I helped a client with Awish to remove the most repetitive part of their support process
A client told me their support team was spending too much time before they could even start helping customers.
The problem wasn’t writing the reply. Every ticket had to be read, understood, checked for urgency, compared with their documentation, and routed to the right person.
So I built an Awish workflow for them.
Now, when a new Zendesk ticket comes in, Awish understands the issue and urgency, checks their Notion documentation, and prepares a response.
I
f it looks like a product bug, it creates a Jira issue with the customer context. Important cases go straight to the team in Slack, while customer facing actions still require approval when needed.
We also connected WhatsApp, so they can follow and manage the workflow without opening Awish.
The biggest change is that their support team no longer opens a ticket and starts from zero. Most of the context and next steps are already there.
Which part of your support process still takes more manual work than it should?
Apps used: Awish, Zendesk, Notion, Jira, Slack, WhatsApp
Replies
@alpertayfurr The thing I'd watch is the pre-filled context quietly becoming the answer. Once the draft and the urgency label are already sitting there, the agent's job stops being work out what's true and becomes does this look right, and those two produce very different error rates. The number I'd instrument is how often a human changes the urgency label after the fact. Near zero means either the classifier is excellent or nobody is really checking, and those look identical on a dashboard. Related, what happens when the Notion page is stale? You get a confident reply that cites your own documentation, which is worse than no reply, because it's wrong with a source attached.
@asadmalik901 Good point, especially on stale Notion content.
I actually handled that by telling Awish in chat to pause whenever the source looks outdated, conflicting, or low confidence. It added that rule to the workflow memory, so those cases go to human review instead of becoming a confident draft.
@alpertayfurr That helps, but it's the model judging its own confidence, and that's the same model that would have written the confident draft. Stale isn't something you can read off the text, a page can be beautifully written and eighteen months wrong. The cheaper rule is the Notion last-edited date, anything older than 90 days goes to review regardless of how sure the draft sounds. Facts beat vibes for that one.