The recovery playbook
Every setting on the playbook card, and what each one changes.
The playbook lives in one place: Settings → Resolutions (/settings/resolutions). It says how the AI talks to an unhappy customer, what it may do without asking you, and when a conversation stops. The offers it can give are set up on the same page, below the playbook — see Offers and promo codes.
Turning it on
If you’ve never enabled it you’ll see a single button, Enable AI Resolution. That switches it on with sensible defaults — friendly tone, short messages, conversations starting automatically, everything drafted for your approval.
The master switch is Recovery conversations.
Voice
Tone of Voice — Formal, Friendly, Optimistic or Serious. Pick the register you actually use with customers.
Custom tone prompt (Pro) — describe your voice in your own words instead of a preset. “Warm but brisk, we never over-apologise, we always offer something concrete” produces noticeably better copy than any single-word tone.
Message Length — Short, Medium or Long. Short is usually right; a long apology reads as defensive.
Whatever the voice, every reply follows the same rules: it names the specific thing that went wrong, owns it once, says what happens next, asks at most one question, and never asks whether the issue is resolved or whether the customer is satisfied.
When a negative response comes in
Start a conversation automatically — the default. Every negative response gets a first message drafted as it arrives.
Only when I click Resolve — nothing happens until someone starts the conversation from the response. Use this for your first week while you calibrate, or if nothing should go out unreviewed.
What the AI may do on its own
This is the setting to think hardest about.
- Draft everything for me — every reply and every offer waits in the Inbox until someone approves it. The default.
- Send replies, ask before offering — replies go straight to the customer; any offer the AI wants to give still waits for approval.
- Send replies and give offers — replies go out on their own, and the AI may give the offers you marked AI may give it — within each offer’s own rules (rating cap, once per customer, monthly limit).
Offers marked Team only are never given by the AI in any mode. A hand-off to a person always sends immediately, whatever the mode, so a customer asking for someone is never left waiting on an approval.
When to stop
AI replies per conversation — how many messages the AI may send in one conversation (four by default). When it runs out, the conversation is handed to your team. Switch Hand off to the team when out of replies off and it sends a short wrap-up and closes instead.
Close after silence of — how many days without a customer reply before the conversation closes on its own (five by default). Silence after an offer counts as settled; silence with no offer counts as unresolved. If the customer writes back later, the conversation reopens and the AI picks it up again.
Let the AI promise a callback — for disputes or anything a call would settle faster, the AI may tell the customer a manager will phone them, and flags it for you. See Callbacks and hand-offs.
What happened to apology, thank-you and promo-code resolutions?
They’re gone. An apology is part of every reply, not a type you pick; a thank-you for positive feedback belongs in automations alongside the review request; and promo codes are now real offers with codes, rules and a record of who received what.
Was this article helpful?
Thanks — that helps us fix the gaps.
Still stuck? Talk to us — or open the chat in the corner.