AI Is Getting Better at Customer Service. Is It Getting Better at Resolution?
Closing a support ticket is not the same as resolving a customer’s problem. And unresolved problems come at a cost: repeat contacts, additional operational effort, customer frustration, and ultimately, revenue at risk. As enterprises adopt AI for customer support, that distinction will define whether the technology delivers meaningful business value.
Our CEO, Atreya Chaganty, discussed this recently with Sharang Sharma of Everest Group, who has tracked the CX market for over a decade. Everest’s research finds that around half of large enterprises now have AI live in agent assist and knowledge discovery. For end-to-end resolution, outreach and retention, that falls to 23%. And 63% identified data unification and real-time workflow orchestration as critical challenges to scaling AI-driven resolution.
Around half of large enterprises have AI live in agent assist and knowledge discovery. For end-to-end resolution, outreach and retention, adoption falls to 23%. Source: Everest Group enterprise survey.
“The constraint isn’t whether AI is smart enough. It’s whether it has the connectivity and the ability to act.”
Much of the industry’s investment has focused on AI-powered conversational tools that automate customer response across chat, email, and messaging channels. AI email automation has real value. It interprets requests, tailors responses and reduces the time spent on simple queries. But the metrics used to measure it, customers responded to, time to respond, chats closed, say little about whether the customer’s problem was solved.
The measure of success isn’t whether a response was sent. It’s whether the customer’s problem was resolved.
Let’s take a common example
Consider a customer asking, “Where’s my order?” Across the enterprises we support, it is about a third of all inbound requests, even today. Many are simple, but many are complex.
An accurate response explaining the five-day shipping policy may answer the question. It does not however establish whether the order is on track, and it certainly does not act if it isn’t. The same sentence sent on day one and on day six reads identically to a platform. It is not the same problem.
When those issues remain unresolved, customers contact support again, increasing cost to serve while the underlying delivery problem persists.
Resolution has three parts
Atreya breaks it into three: “intent and knowledge, access permissions and integration, and third, verification.”
Intent and knowledge is increasingly well served by today’s AI models, particularly for straightforward customer requests. But recognizing what a customer wants is only the starting point.
Permissions and integration is where most deployments stop, because it depends on the enterprise’s own systems rather than on the vendor’s model.
Verification is rarely discussed, and in our view it is the most necessary of the three. Once the action is taken, can we confirm it actually happened, and that the problem has not quietly recurred?
A polite AI response can mean nothing
A customer of one of the large brands we support wrote in to say a refund still had not reached their card. The platform handling that queue did everything it was designed to do. It read the request, identified the intent, retrieved the policy, which allowed seven to ten business days, apologized with appropriate tone, and produced a well-articulated response.
The customer had been waiting thirty days.
Two things had gone wrong. The platform missed that this was not a new problem: the same issue had been raised before, and answered by the same platform. Nobody had asked the question that mattered, which is why the refund had not been processed. Answering that means going into the enterprise’s systems, then confirming the money has moved.
The response was polite, but the problem remained unresolved. The customer was forced to follow up again, generating additional service costs while further eroding trust.
More importantly, the system failed to recognize and address the underlying cause. It automated the response, not the resolution.
These are not edge cases
The usual response is that cases like this are exceptions, to be handled later. We are not sure that is a good way to think about it.
“Normal scenarios do not cause dissatisfaction. They are easy to handle and the upside of handling them is not that high. But if you are able to handle these complicated scenarios, which you might consider edge cases for automation, that is the biggest source of dissatisfaction.”
These are precisely the issues where better resolution can deliver disproportionate value: reducing repeated effort, protecting customer relationships, and preventing avoidable costs.
Measure the verified outcome
Many organizations still measure AI success through operational metrics: closed cases, automation rates, and first contact resolution. These are useful indicators, but they don’t always tell us whether the customer’s underlying problem was solved.
A high automation rate means little if customers return with the same issue, driving up cost to serve and eroding the efficiency gains AI was intended to deliver.
The metric worth becoming fixated on is the verified outcome: did the action complete, and are we confident the problem is not recurring?
It is easy to tell a customer that shipping takes three days. It is harder to go back to the courier, secure a new date, confirm it, and then reach out on the third day to ask whether the package arrived. That last step is what turns an action into an outcome.
Three questions for enterprise leaders
What do we count as a resolution? A closed ticket is easy to measure. A resolved problem requires evidence. Are we tracking repeat contacts, reopened cases, and whether the intended customer outcome was actually achieved?
What behavior are we paying for? Where pricing is tied to “resolutions,” the definition matters. How are repeat contacts for the same unresolved issue counted and charged? Are we paying for completed interactions or verified outcomes?
Are we learning why customers need support? Faster responses do not explain why deliveries fail or why policies confuse people. Understanding the root causes of recurring issues, their operational and commercial impact, and where corrective action can deliver the greatest value is what drives lasting improvement.
Need for human judgement
As AI handles more predictable interactions, human expertise becomes increasingly valuable in diagnosing complex problems, exercising judgement, and determining how to prevent recurring failures.
The objective isn’t simply to route exceptions to people. It’s to combine AI’s ability to identify patterns at scale with human expertise that can interpret those patterns, validate root causes, and drive meaningful improvements.
This is where QiQ brings a different kind of intelligence to customer experience. Rather than replacing AI automation platforms, QiQ complements them by identifying the root causes of recurring customer issues, quantifying their operational and commercial impact, and helping teams prioritize corrective action.
By connecting customer intelligence with operational expertise and measurable outcomes, QiQ helps organizations move beyond resolving individual interactions toward preventing the problems that generate them.
The future of customer experience
Automation and root-cause improvement belong together. AI can execute known solutions at scale. Combined with human expertise, it can also help uncover why problems occur, identify patterns across customer interactions, and inform better solutions.
The opportunity is to connect these capabilities: use intelligence to understand recurring failures, develop and validate corrective actions, and turn successful interventions into repeatable processes.
Over time, organizations can do more than expand what they automate. They can reduce the volume of problems that require support in the first place.
Sharang landed the session on the right question.
“If we’re only measuring whether the conversation was automated, we’re measuring the wrong thing. The question isn’t did AI handle it. It’s did the customer actually get what they needed.”
The ambition should be to resolve problems effectively today, while eliminating the root causes that bring customers back tomorrow.
See what your conversations say about resolution
A short walkthrough of QiQ on your own customer conversations, and the recurring problems sitting inside them.
Book a demo
