AI reply assistant
A first draft that already knows your answers
The assistant reads the thread, your knowledge base, your canned responses and how similar tickets were resolved, then writes a reply in your tone. The agent edits and sends. Nothing leaves the desk without a human pressing send.
- Grounded in your docs, canned responses and resolved history
- Cites the sources it used so the agent can check them in a click
- Summarise a long thread before you reply, or before you escalate
- Human approval required - no autonomous sending
Reply assistant, grounded in your docs
I have been billed twice this month, can you check?
Inside the workspace
What your team actually sees
The assistant sits inside the composer: draft, adjust tone, insert a canned response with variables filled from the ticket, then send.
- HD-418212m left
Invoice total looks wrong
Priya - Northwindbilling - HD-4181breached
Cannot sign in to the portal
Marcus - Arclightaccess - HD-41781h 40m
Order SO-88213 never arrived
Dana - Kestreldelivery - HD-41743h 05m
How do I add a teammate?
Sam - Fielderhow-to - HD-41706h 12m
Refund request for March
Ana - Looprefund
Invoice total looks wrong
HD-4182 - Northwind Ltd - priority: high - owner: Rae S.
Priya Raman - customer
09:12
Our March invoice shows 14 seats but we only have 11 active agents. Can you check before we pay?
Internal note - not visible to the customer
Billing exported the seat log: 3 seats deactivated on 12 March, pro-rata credit due.
AI draft - grounded in docs + 42 past resolutions
Hi Priya, you are right - three seats were deactivated on 12 March. I have raised a pro-rata credit of £186 against invoice INV-2291 and it will appear on your next statement...
Customer
- Priya Raman
- priya@northwind.co
- Client
- Northwind Ltd - Tier 1
- Open tickets
- 2
- CSAT
- 4.8 / 5
SLA
- First response
- 12m left
- Resolution
- 5h 40m
- Policy
- Tier 1 - business hours
Audit trail
- 09:12 captured from support@
- 09:12 rule "billing" set priority high
- 09:12 round-robin to Rae S.
- 09:14 AI draft generated
Capabilities
Built for this job, not bolted on
Draft in one click
A complete reply, not a sentence starter, built from the specific facts on this ticket.
Knowledge grounded
Answers come from your internal docs and past resolutions, so the assistant cannot invent a policy.
Canned responses
A managed library with variables for customer, ticket and agent, filled in automatically as you insert.
Thread summaries
Turn a forty-message escalation into five lines before handing it to an engineer.
Tone and length
Shift a draft warmer, shorter or more formal without rewriting it from scratch.
Reviewable
Every AI-assisted reply is marked in the audit trail with who approved and edited it.
How it runs
From arrival to answer
- 1
Read
The assistant parses the customer thread and the account context beside it.
- 2
Retrieve
It pulls matching docs, canned responses and similar resolved tickets.
- 3
Draft
A reply appears in the composer with its sources listed underneath.
- 4
Approve
The agent edits, sends, and the outcome feeds back into future suggestions.
End to end
Everything this part of the platform covers
Each block below is a working part of the product, not a roadmap item.
Grounded drafting
Drafts are written from your own material, so the answer matches how your team actually replies.
- Context from the live thread, the customer record and their history
- Retrieval across your docs and previously resolved tickets
- Citations to the source used, so an agent can check before sending
- Tone follows your saved replies rather than a generic assistant voice
Human approval
Nothing is sent automatically. The assistant produces a draft; a person decides.
- Approve and send, edit, or discard in one click
- Edits are the reply - no separate copy-paste step
- Insert a doc or canned response into the draft mid-edit
- Every generated draft recorded in the ticket audit trail
Triage assistance
The assistant helps before the reply too, by summarising and classifying inbound work.
- Thread summaries for long or escalated tickets
- Suggested tags and priority based on content
- Duplicate detection that points at the merge candidate
- Suggested next action when a ticket has gone quiet
Control and cost
AI is a feature of the workspace, not a separate product with its own rules.
- Turn assistance on or off per workspace
- Admin-only configuration, enforced server side
- No customer data used to train third-party models
- Usage visible so the value is measurable against volume
Who it is for
What each person gets out of it
Agents
A solid first draft in seconds, with the sources to sanity-check it.
New starters
Ramp on how your team answers by editing rather than starting blank.
Team leads
Consistent tone and accuracy without policing every reply.
Customers
Faster answers that still read like a person wrote them.
Features inside this area
Go a level deeper
Knowledge search
The retrieval layer drafts are grounded in.
Canned responses
Your approved wording, reusable and variable-driven.
Merge and split
Act on duplicate detection in one step.
Audit trail
Every draft and send recorded.
Tags and catalogue
Where suggested classifications land.
CSAT and feedback
Check that faster still means better.
Detail
The small print that matters
Data handling
Drafting happens per request against your workspace content; ticket content is not used to train shared models.
Coverage gaps
When the assistant cannot find grounding it says so and suggests a doc to write, instead of guessing.
Per-workspace tone
Set the house voice once - greeting, sign-off, formality - and every draft follows it.
Measurable
Reports show how many replies started as drafts and how heavily they were edited.