Key takeaways
- Customers don't hate AI; they hate being trapped by it. A visible way to reach a person, in every conversation, is the single most important handoff rule.
- Hand over on clear triggers, such as an explicit request, no confident answer, a loop, frustration, sensitive topics and valuable customers.
- Pass the whole conversation and the customer's details to the person, so nobody has to repeat themselves.
- Out of hours, be honest. Collect the details, create a ticket and give a real reply time, then keep it.
- Tidio is our top pick for small and mid-size businesses because Lyro's handoff rules, separate online and offline behaviour, and "Talk to a person" option are easy to set up without a developer.
A customer types "agent". The AI replies with a help article. They type "talk to a human". The AI apologises and suggests the same article. They type "HUMAN" and close the tab. Somewhere in a dashboard, that conversation counts as "handled by AI".
That moment, when the AI should step aside and doesn't, is where most AI support projects lose customers. The AI can answer most of the questions well and still damage your reputation through a handful of bad handoffs. With the holiday peak about ten weeks away, now is the time to fix it, before volume multiplies every weak spot.
Why handoffs go wrong
The failures follow a pattern:
- There's no visible exit. The option to reach a person is hidden, or only appears after several failed attempts.
- The AI keeps trying. It rephrases the same wrong answer instead of admitting it can't help.
- Context gets lost. The person who picks up asks for the order number the customer already gave.
- Nobody is there. The AI hands over at 11pm to an empty inbox, and the customer waits in silence.
- The wrong person picks up. A billing question lands with someone who can't issue refunds.
- The tone breaks. A cheerful bot hands an angry customer to a scripted apology.
Each of these is a design decision, so each can be fixed.
Step 1: Decide when the AI should hand over
Write down your triggers, and configure them in your tool. Most businesses need these:
| Trigger | Example | What should happen |
|---|---|---|
| Customer asks for a person | "Can I speak to someone?" | Hand over immediately, no questions asked |
| AI has no confident answer | A question your content doesn't cover | Say so plainly and offer a person |
| The conversation loops | Same question asked twice, or two unhelpful answers | Offer a person after the second attempt |
| Frustration | "This is useless", capital letters, repeated "??" | Apologise briefly and hand over |
| Sensitive topic | Complaints, refunds outside policy, legal threats, safety | Collect the basics, then hand over |
| Valuable customer | Large order, VIP or B2B account, high cart value | Hand over early, or route to a senior person |
| Restricted topic | Anything you've told the AI not to handle | Hand over with a short explanation |
When in doubt, hand over. A good handoff costs you a few minutes of someone's time. A bad loop can cost you the customer.
Step 2: Make the exit visible
The option to reach a person should be available in every conversation, from the first message. A button works best on website chat. On messaging channels, a clear line in the greeting ("Type 'person' at any time to reach our team") does the job. If your tool offers the handoff in the AI's reply text rather than with a button, test it with real customers and check they notice it.
Don't make customers earn the human by failing three times first. People who want a person straight away usually have a reason.
Step 3: Pass the full context
When the conversation reaches a person, they should see, without asking:
- The full transcript, including what the AI already suggested.
- Who the customer is: name, email, past conversations, and for stores, recent orders.
- Why it was handed over: the trigger, as a tag or note ("loop", "refund request", "asked for a person").
- Any details collected: order number, photos, account ID.
The first message from the person should prove they've read it: "Hi Marta, I can see your parcel shows as delivered but hasn't arrived. Let me check with the courier." Never start with "How can I help?"
Step 4: Be honest out of hours
A handoff to an empty inbox is worse than no handoff. Set separate rules for online and offline hours:
- Tell the customer the team is offline, and when it's back.
- Collect what the team will need: email, order number, a description of the problem.
- Create a ticket so it's first in the queue in the morning.
- Give a specific reply time ("by 10am tomorrow") and keep it.
Urgent issues, such as a security problem or a missed delivery for an event, may justify an on-call person. Decide in advance what counts as urgent.
Step 5: Route to the right person, fast
Use the handoff reason and the customer's details to route the conversation: billing to someone who can refund, technical issues to the person who knows the product, VIPs to your most experienced agent. Set a target for how quickly a person picks up after a handoff during business hours. Two minutes on chat is a reasonable goal, and it's the number customers will judge you on. Our guide to customer service SLAs covers how to set targets your team can keep.
Step 6: Write the handoff messages
The words matter as much as the logic. Rewrite every message your AI uses around a handoff:
| Moment | Weak | Better |
|---|---|---|
| AI can't answer | "I'm sorry, I didn't understand that." | "I don't have a reliable answer for that. Let me get someone from the team." |
| Handing over | "Transferring you to an agent." | "I've passed this to Jakub with everything you've told me. He'll reply here in a couple of minutes." |
| Out of hours | "Our team is offline." | "We're offline until 9am. Leave your email and we'll reply by 10am tomorrow, so you won't need to explain this again." |
| Customer is upset | "I understand your frustration." | "Sorry, this isn't the help you needed. I'm getting a person for you now." |
Step 7: Measure handoff quality
Add these to your monthly review:
- Handoff rate: the share of AI conversations that reach a person. Falling over time is good; zero is a warning sign.
- Time to a person after handoff, during and outside business hours.
- Satisfaction on handed-over conversations, compared with the rest.
- Repeated information: how often the person asks for something the customer already gave.
- Abandonment before handoff: customers who leave while the AI is still trying.
Read ten handoff transcripts a month. They'll show you more than any chart. When you calculate what your AI saves, count handoffs correctly too; our guide to measuring the ROI of AI in customer service explains how.
Common mistakes
- Optimising for fewer handoffs. A low handoff rate achieved by hiding the exit isn't a success.
- Asking the customer to start again in a new channel or form.
- Handing over to nobody outside business hours.
- Letting the AI argue with a customer who has asked for a person.
- Never reviewing handoff transcripts, so the same gaps repeat every week.
The best tools for AI-to-human handoff
We compared how much control each tool gives over when the AI hands over, how much context reaches the person, how it behaves out of hours, and how easy it is to set up for a small team.
Tidio gives Lyro default handoff rules for when a visitor asks for a person or Lyro can't answer, plus custom handoff audiences and instructions. You set separate behaviour for online and offline hours, so the same question can go to the live inbox during the day and become a ticket at night. When a customer chooses "Talk to a person", the whole conversation moves into the regular inbox for your team to pick up. Higher-volume brands can go a step further on the Premium plan, where Tidio's own services team sets up and keeps tuning Lyro, handoff rules included.
Why we like it
- Separate handoff rules for online and offline hours
- Full conversation passed to the inbox, or a ticket created instead
- Custom handoff audiences and instructions without code
- AI agent, live chat and inbox in one tool, so context never changes systems
Watch out for
- AI conversations are billed by volume, so model your busiest months
- Routing is simpler than Zendesk's for large multi-team operations
Intercom lets you write escalation guidance for Fin and set rules that trigger a handoff when specific customer or conversation data is present. Fin can offer a person when it detects frustration or a loop. In chat, Fin offers the handoff in its reply rather than with a button, so test that customers notice it.
Why we like it
- Escalation guidance plus data-based escalation rules
- Can offer a person when frustration or loops are detected
Watch out for
- Handoff offered in text rather than a button in chat
- Seat plus per-outcome pricing is expensive for small teams
Zendesk lets you configure when the "talk to a human" option appears, build escalation flows, and set up asynchronous ticket escalation for when nobody is available. Combined with its routing, it suits large teams with many queues and skills.
Why we like it
- Control over when the human option appears
- Asynchronous escalation to tickets out of hours
Watch out for
- The visual dialogue builder needs an advanced AI add-on
- Setup needs a dedicated admin
Help Scout's AI Answers works inside its Beacon widget, and you can show an escalation button so customers can chat with your team or send an email. Handoffs to a person aren't charged, and you can customise what the AI says when it can't help.
Why we like it
- Optional escalation button in the widget
- Not charged for handoffs to a person
Watch out for
- AI Answers is a paid add-on billed per resolution
- Chat and social channels are secondary to email
Run a handoff test this week
Before your busiest months, test your setup as a customer would. Try each of these and note what happens:
- Ask for a person in the first message.
- Ask the same question twice in different words.
- Type an angry message in capital letters.
- Ask for a refund outside your policy.
- Ask a question your content doesn't cover.
- Repeat all five outside business hours.
- Check what the person who picks up can see, and how long it took.
Fix whatever surprised you, then repeat the test after each major content or setting change. If your AI is in front of EU customers, also check that the first message says it's AI, as covered in our guide to AI chatbots, GDPR and the EU AI Act. For the bigger picture, see how to automate customer support.
Frequently asked questions
When should an AI chatbot hand over to a human?
When the customer asks for a person, when the AI doesn't have a confident answer, when the conversation is going in circles, when the customer is frustrated, and for sensitive topics such as complaints, refunds outside policy or legal issues. Many businesses also hand over valuable customers early.
Should the "talk to a person" option always be visible?
For most businesses, yes. Hiding it reduces handoffs in the short term but costs trust and repeat business. If you're worried about volume, improve the AI's answers rather than hiding the exit.
What should the AI say when no one is available?
Tell the customer honestly that the team is offline, collect their email and the details of the problem, create a ticket, and give a specific reply time, such as "by 10am tomorrow". Then make sure someone replies by then.
How do I know if my handoff process is working?
Track how long customers wait for a person after a handoff, satisfaction on handed-over conversations, how often customers have to repeat information, and how many leave before reaching anyone. Read a sample of handoff transcripts every month.