A playbook is a short conversation your bot runs when a visitor's message calls for it: book a call, qualify a lead, take a quote request, collect an order problem for your team. The bot still does the talking and still answers anything else the visitor asks. tufu keeps the order of the steps, checks every answer and does each action once.
Find them under Bots → your bot → Playbooks.
Start one#
Pick a template (Book a call, Qualify a lead, Quote request, Order problem), start from blank, or describe it in a sentence ("When someone asks about catering, ask how many guests and the date, take their email, and hand it to us") and tufu writes a first draft of the steps. A description that doesn't make a complete playbook gives you an error, never a half-finished one.
When it starts#
Each playbook has a line saying when to use it, in plain words ("the visitor wants a demo"), and optional keywords. A message with a keyword starts the playbook straight away; otherwise the bot reads the "when" lines and starts the one that fits. A bot runs one playbook at a time in a conversation.
Steps#
- Say: the bot says something, in the visitor's language.
- Ask: a question whose answer is checked: text, an email address, a phone number (with the country code, or a country you set for numbers without one), a number (with a smallest and largest if you like), or a choice of 2 to 10 options. In the web chat a choice shows as buttons; elsewhere the bot lists the options and the visitor replies with a number. An option can jump ahead to a later step.
- Save as lead: the answers so far become a lead, with the conversation, like the lead form's.
- Offer booking: a button to your booking page (Calendly, Cal.com or any https link).
- Hand off: a person takes the conversation, with every answer in the handoff's summary.
- Send webhook: sends
playbook.completedwith the answers to your webhooks. - End: the playbook ends, with optional closing words.
A playbook has up to 30 steps. Move a step with its arrows, or focus it and press Alt+↑ or Alt+↓. How it reads, beside the steps, shows the conversation as a visitor would see it; pick an option at each choice to follow that branch.
When things don't go to plan#
- If the visitor asks something else first, the bot answers it and then comes back to the question.
- After 3 answers it can't accept, the conversation goes to a person (or, where nobody can take it, the playbook stops) and it shows up in Knowledge gaps, like a question the bot couldn't answer.
- A visitor who stops a playbook before answering anything shows up there too: it may have started on the wrong question.
- A visitor who leaves 30 minutes after giving their email is saved as a lead with the answers so far, marked partial.
Each playbook shows how many runs it had, how many completed and how many were handed off, and how many runs reached each step, so you can see where visitors drop off.
Publishing changes#
Changes to a bot's behavior and its playbooks wait in a draft. The Try it drawer and Be your customer run the draft, so you can test a change before visitors see it. When it's ready, Review and publish shows what changes; your visitors get it as soon as you publish. Discard goes back to what's live.
Each publish keeps a version: the last 20, each one restorable. Restoring puts that version in your draft, to try and publish again. Your knowledge sources and the bot's look aren't part of the draft: they change for everyone as soon as you save them.
Test questions#
On the Team plan, a bot can have up to 25 test questions: things visitors ask, each with what a good answer does (cites a page, mentions a phrase, hands off to a person, or starts a playbook). tufu suggests some from the pages your bot cites most and the knowledge gaps you've answered.
When you review a publish, your draft and the live bot each answer every question once, and you see what got better or worse. They warn; they never stop a publish. They run at most once every 10 minutes per bot, and they're never counted as answers.