I spent an evening last month with a small group of CX and contact center leaders at a CCAIA (Contact Center AI Association) chapter meeting in Seattle. The topic was "AI readiness in the contact center." I walked away from the event with a new definition of "readiness" and what it takes to earn it.
Here are five takeaways that stuck with me, and why I think they matter to anyone running CX operations right now.
1. Readiness isn't a company-wide state
Going in, I expected people to describe their organizations as somewhere on a spectrum — early, experimenting, scaling. What I didn't expect was how often that spectrum existed inside a single company at the same time. One team would be confidently scaling one use case while, a floor away, another team on the same org chart hadn't gotten a pilot off the ground.
Rather than asking "is my company ready for AI,” the better question is "is this specific workflow ready?” The answer changes use case by use case, and treating readiness as an org-wide binary is how companies end up either overreaching in the wrong place or sitting on their hands in the right one.
The data backs this up at scale. A 2026 Publicis Sapient survey of over 1,500 AI decision-makers found AI now touches the majority of enterprise work, yet only about one in five organizations describe it as truly integrated into how they operate.
2. The mandate problem is the thing nobody wants to say out loud
The conversation that generated the most energy that night was about pressure — leadership handing down a directive to "adopt AI" without ever defining the problem it's meant to solve, the scope it should cover, or what success looks like. By the time that mandate reaches the people actually doing the work, it's been reinterpreted a half-dozen ways.
I don't think this is a failure of any particular company. I think it's a natural consequence of how fast the AI conversation moved at the executive level relative to how slowly any organization can actually operationalize a new capability.
According to Gartner, 91% of customer service leaders report being under direct executive pressure to deploy AI, and in our own analysis of the CX market, nearly seven in 10 enterprises are still piloting or experimenting even as leadership expects AI to handle a majority of interactions by year's end. That gap between what leadership expects and what teams have actually built is, in my view, the single biggest theme in CX right now — bigger than any specific technology choice.
The one thing I'd push every leader to do before greenlighting a mandate is to force a translation exercise. Turn "use AI" into a named problem, a specific target, and an agreed definition of success before a single vendor conversation happens. It sounds obvious. Almost nobody in that room said their organization had actually done it.
3. Trust the tools yourself before you hand them to a customer
One pattern that came up repeatedly, in different words from different people: the companies that had already put AI to work on their own internal operations were noticeably more confident and more capable when they turned that same technology toward customers. The ones who'd never used these tools internally were, understandably, struggling to build something a customer could rely on.
That tracks with something broader I've seen in workforce research this year: a lot of employees still can't apply the AI training they've received to their actual jobs. The tools get deployed. The fluency doesn't automatically follow. If your own team isn't comfortable with a capability, I don't think you're ready to put it in front of a customer yet, no matter how good the technology is on paper.
4. AI itself is almost never the blocker
If there was one moment that reframed the whole evening for me, it was hearing, repeatedly, that the actual AI part of CX projects tends to be the easy part. What actually stalls projects is everything around the model: systems that don't talk to each other, data that lives in silos, and approvals that surface a month after a project has already kicked off.
This isn't unique to the companies in that room. Deloitte's 2026 Tech Trends research found that legacy system integration — not budget, not skills — is the primary barrier AI leaders cite for deploying agentic AI. IBM has reported that roughly half of executives say integration challenges with legacy infrastructure have directly hindered their AI initiatives. I've come to think of this as the real "readiness" question — can an organization’s underlying systems and internal alignment actually support what that vendor promises?
5. Governance can't be an afterthought
The conversation around risk, compliance, and quality assurance for AI-driven interactions was a very sobering part of the evening. Several people described governance functions getting pulled in far too late, after a pilot was already built or when it's expensive and slow to retrofit real oversight, for example. The organizations that seemed furthest along were the ones that brought legal, risk, and compliance in at the very start of a project and gave them real ownership of the outcome, not a rubber-stamp role at the end.
There was also a striking consensus on a specific question: does anyone actually want AI agents that improve themselves without a human checking the work? The room's answer, without much hesitation, was no. I left that conversation more convinced than before that control, not autonomy, is what buyers actually want right now, whatever the marketing around "self-improving agents" might suggest.
A few practical habits worth stealing
Beyond the bigger themes, a handful of concrete habits came up that I think are worth adopting regardless of where your organization sits on the readiness spectrum:
Define what's explicitly out of scope before you talk to a partner/vendor. Knowing what you “won't” automate is often more clarifying than knowing what you will.
Don't lock into long-term contracts in a market moving this fast. Pilot, prove it, and keep an exit available — advice that, notably, came from vendor-side people in the room, not just buyers.
Not every workflow deserves generative AI. Reaching for the most sophisticated tool available isn't the same as reaching for the right one.
Benchmark external performance numbers against your own mixed reality.
Nobody in that room was arguing against AI. Every person there was working through how to make it succeed inside an organization that wasn't built for it yet — and being unusually candid about how far there still is to go.
