Customer Service AI Has a Resolution Problem
By John McMullan, Director of AI Agent Marketing, Observe.ai
Companies have invested heavily in AI for customer service, but new research shows customers are taking their problems somewhere else. A 2026 Gartner survey of 3,566 customers found that people were about three times more likely to use third-party generative AI (GenAI) than a company-provided AI tool when trying to resolve a service issue. Service leaders devoted a median of 12% of their 2025 budgets to AI, yet only 24% reported positive financial returns.
The findings point to a disconnect that contact center leaders cannot solve simply by putting a better self-service bot on the website. Customers are clearly willing to use AI when they need help. The harder question is, “Why are they often more willing to bring a problem to ChatGPT or another outside tool than to the AI experience offered by the company they are trying to reach?”
Part of the answer may reside in what customers expect AI to accomplish. For years, self-service was primarily designed to help people find information without speaking to an agent. GenAI has made those interactions more conversational, but a better conversation does not necessarily bring the customer any closer to resolving their problem.
Self-Service Has to Go Beyond Answering Questions
Customer service requests typically begin with a question, but the answer is only one step in a larger process. A banking customer disputing a transaction needs their identity verified, their card secured, and a claim opened. A patient calling about a prescription refill may need eligibility checked, a request routed to their provider, and a status update.
Traditional self-service technology can handle the informational side of these interactions reasonably well, but the experience becomes more complicated when resolving the issue requires the company to act. Contact centers look at measures such as deflection and containment to gauge whether self-service is working, yet those metrics mainly show whether a live agent was involved. They do not always reveal whether the customer completed the task or simply left the interaction without getting what they needed.
A customer who gets an answer and completes the task is very different from one who leaves the service chat, searches the web, and asks ChatGPT what to try next. Both may appear to have been contained within self-service, even though the experiences had very different outcomes.
As customer expectations shift to action-oriented support, contact center leaders may need to put more weight on resolution.
The Real Work Takes Place Behind the Conversation
Improving resolution requires organizations to look beyond what happens in the chat window or during voice interactions. Even a routine request may require an agent to verify an identity, review account history, or make a change in another system before the issue can be closed.
AI that can recognize a request but cannot interact with those systems runs into the same limitation as earlier generations of self-service. It may be able to explain what needs to happen without completing the work. Connecting the conversation to the systems behind it enables AI to gather the necessary context and execute approved actions during the interaction.
Company-provided AI has an advantage that third-party tools cannot easily match because it can connect directly to customer accounts and the systems where service work happens. ChatGPT may help someone understand a policy or determine what to do next, but it generally cannot carry that request through to resolution inside the company’s systems. For businesses, the opportunity is to use those connections to resolve more issues within the service experience while still giving customers a clear path to a person when human judgment or support is needed.
Voice Is Where Resolution Gets Harder
The difference between answering a question and resolving an issue becomes harder to hide when a customer picks up the phone. People often call because a digital option did not get them where they needed to go or because the request requires more than a straightforward answer, and the real-time nature of the conversation makes every delay, transfer, and repeated question much more noticeable.
Automating those interactions requires AI to do more than understand what the caller is saying. The system may need information from the customer’s account before it can determine what should happen next, and completing the request can require action in another business system while the conversation is still underway. That is a much higher bar than providing a relevant response from a knowledge base.
Not every call needs to be automated from beginning to end, nor should that be the only measure of success. AI can handle work within its scope while recognizing when an issue requires a person, whether because the situation calls for additional judgment, authority, or a different kind of support. What matters is that the customer continues moving toward a resolution rather than reaching the limits of the technology and having nowhere useful to go next.
The Handoff Is Part of the Customer Experience
Moving a customer to a live agent should not mean sending them back to the beginning of the interaction. If someone has already explained the problem and provided the information needed to get started, asking for the same details again makes the automated portion of the experience feel like an extra step rather than a useful one.
A better handoff carries that context forward so the agent understands why the customer reached out, what has already happened, and where the automated interaction stopped. Instead of reconstructing the issue, the agent can continue the work from there and make the transition from self-service to human support feel more natural.
Resolution Is the Better Test
For contact center leaders, the real question is what AI is truly improving once it is in use. A smoother conversation is helpful, but it only matters if customers are getting closer to resolving the issue that brought them there in the first place.
That makes resolution a more practical way to judge where AI is delivering value and where it is still falling short. It also gives leaders a clearer basis for deciding which use cases are worth expanding, especially as pressure grows to show that AI investments are improving the customer experience and the operation behind it.
Getting the most from customer service AI will mean less focus on how much of the interaction can be automated and more on whether the technology helps customers reach an actual outcome.