
A lot of business owners picture an AI agent as all or nothing: either it answers every customer, or it's not worth having. In practice, the best setups are a partnership. The AI handles the questions it's good at, around the clock, and passes the rest to a person at the right moment.
This post is for owners who already use an AI agent, or are thinking about one, and want to get that partnership right. You'll get a simple way to decide which conversations belong with your team, and the habits that make a handover feel smooth instead of awkward.
It's tempting to measure an agent by how few conversations reach a person. But a customer who is upset, confused or asking for something unusual doesn't want to argue with a chatbot. They want someone who can make a decision.
When an agent keeps trying past the point where it can help, three things happen:
A good handover isn't a failure of the AI. It's the AI doing its job: recognising its limits and getting the customer to the right place quickly.
Your agent is at its best with questions that have clear, factual answers in what you've taught it:
These are the conversations that pile up after hours and on busy days. Letting the AI answer them frees your team for the conversations that need a human touch.
Some conversations should reach your team every time, however good the AI is. A useful rule of thumb: if the answer needs judgement, an exception or empathy, it belongs with a person.
Write these down in the agent's instructions, in plain words. The Boundaries section is a good place for "never do" rules, and the Conversation rules section for when to escalate.
In iGentrix, a conversation can reach your team in several ways, and it's worth using more than one:
The mood signal is the one owners most often overlook. A customer can be polite in their words and still be losing patience, and catching that early is far better than reading about it in a low rating later.
A handover goes wrong when the customer is left wondering what's happening, or has to explain everything again. A few habits prevent both.
Tell the customer where they stand. The customer should know straight away that a person is coming, and roughly when. iGentrix sends a notice at each step: when the conversation is passed on, when someone joins, if it moves to someone else, and when it's handed back to the AI or resolved. You can edit each message so it sounds like your business.
Be honest about office hours. Nobody is available all the time, and customers understand that. Set your opening hours, holidays and special hours, and outside them the customer is told when your team is back while the AI keeps helping with what it can. If you don't have their email or phone number, they're asked for one so you can follow up.
Never make the customer repeat themselves. When your team picks up a conversation, an AI summary shows what the customer wants, what the AI already tried and the key details. Your team starts with context, not "How can I help you today?"
Reply quickly and consistently. The copilot can suggest replies and draft a response from your knowledge, and saved replies give your team ready-made answers they can edit before sending. Nothing goes out until a person presses Send.
Hand back when you're done. Once the tricky part is sorted, click Return to AI so the agent can carry on with the simple follow-up questions, or Resolve the conversation.
Handover rules aren't set-and-forget. Every few weeks, look at:
Each review makes the line between "AI handles it" and "a person handles it" a little sharper.
The goal isn't an agent that never needs help. It's one that knows when to ask for it. Let the AI answer the clear, factual questions, send judgement calls and unhappy customers to your team, and make every handover clear, quick and free of repetition.
If you'd like to try this with your own business, iGentrix includes human handoff on every plan, and you can set it up from the Human handoff page in your agent's menu.

