What are your metrics hiding from you? Justin Robbins on the importance of CX observability

Your CX dashboard looks healthy. But if someone in your next leadership meeting asks whether your AI agents are giving customers accurate answers, could you show them?

And when customers keep calling about the same product issue, can you give the product team compelling data to prioritize a fix?

Those are the kinds of questions Metric Sherpa founder Justin Robbins and Parloa CMO Latané Conant tackle in episode 5 of The Agentic Wave. Drawing on their joint research, Justin and Latané discuss how better visibility can help CX leaders catch problems sooner and bring stronger recommendations to the business.

For leaders trying to get customer experience taken seriously, the conversation offers practical ways to spot what your metrics miss and make a stronger case for acting on customer feedback.

See what your metrics are missing

As companies deploy AI in customer service, leaders have more to keep track of: how the technology behaves, how it changes their operations, and what customers expect from it. Amid this “chaos,” as Justin calls it, are familiar metrics still telling them what they need to know?

The research found that 95% of CX operators expressed high confidence in what they were doing, while 85% acknowledged blind spots. Confidence in the measures you track doesn’t mean those measures capture everything you need to know.

Containment, for example, tells you whether an interaction stayed with the agent. It doesn’t establish whether the customer received an accurate answer.

Latané understands why familiar metrics persist: they’re trackable, understandable, and useful for spotting trends. The question is what else teams need to measure as AI takes on more customer interactions.

Measure what your business cares about

“The last thing you want to be is ignorable,” Justin says.

Latané agrees: “We want to set ourselves on fire and not be ignored!”

Fortunately, there are less combustible ways to get attention.

Customer conversations contain evidence of friction across the business. But when CX leaders present that evidence as isolated anecdotes, or without connecting it to business priorities, recommendations can be easy to dismiss.

Justin argues for making the work defensible: give colleagues evidence they can believe and get behind. Tell them why they should care.

That means showing what customers encounter, explaining why it matters, and identifying a decision the business can make. A recurring service issue becomes more compelling when you can explain its implications for repeat purchases, customer relationships, or the effort required to resolve a problem.

“I want the service team on the revenue team!” Latané says.

Better visibility gives CX leaders more substance to bring to that conversation. It can help them move from reporting what happened to influencing what happens next.

Defuse the AI ticking time bomb

Here’s one question worth taking into your next meeting: How would we know if our AI were giving customers inaccurate answers?

Among surveyed organizations running AI in customer service, 78.5% reported having no way of monitoring factual accuracy.

An AI agent can deliver an incorrect answer while the interaction appears successful. The conversation ends, containment looks good, and the error goes unnoticed.

Latané calls the finding a “ticking time bomb.” Justin points out that although people make mistakes too, AI changes the scale and pace at which errors can spread.

CX observability means monitoring and analyzing customer interactions across human and AI agents to understand the experience, identify risks, and determine where improvements are needed.

Justin likens it to a carbon monoxide detector: it reveals problems you might otherwise miss, early enough to act. For CX leaders, that means checking whether an interaction marked as resolved also gave the customer an accurate answer and solved their problem.

Make the metric lead somewhere

Which measures help your team decide what to improve? Which generate discussion without changing a decision?

Justin recommends starting with a clear view of current interactions, testing where accuracy and visibility are missing, and identifying who will help address the problem. Someone needs both accountability and the authority to act.

Watch episode 5 of The Agentic Wave for the full conversation, then read the 2026 CX Observability Study for findings you can bring into your next discussion about CX performance and priorities.