Automation
Playbook editor
A playbook is the job description you write for a bot — in plain language, not code. Describe what it may do, how it should sound and where it must stop, and the bot follows the script on every conversation, at 3 a.m. exactly as at noon.
What a written playbook buys you
Why plain language is the whole point
The person who knows how support should work — you, or your support lead — is rarely the person who writes automation scripts. Every translation step between 'what we want' and 'what the bot does' loses nuance and adds delay. Botdesk removes the translation: the playbook is the instructions, written the way you would brief a person.
That has a quiet side effect worth more than the automation itself: writing playbooks forces your policies to become explicit. Half the teams that draft their first playbook discover their return rules lived in three heads in three versions. After the playbook, there is one version — and both bots and humans follow it.
Rules the bot cannot bend
There is a difference between guidance and guardrails. Tone and phrasing are guidance — the bot adapts them to the conversation. Boundaries are guardrails: maximum refund amounts, forbidden topics, mandatory handoffs. The bot cannot be talked around them by a clever customer or a long conversation, because they are checked outside the conversation itself.
This is what makes delegation feel safe. You are not hoping the bot behaves; you wrote exactly where its authority ends, and everything past that line lands with your team — carrying a summary, so the person picking it up already knows the story.
A playbook's first month
Day 1 — draft and rehearse
Write the first version in plain language and replay it against last week's real questions. Rough edges show up in the preview, not in front of customers.
Week 1 — tune from real chats
The bot is live on its lane. Daily reviews of its conversations point at the two or three rules worth sharpening — each is a one-minute edit.
Week 2 — extend the lane
Confidence grows: raise the refund ceiling the bot may approve, add the follow-up questions it now answers well, let it close more without checking in.
Month 1 — the lane runs itself
The playbook has settled. You skim its conversations a couple of times a week, and edits become rare — mostly seasonal: holiday hours, promo terms, new products.
From blank page to a bot on shift
Describe the job
Write what the bot handles the way you would brief a new hire: 'You answer questions about returns. Our window is 30 days. Always offer an exchange first.' That is genuinely the format.
Set the boundaries
Spell out where the bot must stop: refund ceilings, topics that always go to a person, promises it may never make. Boundaries are rules, not suggestions — the bot cannot talk its way around them.
Rehearse before launch
Test the playbook against real questions from your history in a private preview. You see exactly what the bot would have said, adjust the wording, and nothing reaches a customer until you are happy.
Put it on shift and refine
Go live on one lane. The first week of real conversations will show you two or three rules worth tightening — each edit takes a minute and applies to every conversation from then on.
Frequently asked questions
- I'm not a writer. What if my playbook is bad?
- Start from a template — returns, order status, FAQ and other common lanes come with a solid draft you edit rather than a blank page. And the rehearsal preview shows you exactly how the bot behaves before customers ever see it, so a weak first draft costs nothing.
- Can the bot go off-script if a customer pushes it?
- No. Boundaries are enforced outside the conversation: a bot with a $50 refund ceiling cannot approve $60 no matter how the chat goes. When a customer needs something beyond the script, the bot hands off to your team — that path is part of every playbook.
- How do I stop the bot from overpromising?
- Overpromising is a boundary like any other. List the promises the bot may make — and it will make no others. Where the playbook is silent, the default is to step aside, not to guess. That default is the single biggest difference from the chatbots that earned the bad reputation.
- How often do playbooks need updating?
- Heavily in the first two weeks — that is where the sharpening happens — then rarely. Mature playbooks change when your business changes: new policy, new product line, holiday schedule. Each edit takes minutes and applies instantly to every future conversation.
- Can I reuse a playbook across bots or brands?
- Yes. Duplicate any playbook as the starting point for a new lane, and on the Fleet plan the same structure carries across brands — each copy stays independently editable, so the outlet store can be more generous than the flagship without either script leaking into the other.