The loop in six steps
The goal is resolution rather than deflection: the AI closes routine requests completely, from reading the message to looking up the order, applying your policy and sending the reply. Every request follows the same path, whichever channel it arrives on. This page gives that path in words; the animated version, with four worked requests, is on how it works.
The AI supervisor reads the intent
It analyses the request and routes it to the right specialist agent, or to a person when the request calls for one.
The specialist acts across your systems
A specialist handles one topic, such as billing, returns, warranty or bookings. Knowledge agents retrieve information from your sources; action agents look up and change data in your CRM, ERP, billing, logistics or commerce systems, within the permissions you give them.
A review agent checks the response
Before anything goes out, a review agent checks the outgoing response against your guidelines.
The customer gets the reply
In the customer’s language and in your tone of voice, in the same conversation they started.
A person steps in when it matters
Approvals, judgment calls and full takeovers hand off to your team with the full context attached.
The AI picks the ticket back up
Once a colleague replies or approves, the AI continues and takes the ticket the rest of the way.
Examples of what a specialist completes on its own, where your systems offer the operation and you allow it: rebooking a trip to new dates, checking warranty coverage and creating a replacement order, prorating and applying a subscription upgrade, or recommending products from your catalogue and stock.
What the AI reads, and what it changes
- KnowledgeYour website, help center and knowledge articles, uploaded documents, or a shared Google Drive folder. Each source has an access level, so internal material stays out of customer answers.
- LookupsLive data read from the systems you connect at the moment the customer asks: an order, a booking, an invoice, a parcel.
- ActionsChanges in a connected system, such as an address, a booking date or a refund. Each is configured on its own, with approval by a person on by default.
- Verified valuesAn input can be limited to a verified value, such as the ID of a signed in customer, so an action only ever touches that customer’s own order.
Connections use native integrations and the Model Context Protocol (MCP), so a system with an API can be connected as lookups and actions. The current list is on the integrations page.
What a hand off carries
Hand off happens when a person adds value, not only when the AI cannot continue. You define the conditions per specialist: which topics people handle, which actions need approval, and which answers the AI gives on its own. Every hand off arrives with three things.
- The conversationThe full history so far, with a summary, in one thread the customer keeps.
- The customer and order dataWhat the specialist already looked up, so nobody asks the customer twice.
- A drafted replyYour agent edits, approves and sends it in a click, and the AI continues from there.
It lands in the built in Typewise Inbox or in the helpdesk your team already uses, such as Salesforce, Zendesk, Freshdesk or Outlook, and tickets and their context flow both ways. More on ticket management.
How a resolution is counted
Resolution is also how Typewise is priced, so the loop above decides the bill. A resolution is counted per ticket, not per message.
| Outcome | What happened | Counts as |
|---|---|---|
| Full resolution | The AI closed the ticket, or a colleague approved its reply without changes. | 1.0 |
| Partial resolution | The AI did part of the work and a person finished the ticket, for example by editing the reply. | 0.5 |
| Not counted | The ticket went straight to a person without the AI interacting with the customer, the AI never touched it, or your team flagged the reply or action as incorrect. | 0 |
The Insights Hub tracks resolution rates, counts partial resolutions at half weight, and shows where conversations escalate to human support. The definition with worked tickets is at what counts as a resolution; plans and the rate are on pricing.
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.