Playbooks

A playbook tells your agent what to do, in order, when a particular kind of conversation comes up. Write it the way you would brief a new teammate.

Open Playbooks in the Build menu, after Behavior, and choose New playbook.

What a playbook has

• Name — for you, such as "Delivery check".
• When to use it — the situation, in a sentence: "A visitor asks whether we deliver to their area."
• Steps — up to 15, followed in order. A step is a plain instruction ("Ask for their postcode."). Add if / otherwise adds a step with two outcomes: If we deliver there, then tell them the next delivery date, otherwise suggest picking up in store.

Your agent takes one step per reply where it needs something from the visitor, skips a step that is already done, and still answers what the visitor asks along the way. A playbook never makes the agent invent an answer: if a step needs something it does not know, it says so.

Using your own data in a step

If your agent has endpoints in Your API, each step has a menu to pick one: "Check whether we deliver there", using your delivery endpoint. Your agent calls that endpoint at that step.

If Order help is on, the same menu offers look up an order and, where address changes are allowed, change a shipping address.

A playbook that uses an endpoint which is switched off or deleted, or Order help while it is off, is skipped, and the page says why. You cannot switch such a playbook on until you pick another endpoint or switch the endpoint back on. Deleting an endpoint tells you which playbooks use it.

Limits

Up to 10 playbooks per agent, 15 steps each, and 5 if / otherwise steps per playbook. Your custom instructions still take priority over a playbook.

Playbooks guide what your agent says. Which card a visitor sees (a booking calendar, the contact card, a ticket) is still decided the same way as before, so write those as plain steps: "Offer to book a call."