Is your AI handoff making customers repeat themselves?

A customer opens your chat window and gives the bot their order number. They explain that the package never arrived and confirm their email address. The bot transfers them to a human agent, whose first question is: "Can I get your order number?"

Most customer service leaders are sure this doesn't happen on their watch. Their customers remember it differently, and Five9's 2026 survey data shows how wide the gap is.

Key takeaways

• 96% of customer service decision-makers say their organization preserves context during AI-to-human handoffs, according to Five9's 2026 Business Leaders CX Report.
• 83% of consumers in the same survey say they at least sometimes have to repeat themselves after a bot transfers them to a human agent, and 35% say it happens often or always.
• 28% of consumers say the handoff experience makes them unlikely to use AI channels for customer service in the future.
• To find out whether your own handoffs work, ask customers right after the conversation: a CSAT or CES survey, plus a yes/no question about whether they had to repeat anything.

What is an AI-to-human handoff?

An AI-to-human handoff is the moment a bot (an AI agent such as a chatbot or voice assistant) transfers a customer to a human agent. A good handoff passes along everything the customer already told the bot, so the agent can pick up mid-conversation instead of starting from hello.

How often do AI handoffs go wrong?

They go wrong more often than the people running them think. According to Five9's 2026 Business Leaders CX Report, 96% of customer service decision-makers say their organization preserves context effectively during AI-to-human handoffs. Yet 83% of consumers say they at least sometimes have to repeat themselves after a bot transfers them to a human agent, and 35% say it happens often or always.

The decision-makers' other answers contradict the confidence. In the same survey, 37% reported that customers have to repeat information during handoffs, and 37% said their agents can't see the earlier conversation with the bot.

About the data: Hanover Research surveyed 3,000 consumers and 600 customer service decision-makers (manager level or above) in the US, UK and Germany for Five9 in April 2026. Five9 sells contact-center software, including AI agents.

Why do leaders think their handoffs work when customers disagree?

From inside the company, a handoff looks fine. It's on the process diagram, and the dashboard logs each transfer as complete. Neither tells you what the customer went through.

Five9's chief technology officer, Niranjan Vijayaragavan, told CRM Buyer that leaders' confidence rests on the handoff being in place, with no proof behind it. He pointed to two places where context gets lost: customer service systems that don't share data cleanly, and AI-generated summaries that leave out details the agent needs.

Privacy filters cause their share of trouble too. In Five9's survey, 30% of decision-makers reported context lost to redaction, where sensitive details are stripped out before the agent sees the conversation. Say a customer types their account number for the bot, and a filter masks it before the transcript reaches the agent. The filter did its job, and the customer still has to read the number out a second time.

A completed-transfer metric can miss all of these problems.

What does a bad handoff cost?

A bad handoff frustrates customers in the moment, and it can put them off your AI channels afterward.

Only 12% of consumers in Five9's survey said they weren't frustrated at the point a bot switched them to a human agent, and a third said they were very or extremely frustrated. Asked how likely they were to use an AI channel for customer service in the future after that experience, 28% said they were unlikely.

That number looks worse next to the investment plans. More than half of the decision-makers Five9 surveyed plan to increase spending on chatbots and virtual agents over the next 12 months. A disappointed customer is less likely to try AI a second time, so the spending lands on a channel some customers have already written off.

How do you measure whether an AI handoff worked?

The most direct way to find out is to ask the customer as soon as the conversation ends. Your helpdesk reports show that a transfer happened and how long it took, but they won't tell you whether the customer had to start over.

Five9's report recommends checking whether context survives every handoff, and it names the repeat-information rate (how often customers have to give the same information twice) as the clearest sign that a handoff broke. You could estimate that rate by combing through transcripts. The customer already knows the answer, though, and a short survey collects it from far more conversations than anyone could read by hand.

Send the survey as soon as the conversation closes, while the details are fresh.

CSAT vs. CES: which should you use after a handoff?

Both surveys go to the customer, but they ask about different things: CSAT about how satisfied they were, CES about how hard it was. Whichever you choose, pair it with the repetition question below.

CSAT (customer satisfaction score)

CSAT asks how satisfied the customer was with an interaction, usually on a 1-5 scale. A typical question: "How satisfied were you with the help you received today?"

Because CSAT covers the whole interaction, a good outcome can lift the score. A customer who got their refund may rate the conversation well even after explaining everything twice.

CSAT scores also tend to be high, which makes a broken handoff easy to miss. In our 2026 CSAT & NPS benchmark report, which draws on 2.9 million CSAT responses from companies using Simplesat in 2025, nearly two-thirds of companies scored 90% or higher. When most scores sit up there, a few points of difference won't tell you which conversations went badly. A 92% tells you customers were satisfied, not whether they gave their account number once or twice.

If you already send CSAT after every conversation, keep it and add the repetition question rather than switching surveys.

For setup details, see how to survey customers automatically after every resolved ticket.

CES (customer effort score)

CES asks how easy it was to get the issue resolved. A widely used version, CES 2.0, asks customers how strongly they agree with a statement like "The company made it easy for me to handle my issue," on a scale of 1 to 7.

Go back to the customer with the refund. To get it, they read their account number to the bot and then again to the agent. They may say they're satisfied but still rate the experience as difficult. The repetition question or a follow-up comment can tell you why.

If you're adding a survey specifically to check your handoffs, use CES.

For setup details, see our guide to measuring customer effort after support interactions.

Add a direct question about repetition

For the most precise read, follow your CSAT or CES question with a yes/no question about repetition. If your helpdesk can flag conversations that included a bot-to-human transfer, send it only on those and ask about the bot specifically: "Did you have to repeat any information you had already given the bot?" The share of yes answers is your repeat-information rate among the customers who responded, and it's the metric Five9 recommends tracking.

If you can't separate those conversations, ask the general version ("Did you have to repeat any information you'd already given?") and read the result as repetition reported by respondents overall. It will also pick up repetition unrelated to the bot, such as a transfer between two agents.

Either way, don't compare your rate with Five9's 83%. That figure counts customers who say they've had to repeat themselves at least sometimes, while your rate counts responses about specific conversations.

What should you do with handoff survey results?

Treat a high or rising repeat-information rate like a bug report and send it to whoever can fix it. The causes covered above are setup problems, so start with the team that owns the bot and its connection to your helpdesk rather than with your agents.

Read the comments next to the scores. If several customers mention re-entering an order number, check that field first. If they mention re-explaining the whole problem, look at the summary the bot hands to the agent.

Track the rate over time, and check it again after every change to the bot or the integration. Before reading a move as progress, look at two things: how many responses the rate is based on, and what kinds of conversations they came from. With only a handful of responses, one or two answers swing the number. And if fewer of this month's conversations went through the bot, or they were simpler ones, the rate can fall without the handoff getting any better.

Your agents see the other half of every handoff, so ask them to flag conversations where they had to re-ask for something the customer had already given the bot.

Frequently asked questions

What makes a good AI-to-human handoff?

A good handoff gives the human agent everything the customer has already shared, including the conversation with the bot and any steps the bot took. Five9's report proposes a simple rule: customers should never have to repeat what they've already told the AI.

Why do customers have to repeat themselves after a chatbot transfer?

Common causes include systems that don't share data cleanly, AI summaries that leave out key details, and privacy redaction that removes information the agent needs. In Five9's 2026 survey, more than a third of customer service decision-makers said their agents can't see the earlier conversation with the bot.

What is a repeat-information rate?

A repeat-information rate measures how often customers have to give the same information twice after a bot transfers them to a human agent. Five9's 2026 Business Leaders CX Report calls it the clearest sign that a handoff has broken down. If your helpdesk can flag conversations that included a bot-to-human transfer, ask those customers, "Did you have to repeat any information you had already given the bot?" The share who answer yes is your rate among respondents. A general version of the question works too, but it will also pick up repetition unrelated to the bot.

Should I use CSAT or CES to measure AI handoffs?

Either works. If you already send CSAT after every conversation, keep it. If you're adding a survey specifically to check your handoffs, use CES, which targets the extra effort a broken handoff creates. Whichever you choose, add a yes/no question about repeated information, which tells you whether the handoff was the problem.

When should I send a survey after an AI handoff?

Send it as soon as the conversation closes, while the customer still remembers the details. If your helpdesk can flag conversations that included a bot-to-human transfer, send the repetition question on those, so the answers are about the handoff.

Simplesat helps support and CX teams collect and act on CSAT, NPS, CES, and 5-star ratings, embedded in the helpdesk tools they already use. Start with our guide to measuring customer effort after support interactions.