AI has become the loudest voice in customer experience, but it is also becoming the easiest excuse.
Need to cut costs? Add an agent. Need to improve service? Add an agent. Need to prove innovation to the board? Add another agent.
But the CX leaders and professionals that CX Today has spoken with this year have been far more cautious.
The technology can help, certainly. Yet the same themes keep resurfacing when projects move from demos to real customer journeys:
-
Who is watching?
-
Does the customer still have control?
-
What happens when AI is wrong?
-
Did the interaction actually solve anything?
For World CX Day, here are five lessons from the past few months that deserve just as much attention as the latest AI launch:
1. AI Governance is Now Part of the Customer Experience
Agentic AI is moving from answering questions to taking action. That makes governance an outright CX issue, not simply a security or compliance concern.
In discussing the topic, Heath Ramsey, GVP Product Management and AI Platform at ServiceNow, warned that “the gap is the governance and the visibility,” particularly in customer-facing use cases.
Thames Valley Police offers a more practical model. Its AI assistant, Bobbi, automates around 75% of eligible contacts but escalates the cases involving vulnerability or higher harm to human operators.
Tom Boyd, Digital Product Manager for Salesforce products across the force, described Bobbi as “the first digital member of our staff.”
While this may sound like a playful soundbite, it speaks to the organization’s AI agent management strategy. The team monitors, coaches, and updates the AI as it would a human employee.
The risk of not implementing sufficient AI oversight and guardrails was also discussed by Vangie Cleversey, Chief Customer Officer at Acquia, who said:
“The companies I worry about are the companies where agents run free.”
2. Deflection is Not the Same as Resolution
For years, contact centers have measured success through containment, average handle time, and reduced volume. AI has made those metrics easier to improve, but not necessarily more meaningful.
Antoine Nasr, Head of AI at Forethought, argued that “deflection is a broken metric,” because a deflected ticket says nothing about whether the customer’s issue was resolved.
That point was reinforced by Bluecrest’s 100% interaction analysis program. The healthcare provider expected online self-service for appointment rescheduling to reduce calls. It did not. Instead, analytics showed that customers were using the new option because it gave them more flexibility to reschedule.
The initiative still improved the journey. It just did not produce the metric Bluecrest initially expected.
As Russell White, Transformation Lead at Bluecrest, explained:
