You configure the AI qualifier from one playbook: a persona describing how it should sound, your qualification criteria, and a numbered list of questions with unique keys. Save it as a draft, try it out in the built-in test chat, then publish it live once it has a persona, criteria, and at least one question.
Setting up the playbook needs admin access on your account; agents can view a conversation's AI activity but can't edit the playbook itself. Before you can publish a live playbook, AI qualification also needs to be turned on for your account. If you don't see the option, contact support.
| Field | What it's for |
|---|---|
| Persona | How the AI introduces itself and the tone it replies in |
| Qualification criteria | Plain-language description of what makes a lead a good fit |
| Questions | A numbered list the AI works through, each with its own key |
| Success / disqualification actions | What happens to the lead's stage and tags on each outcome |
| Turn cap | How many back-and-forth turns the AI gets before it stops on its own |
| Handoff message | What the AI sends when it hands a conversation to a human |
| Read customer images | Optional toggle letting the AI look at photos the customer sends |
Each question needs its own key (a short internal label, like budget or timeline) so its answer can be matched to a field on the lead. Saving two questions with the same key is rejected.
Criteria are plain language, not a formula, the AI uses them to judge whether an answered set of questions adds up to a qualified lead. For example, a solar installer's playbook might use:
homeowner), "What's your average monthly electricity bill?" (key monthly_bill), "When would you like to install?" (key timeline).As the customer answers, each answer is extracted onto the matching field on the lead, and the AI weighs the answers against your criteria to decide qualified, disqualified, or still pending.
You can edit a live playbook's wording freely. What you can't do is edit a live playbook down to missing its persona, criteria, or questions, that edit is rejected until you pause the playbook first. This stops a live playbook from silently going incomplete mid-conversation.
If the AI hits the turn cap on a conversation without reaching a decision, it stops and leaves a note for your team rather than continuing to loop.
A playbook needs a persona, qualification criteria, and at least one question before it can go live. Use the test chat to try sample conversations first, test-chat turns don't touch a real customer and don't count against your usage. Once you're happy with it, move the playbook from draft to testing and then to live.
For how the AI behaves once it's live, see What the AI qualifier does. For how the AI extracts a quick read of any conversation on demand, see Conversation insights.