This article was originally published on  LinkedIn .

I have sat in enough AI strategy meetings now to notice a pattern. Someone presents a roadmap for what customers want from AI, and the assumptions behind it rarely started with a customer conversation. They started with a competitor benchmark, a vendor deck, or a leadership hunch about where the market is heading. The plan can be thoughtful and well resourced, and it can still be built on a guess dressed up as insight.

That gap between assumption and reality turns out to be measurable, and the numbers are more revealing than most internal debates about AI strategy tend to be.

What Leaders Think Customers Want

Metrigy’s  AI’s Role in Customer Experience 2026–27  study—a global survey of 759 companies—paired with its  Q2 2026 AI Consumer Experience Index  of 1,000 U.S. consumers, put a figure on exactly this gap.

IT, CX, and AI leaders estimated that 42.7% of their customers prefer interacting with an AI agent and that only 10.3% actively avoid one. Consumers themselves told a different story. Genuine preference for AI agents sits at 21.7%, while 32.7% say they avoid them where they can.

Robin Gareiss, CEO and Principal Analyst at Metrigy, has been direct about the consequence: organisations that believe the higher number will overinvest in customer-facing automation and underinvest in the AI that helps human agents perform better.

That is not a small miscalculation. It is close to double the real preference and roughly a third of the real avoidance—in opposite directions—and it shapes real budget decisions.

What Customers Actually Say When Asked

The second half of the picture is just as telling, because it is not really about AI adoption at all. It is about what customers are frustrated by.

ServiceNow’s  The CX Shift: A Study of Customer Expectations in the AI Era  surveyed 27,000 customers, 3,500 service representatives, and 3,900 executives globally. Half of customers identified a lack of empathy as their biggest frustration with service, while only 23% of executives ranked empathy as a priority.

That is the real gap hiding underneath the AI conversation. Customers are not primarily asking whether AI or a human answers the phone. They are asking whether the response—whoever or whatever delivers it—actually understands what they need.

Motive matters here too. A  2025 consumer study commissioned by Kinsta  found that 80.6% of consumers believe businesses implement AI mainly to save money rather than improve service, and 88.8% said companies should always keep the option of speaking to a human available.

This was also a running theme at the 9th Annual Africa CX Conference. Customers are not against the technology. They are alert to why it is being deployed, and they notice quickly when the answer feels like cost reduction wearing a service label.

Verint’s  State of Customer Experience 2026  report, based on 5,000 U.S. consumers, adds another data point worth considering. For the first time in five years of that research, a majority—51%—said businesses fall short when customers actually need help.

That finding arrives at a time when AI investment in service has never been higher, which tells its own story about where the effort is being aimed.

Why the Gap Exists—and Why It Is Closable

None of this suggests AI is the wrong direction.  Metrigy’s research on the state of AI in customer experience  indicates that companies deploying AI strategically—blending it thoughtfully with human expertise rather than treating it as a replacement—are seeing improvements across business and customer outcomes.

The technology works when it is built around what customers have actually said rather than what leadership assumed they would say.

The gap exists for an ordinary reason. It is far easier to build a roadmap from a vendor pitch or an internal hunch than to sit down and ask customers directly—and then genuinely act on an answer that might contradict the plan already in motion.

Asking is not complicated. It is just uncomfortable, because the honest answer sometimes means slowing down a rollout that already has momentum behind it.

The organisations closing this gap are the ones treating customer input as a required part of AI strategy, not as a validation exercise conducted once the direction has already been set. That is a solvable problem, and solving it tends to pay for itself—in trust as much as in the numbers above.

If you asked your customers directly what they want from AI in their experience with you, how different would the answer be from the plan currently on your roadmap?

At Cybit, this is one of the first questions we ask before any AI conversation goes further: not what the market assumes customers want, but what your customers have actually told you—and whether anyone has checked recently. That single step changes more roadmaps than any framework we could bring to the table.

If any of this resonates and you would like to talk through what it looks like inside your own organisation, I would genuinely enjoy the conversation. Feel free to reach out and let’s talk.