AI Support+ for Jira

10x your support game using AI powered by ChatGPT

Comment

Try it now: https://marketplace.atlassian.com/apps/1232262/ai-support-for-jira

Inspiration

I run one of the largest Atlassian Marketplace app vendors (Apps+) and have made it a priority from day one to deliver first-class rapid support. I give every customer my personal email.

We provide 24/7 support and typically respond within minutes-hours. The maximum response time is 12 hours. And we don't only respond, we aim to solve the issue and have it deployed within that time window.

However with all of these new AI tools, even I find myself first going to Google instead of ChatGPT. That behavioural habit is very much wired into all of us and has cost me countless hours of wasted efficiency.

In one support ticket example I spent 3 hours searching Google and running tests in code before I finally decided to dump the ticket text into ChatGPT, which returned an accurate diagnosis of the problem within seconds. That was more than enough inspiration to build this app.

What it does

The core feature and value offering is automation and saving time. When you open a new support ticket for the first time there is an internal comment that has already translated the customer request if needed, and has already attempted to diagnose the issue. No information hidden behind UI buttons.

Additionally for support teams the app can also automatically assign new support tickets based solely on natural language; eg Mary: "design and implementation". Bob: "sales and billing". Those can be configured along with other settings on the admin config page.

And finally the app features an inbuilt ChatGPT-like interface (click "AI Support+" in the activity section of a Jira issue) to continue to solve the support ticket in unison with AI and your teammates. No need to copy-paste between browser tabs. Then when you're ready to respond to the customer, there's a second interface within which you can use AI to help quickly craft a response.

How we built it

Forge, Jira APIs, Custom UI (React), UI Kit (settings), OpenAI APIs.

Challenges we ran into

Non-determinate responses and behaviours from AI models is always a novel pain. Also I've never been a fan of the server-client separation in the Forge platform using invoke with @forge/bridge. For me it massively slows down development time. Tunnel dropouts and non-descript errors are also frustrating on the platform because you're never quite sure if it's on your end or Forge's. Hopefully that improves over time.

Accomplishments that we're proud of

We've managed to move much of the complexity in UI/UX to automations that occur behind the scenes. And will continue to do that wherever possible with new feature requests. Seeing GPT4 automatically assign a ticket based on a simple natural language instruction was pretty cool to see work for the first time :)

What we learned

While the OpenAI APIs and models have improved since I first used them many years ago (before ChatGPT existed), working with AI models is still a difficult developer challenge given their indeterminate responses.

What's next for AI Support+ for Jira

Iterating features and bug fixes based on customer feedback. The next core feature will be perfecting fine-tuning on Jira and Confluence data when given customer opt-in. We did some basic tests on fine-tuning but it's one of those things that is so fickle to get right that you can't rush it.

Built With

Share this project:

Updates