AI has become a standard part of customer service. As Five9’s 2026 Business Leaders Customer Experience Report found, 92 percent of organizations have implemented or piloted AI use cases in CX. But adoption alone does not guarantee a better experience.
83 percent of consumers still have to repeat themselves when moving between AI and a human agent. That's despite nearly all CX decision-makers believing their organization preserves context during the handoff.
That disconnect exposes one of the biggest challenges facing AI-powered CX: businesses are measuring whether the technology works, while customers are judging whether the experience works.
The tension is not simply that organizations are unaware of the issue. Around one third of decision-makers also acknowledge problems including customer repetition, limited agent visibility into earlier AI interactions and transfer glitches. The gap is in how success is being defined.
Sam Counterman, SVP of GTM Technology & AI Operations at Five9, told CX Today the industry needs a clearer definition of what a successful handoff actually is.
“When we move that person or that customer from A to B, whether it’s from AI to a human, that doesn’t mean that we’ve solved the problem.”
A handoff is one part of a wider customer journey, and if the transition introduces delay, loss of context, poor routing or irrelevant conversation, it can undermine the relationship with the brand.
Here are three reasons handoffs still fail, and what leaders should look for instead.
1. Organizations Measure the Transfer, Not the Outcome
Many CX teams can report whether an interaction was transferred from a virtual agent to a human, and they may also track the duration of a handoff, the volume of escalations or the percentage of contacts resolved through self-service.
Although these are useful operational measures, they do not show whether the customer’s problem was resolved.
The problem is that many organizations are grading themselves on the wrong assignment. For the customer, context is preserved only when the next person knows who they are, why they contacted the business, what they have already tried, and what should happen next.
An AI system can identify that it cannot complete a task and successfully transfer the customer to a live agent, and on an internal dashboard, the process may appear to have worked. But if the customer has to repeat their identity, explain the problem again or is routed to an agent without the required skills, the experience has failed.
“It’s more of a scoring gap, not necessarily a knowledge gap,” Counterman said. “The knowledge is there. It’s just how and when you want to use it in the right moment based on the nature of the small task or a big escalation.”
In a restaurant, being greeted at the door, shown to a table and served by different people all involve forms of handoff, where the customer does not assess each interaction separately. They assess the experience as a whole.
The same applies in customer service. A customer may start in a web chat, move to a virtual agent, transfer to voice and finally speak to a specialist. While the business may record several successful transfers, the customer may experience a fragmented journey.
This is why leaders need to move beyond channel-level measures. They should assess whether the customer reached an appropriate resolution, how much effort was required, whether the customer made repeat contact and how the interaction affected their likelihood to stay with the brand, Counterman said.
“It’s worse if you have a bad experience through handoff in a non-contextual sense with an existing brand, because that’s damaging in some ways.”
Counterman shared a recent example from a grocery order change. He started in an AI chat flow, was moved between multiple points of contact and had to repeat the same information five times.
“I managed to resolve that with the brand, but am I going to shop with them again? Probably not.”
Resolution alone did not repair the experience, as repeated explanation created friction and reduced confidence in the brand.
What does fixing it look like? Track the transfer, but treat it as the start of measurement rather than the end.
Combine transfer data with three outcome measures: repeat-contact rate in the following 24 to 48 hours, customer effort measured at the transfer point itself, and first-contact resolution after the handoff.
None of these measures is perfect in isolation. Repeat contact can have other causes, effort scores are subjective, and resolution can be defined too loosely. Together, they show whether the handoff made the customer’s situation better or worse.
Advisors should also be able to quality-score the AI handoff. They can assess whether the AI captured the correct intent, passed on usable context, escalated at the right time, and enabled them to resolve the issue without starting cold.
A handoff for a simple order amendment should not be assessed in the same way as one involving a vulnerable customer, a disputed payment or a complex healthcare query.
