Every team that puts AI agents on inbound leads has to answer the same question: when does a person take over? If the answer lives in people's heads, the agent will find the gaps. If it is written down, reviewed and changeable, it becomes the control that makes the rest of the system trustworthy.
This guide gives a simple structure for writing those rules, and the mistakes that make them fail.
What a hand-off rule is
A hand-off rule is a sentence with a trigger and a consequence: when this happens, a person takes over. It is written in the language your sales team already uses, not in technical terms, and it is owned by the people who answer to the customer, not by whoever configures the software.
The five parts of a rule
A rule is complete when it answers five questions. If any of them is missing, someone will have to guess at the worst moment.
- Trigger. What the agent notices: a topic, a score above a threshold, a request for a person, a type of account.
- Owner. Who takes over, by role or by name, and what happens if they are away.
- What is passed on. A summary of the conversation, the lead's answers so far and anything the agent has already promised.
- What the agent says meanwhile. A short, honest message that a person will pick this up, without guessing at an answer.
- Limit. How long before the lead is reminded or someone else is told, so that a hand-off never becomes a dead end.
A template to copy
Write each rule as one block, in this shape:
- When: the lead asks about pricing, discounts or contract terms.
- Hand to: the account owner. If unavailable, the team lead.
- Pass on: a summary, the lead's answers, and the questions the agent asked.
- Agent says: that a colleague will pick this up, and when to expect a reply.
- Escalate if: no one has replied within the time your team sets.
Rules most teams need first
- Pricing, discounts and contract questions go to a person.
- High-fit accounts, above a score your team chooses, get a rep for the first call.
- Topics your compliance or legal team marks as sensitive go to a person.
- A lead who asks for a person gets one.
- New message templates are approved by a manager before the agent sends them.
- A rep's manual edit to a CRM record is never overwritten by an agent.
Routine work, such as answering common questions from approved replies, gathering basic details and booking time on a calendar, can stay with the agent. That split is what frees reps to spend time on the conversations that need them.
Mistakes that make rules fail
- Rules that describe a goal instead of a trigger, such as "hand off when it seems important". An agent cannot test for that.
- Rules with no owner, so the lead is handed to a queue nobody watches.
- Rules written once and never reviewed, while the product, the pricing and the team change.
- Too many rules at the start. Begin with the few that protect you and add more as the data shows where agents are comfortable.
Test the rules before they go live
Rules should be tested the way software is tested. Take a sample of past leads, including the awkward ones, and check what the agent would have done under each rule. Where your team would have acted differently, change the rule or the wording and run the sample again.
Then start with a share of live leads and keep a way to switch agents off. Look at the hand-offs themselves: were they to the right person, with enough context, and fast enough? A rule that fires too often wastes a rep's time. One that fires too rarely is a risk.
Keep them readable and changeable
The best test of a rule set is whether a new manager could read it in ten minutes and say what the agents will and will not do. Keep the list short, name an owner for it, and review it on a schedule. Agents should work to your rules, and the rules should be yours to change at any time.
For how we think about the controls around agents more broadly, see our security page. Read about security and guardrails