Why Customers are Bailing on Your CX Bots

For years, customer service teams have treated the immediate request for a human agent as a design problem. When callers reject an automated experience before it has really begun, the natural instinct is to make the system sound warmer or more human. Rewrite the greeting. Acknowledge frustration. Tell callers they can speak naturally. The assumption is that if the experience feels less robotic, people will be more willing to engage with it.

We recently saw evidence that something else matters much more.

At a large insurance company using AI to handle questions about car rental benefits, a voice agent initially greeted callers by telling them they could “talk to me like a real person.” More than four in ten callers still asked for a human before having a substantive interaction with the system. We changed the greeting to something simpler and more direct, but the result was essentially unchanged. Forty-four percent of callers still immediately requested an agent.

What finally changed caller behavior had little to do with how the AI sounded. We changed what it could demonstrate that it knew. Instead of beginning with a generic greeting, the AI would make a reasonable guess at why the customer was calling, and provide information that would proactively address the issue – for example, it would describe the vehicle associated with a caller’s claim, provide the claim’s status, and check if the customer needed a rental.

Requests for human agents fell by 23%. Substantive engagement with the AI increased from 54% to 66%.

There is a lesson here that reaches well beyond the contact center. As companies race to make AI more conversational, personable and intuitive, they may be underinvesting in something more fundamental: giving AI enough context to prove that it understands the situation in front of it.

AI Is Inheriting a Trust Deficit

People don’t encounter a new AI system with a blank slate. They bring years of experience with phone trees that misunderstood simple requests, chatbots that served irrelevant answers and automated systems that made them repeat information they had already provided. Over time, users have learned that asking for a person can be the fastest route to getting something done.

That behavior can look like resistance to AI, but there is a more useful way to understand it. Before trusting an automated system with a task, people want evidence that the system understands enough about their particular situation to be useful. Researchers have described one element of this resistance as “uniqueness neglect,” the perception that an automated system cannot adequately account for an individual’s circumstances.

This becomes increasingly important as AI moves into more consequential areas of our lives and work. A customer calling after a car accident doesn’t see their problem as a generic insurance inquiry. An employee using an enterprise assistant expects it to understand their role, permissions and current work. A developer using an AI coding tool needs answers that account for the architecture surrounding the code.

In each case, an AI system has to demonstrate that it understands the circumstances surrounding the request before the user has much reason to trust its response.

Context Is Becoming Part of the AI Interface

Personalization has traditionally meant knowing someone’s name, remembering a preference or recommending something based on previous behavior. AI gives us an opportunity to make personalization far more situational. A useful system should understand what someone is trying to accomplish, what has already happened, which information is relevant at that moment and what actions are actually available.

That distinction helps explain why the insurance experiment worked. The AI didn’t simply greet callers more pleasantly. It removed work from the interaction. Someone calling to check whether rental benefits were available no longer needed to explain their claim, identify their vehicle and wait for the system to retrieve the relevant information. By surfacing that context immediately, the system demonstrated competence before asking the caller to trust it with anything else.

This principle applies across enterprise AI. Every time a user has to repeat information a system should already know, correct an assumption it should never have made or reconstruct context that already exists elsewhere in the organization, confidence erodes. For enterprises, that means context should be treated as part of the user experience, not simply as a technical concern behind the scenes.

AI May Change How We Measure Success

Our experiment surfaced another behavior that was easy to overlook. After the contextual greeting was introduced, some callers listened to the opening message and simply hung up.

Traditional contact center metrics could classify those calls as abandonment. But some of those customers had just received exactly what they appeared to need. If someone calls to find out whether a claim has been approved and the AI immediately tells them their claim is authorized and their rental benefits are ready, there may be nothing left to discuss.

Roughly 3% of calls using the dynamic greeting appeared to follow this pattern, with callers disconnecting after receiving the information and not calling back within two hours. It’s a small percentage, but it raises a larger question: What happens when the metrics we’ve historically used to measure engagement stop accurately measuring value?

As AI becomes more capable, we may see more cases where shorter interactions are better interactions. If a system can anticipate a need and surface the right information proactively, conversation length, number of steps completed and even traditional abandonment rates may become less meaningful measures of success.

Every AI Interaction Is an Audition

As AI becomes embedded in more workflows, users will get better at deciding which systems deserve their time. The opening moments of an interaction carry significant weight because people are quickly evaluating whether the system understands their situation, whether its information appears trustworthy and whether continuing will be worth the effort.

The strongest AI experiences will be the ones that can demonstrate useful context early and accurately. In customer service, that might mean understanding the status of a claim before the caller explains it. In an enterprise assistant, it could mean understanding an employee’s role, permissions and current work. In software development, it could mean reasoning about the architecture surrounding a piece of code.

As companies invest in the next generation of AI experiences, making those systems more natural to interact with is worthwhile. But natural conversation can only take us so far. People will give AI a chance when it gives them a reason to believe it can help, and increasingly, that reason is context.