What to ask about human handoff before you buy a support chatbot

Handoff is where a chatbot's promises get tested

Every support chatbot will eventually hit a question it should not answer. What happens next decides whether your customer gets help or gets stuck. Demos tend to skip this moment, because it is the least flattering one, so it pays to ask about it directly before you buy.

The questions below are meant for a sales call or a trial. None of them need technical knowledge, and a vendor who answers them plainly is telling you something useful.

1. What counts as "cannot answer"?

Ask what actually triggers a handoff. Is it only when the customer types the word "human"? Or does the bot also step aside when the documentation does not cover the question, or when the same person has asked twice without getting anywhere? A bot that only hands off on request leaves quiet failures in the conversation, where you will not see them.

2. Where does the customer go?

"We hand off to your team" can mean very different things. It might be a live chat with a person, a form, a support email address, or a ticket in your helpdesk. Ask to see the exact screen a customer lands on, and ask who on your side receives it and how quickly they are likely to see it. A handoff to an inbox nobody watches is a dead end with extra steps.

3. What does the person receive?

Ask what arrives on your side when a handoff happens. The customer's question, the conversation so far, the page they were on? If nothing travels, your customer will have to repeat themselves, which is a common complaint about chatbots in the first place.

A good answer to this question is a concrete example, not a description. Ask for one.

4. Is a person actually available?

A handoff to a form that says "we will get back to you" is fine, as long as it says so. What matters is that the bot does not imply a person is on the line when nobody is. Ask how the bot words the handoff outside your working hours, and test it yourself at a time when your team is offline.

5. Can you change what it says and where it sends people?

Your support setup will change. Ask whether you can edit the handoff wording and the destination yourself, or whether every change is a request to the vendor. Also ask what happens if the destination stops working, such as a retired email address.

6. How do you find out the handoff is failing?

Ask what you would see if handoffs were going wrong. A way to read the conversations that ended in a handoff, and the ones that should have, is one way to tell whether the bot steps aside at the right moments. If the vendor has no answer, plan to read a sample of conversations yourself.

How we handle this

Here are our honest answers, including the weaker ones. When our agent cannot answer from your documentation, it is instructed to say so plainly and point the customer to the contact path configured for your deployment, such as your support inbox, instead of sending them back to search your help site (question 2). That is a pointer to a contact path, not a live transfer of the conversation to a person, so nothing is forwarded to your team on the customer's behalf: they write their question to you themselves (question 3). The wording and destination are set up with us rather than edited by you in a dashboard, so a change is a request to us (question 5). We do not give you a view of conversations to review, so the way to check the handoff is to ask it questions your documentation does not cover and see where it sends you (question 6). We test these behaviors rather than promise them outright.

Related reading

• See all 26 guides in the Actually Helpful library, grouped by what you are actually trying to figure out.
• Why "let me transfer you to a human" so often means starting over, on what gets lost between the bot and the person.
• What should happen when a chatbot can't answer, on the moment before the handoff.
• What a chatbot can't replace on a support team, on the work that stays with your people.