From deflection to outcome: How IT evolves their infrastructure to prove outcomes and ROI

For years, customer service and IT support lived by a comfortable metric: ticket deflection. If a user didn’t open a support ticket, the system was considered a success. But in the era of agentic AI, deflection has exposed itself as a vanity metric. A user closing a chat window out of sheer frustration is not a win, it’s a hidden operational liability.

At Zendesk, we believe true AI value is measured in outcomes: actual, complete solutions where a problem is solved end to end without human intervention. According to our recent study of nearly 900 global IT leaders, they believe performance metrics should reflect this shift, too. More than six in 10 agree that AI metrics should be tied to clear, auditable resolutions. 

But evolving from deflection to outcome requires a fundamental shift in IT infrastructure. Here’s a closer look at what it takes. 

Evolving the IT infrastructure behind service outcomes

The technical reality is simple: AI cannot resolve an issue if it cannot access the system where that issue lives. An AI agent can only process a refund, change a shipping address, or provision an internal server if IT has built secure, real-time pipelines into the underlying CRM, ERP, and identity management systems. Without these connections, AI is just a glorified search engine.

That distinction separates deflection from outcome. Deflection happens at the surface: AI answers a question, recommends an article, or points a user toward the next step. outcomes happen deeper in the technology stack. The AI must understand the user’s intent, authenticate their identity, retrieve the right data, determine the appropriate action, execute it across one or more systems, and confirm that the desired outcome actually occurred.

Consider a customer asking an AI agent for a refund. A deflection-first system might explain the refund policy or direct them to a form. An outcome-first system can verify the order, determine whether it meets the company’s refund criteria, process the transaction, update the customer record, and confirm the refund—all within the same interaction.

Here’s how the infrastructure changes when outcome becomes the goal:

This shift allows IT to move beyond deploying AI as a standalone interface and start treating it as part of the enterprise architecture. APIs and integrations give agents the ability to take action across systems. Reliable, accessible data gives them the context to make the right decisions. Identity and permissions determine which actions they’re authorized to take. And governance, monitoring, and audit trails give IT visibility into what the AI did, why it did it, and whether the issue was actually resolved.

The future of service outcomes is verified

Moving from deflection to outcome gives organizations something else they need: proof that their AI investments are delivering real value.

A closed chat window or avoided ticket says little about whether the underlying problem disappeared—or customers walked away satisfied. Connected systems, however, give IT the ability to verify the outcome itself, answering questions like: Was the refund issued? Was the account updated? Was access restored? 

Verified outcomes provide organizations much-needed clarity on what they’re paying for and whether customer issues are getting resolved. In turn, this creates a much stronger foundation for measuring AI performance and proving ROI.

The bottom line: When AI is restricted to a standalone layer, it remains a novelty. To unlock true enterprise value, IT must build the connected foundation that allows AI to step inside business workflows, take action, and finish the job—transparently, securely, and verifiably.