As organizations expand their use of AI across customer service, efficiency has become one of the clearest ways to demonstrate its business value.
However, improvements in efficiency metrics do not necessarily show whether customers are reaching a resolution, as lower contact volumes can conceal repeat contacts and increased friction.
CX leaders should now be considering whether their AI measures capture the quality and outcome of the customer journey, alongside the operational savings.
Jon Lamont, VP of Client AI Automation at CallMiner, told CX Today that businesses can lose sight of what customers actually want when they focus too heavily on efficiency metrics:
“Customers contact you to solve a problem. Success should be measured by resolution, not efficiency alone. A shorter call and faster response have little value if the customer's need remains unmet.”
The Customer Cost of Efficiency Metrics
Measuring AI through efficiency metrics is key to understanding whether it decreased operational costs, sped up interactions, or reduced employee workload.
For automation performance alone, these metrics are ideal. However, when these measures serve as a proxy for the quality of the journey, they often fail to capture whether a customer achieved resolution, or did so happily.
For example, a customer interaction can be contained by an AI system without the underlying issue being resolved, while a shorter interaction can indicate that they have been moved elsewhere.
"We're seeing organizations celebrate high deflection and containment rates, but when you look deeper, first-contact resolution is often declining," Lamont noted.
"That's a sign customers are still needing help, they're just taking a longer route to get it."
Furthermore, this problem becomes harder to identify when organizations do not make full use of the customer data available to them.
In fact, CallMiner’s 2026 CX Landscape Report found that 68% of CX and contact center leaders do not always use CX data to its best advantage (up from 62% in 2025), leaving important signals outside AI performance assessments.
Evidence of repeated frustration or lack of quality can provide context for an efficiency metric but may be disconnected across systems. As Lamont explained:

