The AI Playbook is a set of instructions you write to tell the AI what to do when it handles a task. Think of it as handing the AI your team's standard operating procedure — if a customer emails about a dispute, here's what should happen first, second, and third. The better your playbook, the more useful the AI's suggestions will be.

Go to AI Control Centre → Task Planning → select a category → Playbook tab. There are two places to write instructions: the main Playbook field and the Rules section below it.

Write it as a clear sequence of steps, using "if / then" logic where the process branches.

What to include:

What to do first — What's the immediate response when a task of this type is created? Should the AI draft a reply straight away, or does it need to check something first?

Decision points — What changes the approach? For example, a Promise to Pay task looks different depending on whether the customer gave a specific date or just said "we'll sort it soon."

What "done" looks like — When should this task be considered resolved?

Example — Promise to Pay:

"If the customer has given a specific payment date, record the promise to pay and draft a reply confirming the date and amount. If the commitment is vague (no specific date), draft a reply asking for a specific date before recording anything. Only record a promise once a concrete date is known."

Example — Invoice Dispute:

"Draft a reply acknowledging the dispute and confirming which invoice is affected. If the customer hasn't provided specifics, ask for the invoice number and the nature of their concern. Once details are received, escalate internally before responding further."

Be specific, not general. "Handle the customer appropriately" tells the AI nothing. "Draft an acknowledgement email and ask for supporting documentation" tells it exactly what to do.

Use if / then logic. Most real processes branch depending on what the customer said. Writing out those branches explicitly gives the AI much better guidance.

Reference action types directly. If you want the AI to suggest a reply email, say "draft a reply email." The AI knows what actions are available for your category — you just need to tell it when to use them.

Keep it concise. A tight 3–5 sentence playbook often outperforms a long paragraph.

Rules are shorter, more specific constraints that sit alongside the playbook. Use them for things that apply regardless of context — thresholds, escalation triggers, or standing instructions. Each rule should be one clear instruction. You can add up to 20.

Good rule examples:

Before your playbook goes live, test it in the Playground. Paste in a sample email that represents a typical scenario for this category. The Playground shows you exactly what the AI would do — which actions it would suggest, what draft it would write, and why — without touching any real tasks.

If the AI's suggestions don't match what you expected, check two things: whether your playbook is specific enough, and whether the action types you're expecting are on the allowed actions list for the category.

Your playbook doesn't need to be perfect from day one. Start with the most common scenario, test it, and refine based on what the AI produces. Add rules as edge cases emerge. The more your playbook reflects how your team actually works, the more useful the AI becomes.