Contact center AI will be judged less by how many interactions machines handle and more by whether customer problems actually get resolved. That conclusion came from Bob Laliberte of theCUBE Research and Zeus Kerravala of ZK Research during The AI ROI in Contact Center Summit, an exclusive broadcast on theCUBE, SiliconANGLE Media's livestreaming studio. The analysts reviewed conversations with Cisco Systems, Talkdesk, Zoom Communications and Five9 and found a shared destination: end-to-end resolution supported by connected data, governance and measurable business outcomes. For companies buying contact center AI, the unit of value is shifting away from automation volume.
What the analysts concluded at the summit
Kerravala framed the change directly: resolution and resolution quality is the new unit of value, and agentic systems should be judged on whether the customer's needs were completed across the full journey. Traditional measures such as containment and deflection no longer carry the weight they once did. Instead, the analysts said organizations need broader scorecards that account for customer satisfaction, effort, employee productivity, cost and growth. The summit gathered four platform vendors with different architectures and deployment strategies, yet the conversations converged on the same requirement, which suggests the shift is not tied to one product line.
Behind that shift sits context and integrated data. Fragmented systems and stale knowledge undermine AI accuracy, produce repetitive interactions and accelerate flawed processes instead of fixing them. Kerravala put it plainly: if a process is broken, AI will reach the bad destination faster. The recommended starting point is a clearly bounded, high-value problem rather than a redesign of the entire customer journey at once. Companies can establish baseline metrics, deploy AI against one specific workflow and measure the improvement before expanding further. That sequencing keeps early failures small and makes results comparable over time.
Initial deployments still need architectures that connect systems and reuse governance controls as they grow. Governance itself has to move beyond a preproduction checkpoint and become an ongoing operating practice: continuous evaluation, observability, policy enforcement and testing. According to Kerravala, proper governance lets an AI initiative move faster rather than slower, because it enables adoption instead of holding it back. The same logic reshapes how work is divided between people and digital agents. Human employees are expected to spend more time on exceptions, emotionally sensitive situations and interactions that require judgment, while AI absorbs standardized processes. Supervisors then have to manage a blended workforce and understand when AI is working, when it is failing and how work should pass between automated systems and people.
What this means for companies deploying AI
For a business that already runs or plans a contact center AI program, the practical consequence is a change in what gets measured and funded. A vendor demo built around containment rates says little about whether a customer's issue was closed, so procurement questions should move toward resolution data and the systems that feed it. Small companies feel this differently from large ones: a smaller operator can pick one journey, wire the data and see results within a quarter, while a large enterprise with fragmented systems faces a longer integration effort before any scorecard becomes meaningful. In both cases, the budget case now rests on outcomes rather than on the number of automated interactions.
What remains open is how much of this is verifiable today. The analysts describe a direction, not a finished standard, and no universal scorecard for resolution quality exists yet. Buyers should ask vendors how resolution is defined and tracked, which systems the platform connects to, how stale knowledge is refreshed, and what governance controls are enforced after launch rather than only before it. It also matters who owns the blended workforce model, because moving work between agents and people requires rules that most organizations have not written. This news does not mean containment metrics are useless; it means they are no longer sufficient on their own.
The marker to watch is whether vendors begin publishing resolution-based results alongside containment figures in their customer references. If that happens over the next few quarters, the buying criteria for contact center AI will have genuinely changed, and budgets will follow the scorecards. Until then, the controlled approach the analysts recommend remains the safer path: one journey, documented workflow and data requirements, baseline metrics, and testing of both routine interactions and edge cases before scaling.
