Three settings decide the automation level
There is no single automation slider. The level the AI works at is the combination of three settings, each decided separately.
| Setting | What it holds | Default | Set per |
|---|---|---|---|
| Human review | Every AI reply on email and WhatsApp, until a colleague approves, edits or rejects it. | Your choice | Channel and topic |
| Action approval | Any change in a connected system, such as a refund, an address change or a rebooking. | Approval on | Action |
| Hand off conditions | The situations that send a conversation to the team you define, with the history and a summary attached. | Your conditions | Specialist |
Your team decides where the line sits: which topics people handle, which actions need approval, and which answers the AI gives on its own. The architecture behind this is on how it works.
Approving a reply
A reply waiting for approval lands in the Typewise Inbox or in your existing helpdesk with the full context: the conversation, the customer and order data, and the drafted reply. Your agent edits, approves and sends it in a click, and the AI picks the case up again from there.
For billing, a reply approved without changes counts as a full resolution, because the AI did the work and a person only approved it. A reply the person edits counts as a partial resolution. The definition is at what counts as a resolution.
Approving an action
Actions change something in a connected system, if that system offers the operation. Each action has its own description, inputs and approval setting, and you can check it in test conversations before it goes live. You can keep a sensitive action, such as refunds above an amount you set, on human approval while routine ones run on their own.
An input can be limited to verified values: a customer signed in to your portal, an email address confirmed with a one time code, or the phone number on WhatsApp. Without a verified value the action does not run, so a change only applies to that customer’s own order or shipment. Where no integration exists yet, the AI collects what the change needs and raises a manual action for your team.
Hand off to the right team
Hand off conditions are set per specialist, and team routing sends each request to the team that owns it, in the Typewise Inbox or in the helpdesk you already use, such as Salesforce, Zendesk, Freshdesk or Outlook. Tickets and their context flow both ways, so nothing is lost in the hand off, and the customer keeps one conversation rather than starting a new one. More on ticket management.
Growing autonomy step by step
Start with one specialist and human review on
Read every reply before it goes out. Agents can rate conversations, and an Unrated filter finds the ones still waiting for a quality check.
Switch review off topic by topic
Once the replies for a topic prove themselves, let that specialist send on its own while the others stay reviewed.
Open up actions one at a time
Keep approval on for sensitive actions and remove it where the action is routine and the input is a verified value.
Let Nova propose the next change
Nova, the AI Operator, watches quality, drafts fixes and waits for your approval. Its changes stay drafts until someone on your team publishes them, and Simulator tests check the expected behaviour again after every change.
Typewise recommends this conservative rollout: a single specialist with human review first, then more scope as the evidence comes in.
Logged and under your control
Every action the AI takes is logged, and the Insights Hub shows where conversations escalate to human support and every case where a guardrail held the AI back. Role based access and SSO or SAML control who can change what, and you can pause the AI at any time. Certifications, hosting and audit logging are described in the trust center.
Where to read next
The platform end to end is on how it works. Security, hosting and compliance are in the trust center. Plans and the per resolution rate are on pricing, and the free trial runs on your real tickets for 14 days with no credit card.