Codegeist Unleashed, our story creating an app in Forge using AI 👩🏻‍💻 🤖

[#1]

Every story has a beginning, a twist, and a conclusion; this is ours.

It all started in the year 2022 when we formed a new Deiser team, product discovery. Responsible for working on new apps to launch into the Atlassian Marketplace, and always keeping Codegeist 2023 in mind.

For the team's idea management, we started using Jira Product Discovery, and there was an idea with significant potential for managing project economics. During the Appweek in Berlin, we started working on it and launched it on the Atlassian Marketplace in April. Then we want to validate its market fit and also presenting it at Codegeist 2023. This application, a native Forge app with CustomUI, not only added value but also provided a good user experience.

We had everything ready to present an app at Codegeist, and we wanted to focus on preparing good content to position the product in the Hackathon.

In September, when Codegeist 2023 Unleash was announced, all our plans fell apart because one of the requirements was to use AI for the product. We thought we had it all, and we had to start from scratch 😨.

First, we thought about how we could apply AI to the application we had already developed. None of the enhancements fell into any of the 3 Codeigts categories 😞.

Which new idea did we select?

With less than two months available to build something, we debated whether to participate in Codegeist or not, as we had to find a good idea and start working on it. Besides, some team members were still on vacation (in Spain, summer vacations usually span from June to September).

We have a Discovery team, we love innovation, and we didn't want to miss out on Codegeist, as it is a unique learning opportunity that creates space for testing and rapid learning 💪. So, we got to work and analyzed the ideas from our Jira Product Discovery.

One idea called "Sentiment Analysis" stood out from the rest because:

• It added value by analyzing texts using OpenAI.

• We had already conducted a technical feasibility study with good results.

• The problem the idea solved had a high confidence index for success.

• And it fell into one of the Codegeist categories: AI Apps for Data-Driven Insights

With all this, our RICE matrix positioned the idea ahead of the rest; we just had to start designing and developing it.

🔍 Research

We needed to start investigating quickly, so we conducted 6 interviews 🎤 with support agents and leaders who work daily in Jira Service Management.

We confirmed that there were issues in teams with more annoying users, although not all companies confirmed that they would prioritize them. We saw an opportunity, but we also knew that it was not something they were currently thinking of solving.

We also conducted desk research and found potential competition in the marketplace, but they didn't seem to have succeeded. We believed that they did not offer users actionable and deep data as we would like to provide. Moreover, we thought they had not achieved success due to a lack of promotion and perhaps launching at a time when people were not sure about investing in AI.

We got to work to offer what would be most relevant to support agents, support leaders, and customer success managers.

On this journey, we conducted a market research:

Two weeks after the announcement, we already had our first designs in Figma 🎉. With various proposals, we knew we had to cut down ✂️ and see what we could develop in the remaining weeks.

We used our development framework to define the minimum we wanted to bring to Codegeist.

• Make it possible:

  • Issue view

• Make it confortable:

  • Insights by project

• Make it delightful ❤️:

  • Aim to offer 7 reports

🧑🏽‍💻 Development begins

On September 18, we started development before having the designs 100% complete. We had clear and defined data to analyze, and it was time to start.

First hurdle

A few days later, we found that we couldn't fit all the information we wanted to show support agents vertically in the issue context.

So, we ultimately decided to place Gomood in the issue activity module.

Refining Insights while development started

While development began with data recording and how it would be displayed in the issue, insights were still to be defined.

We didn't want it to be just another dashboard filled with charts 📊 that didn't add value to users' daily work. That's why we met with several people from the Deiser Solutions and Support team to present our future app and gather feedback. What provided the most value? What were they missing?

Based on their feedback, we selected the reports that would provide the most value to support leaders and customer success managers. Development could now start building the necessary algorithms to create the Insights part of each Jira Service Management project.

Next, we conducted several user tests 🧪 with other team members to ensure the app was understandable. This is where we fine-tuned the texts.

Development progressed well, and it seemed like we were going to present 7 reports on the project's Insights page.

A plug & play app

One of our obsessions was to ensure that users could see the value Gomood provided in just a few minutes. As Oscar Wilde said: “You Never Get a Second Change to Make a First Impression”.

As the app was developed, both at the issue and project levels, users depended on customer comments to begin sentiment analysis. This external dependence meant that users testing the app couldn't do anything to start seeing data in Gomood, and this situation could drag on for days, even weeks 😱.

💡 Solution:

• At the issue level: we analyzed the issue title and description and the last two customer comments. This was done when:

  • The first customer comment was added after installing Gomood

  • When a Jira user clicked on 'Customer Mood'.
    
    
    

• At the project insights page: users might not see data or see very little data when testing the app, all because Gomood had analyzed few issues.
  We decided to allow users to analyze a good volume of issues for the project, specifically five for each of the reports.
  
  

Take care of AI platform quotas

We were testing with a trial account that had very restrictive rate limits, and some of our functionalities exceeded them, particularly the initialization of issues and projects. At this point, we created a permanent account and went from being able to make 3 requests to OpenAI per minute to 3500.

The toughest technical challenge 😰

The most important aspect of Gomood is the insights it provides, dealing with a large amount of data at both the issue and project levels. For this reason, we had to focus on data processing to achieve the best possible performance 🏎 within the Forge platform limits.

Development finished, time to prepare the content 📚

In the last few days of development, we fine-tuned the text and fixed small details that we noticed during testing while preparing the content to present the app at Codegeist. Finally, app development was completed on October 19, within 4 weeks we had everything ready 🙌🏽.

As the final lines of code were being written, the script for the presentation video was being prepared. This stage was facilitated by the earlier ideation phase, where we identified our user personas, established their key benefits, and their value proposition.

This year, we couldn't rely on our marketing team, they are so busy. So for recording 📹, we chose Powtoon, a tool that allows non-marketing experts to create a video in a simple way. The most challenging part of the video was preparing an instance with data that reflected the value in each of the reports 📈.

Impossible alone, possible together 👩‍👩‍👦‍👦

The Deiser product team consists of 13 people. Although four of us were deeply involved in the ideation of Gomood, the rest were also essential. They ensured that the other apps continued running 🏃🏾‍♂️ as usual and helped us with feedback and overcoming obstacles. Gomood was created thanks to this amazing team.

This year we have had a front end profile, @David González, a back end, @Alvaro Aranda, the research and design in charge of @Laura Barreales Rubio and me leading the team and strategy.

Special thanks to @Diana Aceves for helping us in the last part of the Codegeist in the layout of the reports and to @federico_baronti  for giving voice to the video. 

The best of this Codegeist

If we have to keep one thing from the 7 weeks of work, it would be the importance of the new Product Discovery team in making everything flow faster, with a clearer market objective and in a more agile way than ever for our team.

On a technical level, the best thing has been the immersion for the first time in the world of artificial intelligence focused on offering an app with value for Jira users.

[Pablo Beltran _Marketplace Expert_]

Nice! Congrats! 

Did you take into acoount the possible interference of other AI apps inmpact on sentiments?

[Leo Diaz _ DEISER]

Hi Pablo! Thank you!

What kind of interferences?