The edit path is the real product
Perfect suggestions are a minority of any queue. What decides adoption is whether a nearly-right draft is faster to fix than to replace.
Resolution Agent
AI Companion for Zendesk
The same idea applied to Zendesk: suggested answers, similar past tickets and a drafted reply alongside the ticket your agent is working — including the part that usually gets skipped, which is what an agent does when the suggestion needs editing before it goes out.
- Suggestions and drafts alongside the ticket
- Similar past tickets surfaced automatically
- Editing is the normal path, not the exception
- Zendesk stays the system of record
The suggestion is right about the problem and wrong about the customer
Most evaluations test whether the suggested answer is correct. That is the easy half, and it is not what determines whether agents keep using it.
What matters is the far more common case where the suggestion is nearly right. If editing it takes as long as writing from scratch, the agent stops looking. The pitch shows the perfect answer; the queue is mostly the other one.
What an agent actually does with a suggestion
Three of these are preparation. The fourth is where the time is won or lost.
- Step 1
See what resembles this
Similar past tickets are surfaced alongside the current one, matched on the problem rather than on shared wording.
- Step 2
Get a starting draft
A reply is drafted from the resolution and the ticket context, as a starting point rather than a send.
- Step 3
Check where it came from
The sources are attached, so the technical detail can be confirmed before an agent puts their name to it.
- Step 4
Edit and send
The agent fixes the tone, drops what the customer already knows, and sends. How cheap that step is decides everything.
Three things that decide whether agents keep using it
None of them is how good the answer is on a perfect case.
It lives beside the ticket
Everything appears next to the ticket the agent already has open in Zendesk. No second console, and no copying between systems while a customer waits.
Corrections are signal
What agents change is evidence. Consistent edits to the same kind of suggestion show where grounding is thin, and which categories are not ready for more autonomy.
Why the usual approaches fall short
Each one is fine on the easy tickets and no help on the rest.
Macros and canned responses
Fast for the cases somebody predicted. They do not read the ticket or know what was already tried.
A general assistant in another tab
The agent copies the ticket in and the answer out, which costs more than it saves.
Knowledge base search in the sidebar
Returns articles matching the words in the ticket, leaving the judgement where it was.
Auto-replying without review
Fine until the suggestion is confidently wrong to an important customer.
What things are called
Draft is doing real work in this vocabulary.
- Ticket
- The Zendesk record an agent works. Ascendo attaches to it rather than replacing it.
- ResolveGPT
- Ascendo’s resolution capability surfacing real-time recommendations and relevant information inside the tool an agent is using.
- Similar tickets
- Previously resolved records matched on the problem rather than on shared wording.
- Draft
- A prepared reply for an agent to edit and send. Not an automatic response.
- Grounding
- Restricting what the system answers from to your own documentation and resolved tickets.
Frequently Asked Questions
The agent edits it, which is the normal path rather than a failure. What matters is that editing a draft is genuinely faster than writing from scratch — if it is not, agents stop looking, and no amount of accuracy on the easy cases makes up for that.
Resolution Agent
Test it on the awkward tickets
Do not send us the easy ones. Send the tickets where the answer needed judgement, and we will show you how much of the work a draft actually saves.
Talk to us