The confidence threshold — in depth
Every time the AI drafts an answer, it also estimates how certain it is that the answer is correct and complete — a confidence score from 0 to 100. The score reflects several factors: whether the knowledge base contains a clear answer, whether the relevant shop data (order, product, policy) was found unambiguously, and whether the customer's request itself is clear.
Your threshold is the line that separates “send it” from “show it to me first”:
- Score ≥ threshold → the reply is sent automatically and marked “Answered by AI”.
- Score < threshold → the case becomes a ticket with the draft attached, waiting for your approval.
Practical guidance:
| Threshold | Behavior | Recommended for |
|---|---|---|
| 90–95 | Only near-certain answers go out automatically. | First weeks, cautious merchants, sensitive niches. |
| 80–89 | Balanced: most routine questions automated, edge cases escalate. | The sweet spot for most shops after week two. |
| 65–79 | Aggressive automation, more AI-handled volume. | High-volume shops with a mature knowledge base. |
Start high, then lower gradually as you see the drafts consistently match what you'd write. Your ticket view shows the score for every escalated case, so you can judge where your threshold should sit.
Automation levels: full / suggest / off
Independently of the threshold, each mailbox has a master automation switch:
- Full automation — the AI answers above-threshold emails on its own, asks customers clarification questions directly and sends acknowledgements for complex cases. Maximum time saved.
- Suggest-only — the AI does everything except send. Every draft, including high-confidence ones, waits for your one-click approval. Perfect for building trust in the first weeks: you see exactly what the AI would have sent.
- Off — no AI processing for this mailbox. It behaves as a plain mail client with ticketing. Useful for mailboxes that shouldn't be automated (e.g. a partnerships or press inbox).
Tone of voice & emoji control
Under Settings → AI → Tone you choose how the AI writes:
- Like my previous emails — applies the style guide learned from your email history: your greetings, phrasing, length and sign-offs.
- Young & friendly — casual and warm.
- Short & factual — concise, no small talk.
- Formal — traditional business correspondence.
The separate emoji toggle overrides everything: switched off, the AI will never include an emoji, regardless of tone or how the customer writes.
Knowledge base auto-activation
As the AI learns — from shop data, email history and every resolved ticket — it creates new knowledge entries. The auto-activation setting decides what happens next:
- On — new entries become active immediately and improve answers without your involvement. Recommended once you trust the system.
- Off — new entries land in a review queue; the AI only uses them after you approve. Maximum control, slightly more work.
Either way, the Knowledge section always shows every entry, and you can edit or deactivate any of them at any time. A good practice: keep auto-activation off for the first two weeks, review what the AI learns, then switch it on.