precisely what keeps quality programmes from expanding their reach.
The Wrong Metric in the Wrong Seat
QA Score is the most common metric in performance reviews, cited by 52% of agents. Average Handle Time (AHT) comes second at 47%. Most agents trust the metrics used to assess them, with 77% saying they are confident their primary performance metric accurately reflects their support quality.
Among agents for whom AHT is the most important review metric, more than half say they are not confident it accurately captures the quality of their work. AHT is operationally useful since it helps with forecasting, staffing, and cost management, yet agents know it measures speed, not outcomes. When it becomes the primary lens through which their work is evaluated, it sends a message that how quickly a conversation ends matters more than how well it went.
Pressure on handle time has long been associated with agents rushing customers, skipping thorough resolutions, and generating repeat contacts that cost more in the aggregate than the time saved per call. The Solidroad data suggests agents are acutely aware of this issue; they simply have little power to resolve it.
When Training Ends, the Hard Part Begins
Most onboarding happens by watching experienced agents work. Sixty-seven percent of respondents were trained by shadowing peers, ahead of written documentation (57%) and customer support simulations (54%). On its own terms, this approach works reasonably well — 83% of agents said they felt either very prepared or somewhat prepared when they first went live.
However, readiness on day one and consistent performance over time are different things. The hardest part of ramping up, cited by 53.3% of agents, was applying what they had learned to unpredictable real customer situations. Product knowledge ranked second (44.4%), followed by finding the right information quickly during live conversations (43.2%). These are not gaps that onboarding alone can close. They require ongoing feedback, practice, and the kind of targeted coaching that current QA models cannot reliably deliver at scale.
Contact centre agent wellbeing research has consistently shown that insufficient support, unclear expectations, and a lack of development opportunities are among the leading drivers of turnover in an industry that already struggles with high churn. The Solidroad data adds that agents are struggling because the systems designed to help them improve do not reach far enough or fast enough.
Most contact centres have invested in training, metrics and QA programmes, just not in making any of them work together.
