In a world of interactions, how do you know if your customers’ issues are actually getting resolved?
For years, customer service lived by a comfortable metric: ticket deflection. If a customer 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 customer closing a chat window out of sheer frustration is not a win, it’s a hidden operational liability.
Now, organizations want greater transparency when it comes to AI in CX, including what they’re paying for and whether customer issues are actually getting resolved.
A customer closing a chat window out of sheer frustration is not a win, it’s a hidden operational liability.
At the same time, customer trust in AI is at stake. Murky decision-making and clunky chatbot experiences won’t cut it. When an AI agent makes a mistake, customers want to know who is responsible and how it will be corrected. Therefore, actions have to be explainable: 92% of leaders say customers are more likely to trust AI responses when they include an explanation.
Verified outcomes—that is, resolutions confirmed by large language models (LLM)—can satisfy the needs of organizations and customers alike.
Here’s how:
Verified outcomes validate every resolution
Organizations need greater confidence that their AI investment is delivering meaningful value, but many still face uncertainty about what they are paying for. For starters, activity, usage, and deflection metrics do not always show whether the outcome was accurate, complete, and effective. We also recognize that it’s difficult to prove whether automation is truly resolving customer needs or simply deflecting, delaying, or shifting contacts across channels.
We’re not alone in this belief. Our research shows over 60% of service leaders believe AI projects should be judged by resolutions, not deflection. And nearly two thirds (64%) say senior leadership want AI metrics tied directly to resolved issues.
At Zendesk, we’re evolving how we measure success to make the quality and value of automation easier to understand. That’s where our verified outcomes shine, aligning AI investment to measurable, quality-assured outcomes.
Aligning investment to measurable outcomes is the new way forward. By connecting spend to verified outcomes and quality evaluation, we’re ensuring that your finance and procurement teams can assess not only how much automation occurs, but how effectively it resolves work.
Think of it this way: A customer of yours reaches out because their refund hasn’t arrived. With a deflection-first approach, your chatbot recognizes the word “refund” and directs them to an FAQ explaining that refunds typically take five to 10 business days. The outcome is marked as resolved—but is it? The customer still doesn’t know where their money is or when they’ll see it. Ultimately, the customer is left with unsavory choices: contact your support again, or get fed up and give their business to your competitor.
With a resolution-first approach, your AI agent securely authenticates the customer, finds their order in the CRM, and checks the payment and returns systems. It sees that the return was received but the refund failed to process. The AI agent automatically retries the transaction, and confirms when the customer can expect the money back. Instead of simply answering a question, the agent resolves the underlying issue. And in the case of LLM-verified outcomes, the entire workflow is tracked for auditing and ROI measurement.
Customers get proof, not just promises
For organizations, building customer trust in AI is paramount. In fact, 85% of leaders say AI transparency will become non-negotiable for customer-facing use cases.
Here’s why: With AI, customers are being asked to put their faith in a system they can’t always see, understand, or hold accountable. A lack of trust isn’t necessarily with AI itself, but whether a company’s AI understands them, took the right action, and can fix things when something goes wrong. Most importantly, customers want proof their problem has been solved.
We see this in the story above. A traditional chatbot saying, “I’ve processed your refund” requires your customer to trust the chatbot (hint: they won’t). However, an AI agent that executes the refund, verifies the transaction succeeded, and provides a confirmation number gives your customer evidence that the outcome occurred.
Set a new standard with LLM-verified outcomes
Your customers deserve proof that AI actually solved their problems. At Zendesk, we believe our customers do, too.
That’s why we’re going beyond automation, deflection, conversation, or usage metrics. Instead, we’re certifying AI outcomes with an LLM for quality and completeness. That way, you can understand when automated work is complete, accurate, and genuinely resolved—ensuring the value and quality of every customer outcome.
