AI in CX is moving from innovation theater to operational test, but enterprises still overestimate what AI can deliver without the right data, governance, and operating model.
The immediate appeal is obvious. AI can cut costs, improve deflection, and reduce pressure on contact center teams. But leaders that frame AI only as an efficiency play risk missing the larger opportunity, and creating the wrong incentives from the start.
The current moment is a strategic fork for enterprises. One path treats AI as a short-term service automation tool. The other treats it as infrastructure for better resolution, stronger loyalty, and more durable customer relationships. Asked where executives get it wrong, Vinod Muthukrishnan, VP/GM of Webex Customer Experience put it simply:
“The strategic case for AI in CX isn’t a single outcome. It’s the compounding effect across all of them. But here’s what’s most consistently misunderstood: cost reduction is the entry point, not the destination.”
That distinction matters because cost reduction is easy to justify in a boardroom. Loyalty and growth are harder to model, even though they may create more lasting value.
Muthukrishnan argues that the best AI deployments do more than deflect demand. They anticipate customer needs, preserve context, and resolve issues with less friction. In that model, the contact center becomes more than a support function. It becomes part of the growth engine.
That is also where the gap between hype and reality starts to show. Many companies can automate a simple interaction. Fewer can use AI to improve the quality of the experience without making the journey feel colder, harder, or more fragmented.
Model Commoditization Will Shift The Real Competitive Battle
As foundation models become easier to access, the question for enterprise buyers is changing. The issue is no longer just which model performs best in a demo. It is where durable CX advantage actually lives. Muthukrishnan’s answer is clear.
“Model commoditization is real. The durable moats aren’t in the models. They’re in the layers above and below them.”
He points first to proprietary customer signals. An enterprise that can detect intent, maintain continuity, and learn from millions of interactions builds an advantage that is difficult to copy.
He also emphasizes workflow and platform IP. Coordinating backend systems, AI agents, and human agents in real time is a systems challenge, not a prompt design exercise.
That view should resonate with CX leaders trying to separate substance from market noise. Model quality matters, but operational design matters more once AI enters live enterprise environments. If the surrounding systems cannot move context, trigger the right action, or govern the experience safely, the intelligence layer loses value quickly.
Trust and security also remain central. Muthukrishnan sees them as procurement differentiators, but also as architectural commitments. That is an important signal for enterprise buyers, especially as AI risk becomes more visible in regulated and high-volume service environments.
The Federated Model Works, But Only With Real Governance
AI in CX also forces a more practical question: who should own it?
Muthukrishnan does not support a fully centralized AI team as the long-term answer. He says those models can create bottlenecks. He is also skeptical of leaving AI entirely to CX teams, because that can fragment data and duplicate investment.
An IT-only model, meanwhile, may produce technically sound systems that miss the human and journey-level realities of service. From an execution standpoint, Muthukrishnan outlined the goal:
“The federated ‘platform plus business owners’ model wins at scale, but only when the platform is genuinely enterprise-grade and governance is real.”
In his view, IT should own infrastructure and governance. CX business owners should configure and optimize AI agents around specific journeys. Supervisors should manage a blended workforce of human and AI agents in real time.
That structure is compelling because it balances control with operational proximity. But his warning is just as important as his recommendation. Without shared standards and clear ownership boundaries, federated execution quickly becomes fragmented execution.
That is one of the central operational realities in AI today. Many enterprises are moving quickly, but speed alone does not create coherence. AI may appear across channels fast, while the customer experience behind it becomes less consistent.
Data Readiness Is Still The Hardest Truth In Enterprise AI
Muthukrishnan’s comments on data readiness may be the most useful part of the discussion for boards and operators alike. He says enterprises often confuse data investment with data readiness, and that mistake becomes obvious only when AI is deployed in live interactions.
His rubric centers on four non-negotiables: real-time accessibility, identity coherence across channels, context preservation, and operational governance. In an assessment, he warned:
“If your systems can’t surface the right information in the flow of a live interaction, you don’t yet have an AI-ready data foundation, regardless of how much data exists.”
That is a sharper standard than many organizations apply. A data lake or warehouse may look impressive on an architecture diagram, but that does not mean the right system can access the right customer information at the right moment, with the right permissions in place.
His point on identity coherence is equally important. Many enterprises can identify a customer within one system. Fewer can maintain continuity when that customer moves from voice to chat to email, or from digital service to a branch or store.
He also draws a useful line between data and context. Data is stored information. Context is the ability to carry meaning forward into the next moment. In practice, that difference often decides whether an AI interaction feels intelligent or exhausting.
Governance rounds out the rubric. Muthukrishnan argues that consent and permissions must work operationally, not just sit inside a documented policy. That is a meaningful point for enterprise CX teams that are now under pressure to scale automation without losing trust.
Why Resolution Quality Matters More Than Deflection
Measurement is another area where Muthukrishnan pushes back on industry habits. He says average handle time and deflection rates can distort behavior if they become the main story executives tell themselves about AI success.
Those metrics still have value. But they can reward the appearance of efficiency while masking poor customer outcomes. Asked what leaders should elevate instead, Muthukrishnan emphasized:
“If I had to pick one metric to keep teams honest, it’s resolution quality at first contact.”
He also highlights Customer Effort Score, retention signals, revenue expansion indicators, and AI containment quality. The emphasis is not just whether AI contained the interaction, but whether it actually solved the customer’s problem.
That distinction matters. Deflection can be gamed. Resolution quality is harder to fake because it is tied more directly to customer reality.
This is where enterprises can either build trust or quietly erode it. Teams that optimize for deflection may create faster systems in the short term. Teams that optimize for resolution quality are more likely to create experiences customers actually value, and return to.
The CX Market Still Needs More Operational Honesty
What stands out in Muthukrishnan’s perspective is that it is neither anti-AI nor blindly enthusiastic. It is operational. That makes it more credible.
His argument is not that AI in CX lacks promise. It is that the promise gets exaggerated when enterprises overlook architecture, governance, context, and accountability.
That is an important reminder for CX leaders, because the next phase of enterprise AI will be won by the teams that can turn intelligence into dependable execution across real customer journeys.
For enterprise buyers, that means asking harder questions. Can the system preserve context? Can it recognize the same customer across channels? Can it resolve the issue without increasing effort? Can it do so within governance guardrails that hold up under scale?
Those are less glamorous questions than the market often prefers. But they are the ones that separate tech hype from operational reality, and they are the ones more CX leaders now need to ask.
Join the conversation: Join our LinkedIn community (40,000+ members): https://www.linkedin.com/groups/1951190/
Get the weekly rundown: Subscribe to our newsletter: http://cxtoday.com/sign-up
