Playbooks & agent-assist

“Playbook” means two different things in Zubby, and it is worth getting the distinction straight up front:

• Outbound Playbooks (Dashboard → Playbooks) — a library of pre-built automations you install in one click. Each one becomes a real, editable journey that runs on a trigger (cart abandoned, back in stock, post-purchase…).
• Agent Playbooks (Playbooks → Agent) — step-by-step procedures the live chat agent follows inside a conversation when a trigger matches. This is the in-chat counterpart to the outbound library, and the foundation of agent-assist.

Outbound Playbooks (the automations library)

The Playbooks tab is a catalog of curated, high-ROI automations. Each card shows the trigger, the channel sequence with delays, the category, and an illustrative impact estimate. The shipped catalog includes:

• Cart Rescue · 3-touch recovery — email now, SMS at 6h (only with SMS consent), final email with an optional approved discount at 24h.
• Welcome Series · first-time shoppers — a warm intro, a concierge/quiz nudge, and a social-proof email over five days.
• Win-back · 30-day dormant — re-engages shoppers with no activity in 30+ days.
• Browse Abandonment · high-intent viewers — two gentle touches for shoppers who viewed a product repeatedly but never added to cart.
• Back in Stock · subscriber blast — fires the moment inventory flips from zero to positive.
• Post-Purchase Thank-You — a single human thank-you and review request after fulfilment.

Installing a playbook

Click Install playbook and Zubby creates a real journey plus its steps from the template, then drops you into the journey editor to tune it. Installed playbooks run exactly like hand-built journeys — they are not a separate system. The card shows an Installedbadge with the journey's status afterward, and Open jumps straight to it.

Build a custom playbook

Don't see what you need? The Build a custom playbook form at the top of the tab creates an editable draft journey from a name, a trigger, a first channel, and an optional opening message. Supported triggers: cart abandoned, browse abandoned, checkout abandoned, back in stock, price drop, post-purchase, and win-back. First channel can be email, SMS, or web push. Save creates the draft and opens it on the Automations tab, where you add more steps with the normal editor.

Agent Playbooks (live-chat procedures)

The Playbooks → Agenttab is where you author the procedures the AI follows mid-conversation. A classic example: “When a shopper asks about returns — confirm the order number, state the policy from a Locked Fact, then offer to start the return.” A matched playbook is injected into the conversation as an ordered procedure the agent must work through.

Anatomy of an agent playbook

• Name— for your reference (e.g. “Returns help”).
• Trigger type — one of:
  • Always on — runs on every conversation.
  • Keyword match — fires when the message contains any of your comma-separated keywords (e.g. return, refund, exchange).
  • Page type — fires on a specific widget page: product, collection, cart, checkout, order status, post-purchase, account, or home.
• Priority — 0–100; higher wins when several playbooks could match.
• Steps— the ordered procedure (up to 15 steps, 400 characters each). Steps reference tools by intent, e.g. “Confirm the shopper's order number with get_order_status.”
• Active — inactive playbooks are saved but never matched.

Pair playbooks with Locked Facts

Playbook steps tell the agent what to do; they are not the place for exact policy wording. When a step needs precise language (a refund window, a warranty term), point it at a Locked Fact so the agent quotes the approved text rather than improvising. See Training for how facts are managed. Each step is sanitized before it is injected, so playbooks cannot be used to smuggle prompt instructions.

Agent-assist for human reps

When a conversation is handed off to a person, the AI stays on as a co-pilot in the inbox. Reps get:

• Suggest reply — a context-aware draft built from the transcript, shopper profile, and your knowledge base, ready to edit and send.
• Canned replies — insert a saved snippet with a keyboard shortcut.
• Sensitive-language flags — the AI warns before you send a risky commitment (for example promising a refund) so wording stays safe.

Full detail on the rep experience lives in the human handoff and inbox guides.

Outbound vs. agent playbooks at a glance

Related

• Human handoff & agent-assist
• How the AI agent works
• Journeys — what installed playbooks become
• Rules engine — for merchandising and objection controls
• Training & Locked Facts