Kristina Arnold ─ UX / Product Designer

Open AI with it cutting-edge AI technology has the potential to make data analysis and dashboard building easier and faster for users of Astrato. The design challenge was to research opportunities and risks and conceptualize how OpenAI could be integrated into Astrato.

Firstly, I wanted to understand from users where and how this new technology might bring most value and what they care or worry about. One assumption was that users might be concerned about sharing confidential data with OpenAI API.

Also I wanted to understand which prompts might be most useful, as we considered pre-defined prompts to get users started. What would our users be most interested in, how would they intend to use such feature or why they would not use it?

Further improvements were released by OpenAI while working on this feature, which accelerated its potential for assisting users in other areas of the product such as by entirely building a dashboard for them.

Research

Following a first simple Proof of Concept for a chart-based insights feature, I decided to usertest early on by showing different concepts for a dashboard feature to users and stakeholders.

I presented two UI concepts that the PM and me identified as most scalable, testing with Figma prototypes prompt hierarchies, interaction patterns and language options. One concept was a more static dropdown versus a second presented the feature in a personalized chat-like interactive experience.

There was a strong pattern for going with the personalized chat concept that provided scalability as users expect also to freely type their prompt.

Based on the feedback, I was able to create a hierarchy for prompts and moved the language option had to be exposed position.

Insights

After implementation, I validated my designs and conducted both in-person and remote, moderated usertests with five business analysts. II asked them to compare and summarize two charts with an AI feature that they were tasked to find.

I found in those tests that users easily discovered the feature, but that language and advanced options were still too hidden. Also, users expected to click onto a chart to select it for insights and they want to customize the prompt by typing, even though the pre-defined prompts help getting started.

Also, testing revealed a strong preference of users to type their custom prompts, even though the pre-defined prompts were meaningful and helpful.

Even though the feature is easily discoverable, due to familiarity with iconography, a main inight from testing was that selecting objects for insights had to made simpler and more intuitive. Generally, users felt intruiged by the potential of this feature: they felt it can help them design Astrato workbooks and derive insights much faster.

Astrato

Reflections

The project manifests how quickly people embrace new AI technology. This feature has the potential to simplify a tedious and technically challenging dashboard creation process, helping technical and non-technical users kick start their data anlytics journey faster and deliver insights quicker.

As a result from testing, I re-designed the selection pattern for Insights, taking into consideration existing hover states for editing objects. Another iteration looks into auto-generating an entire dashboard based on connected cloud data, which can tremendously simplify the first step to create.

To conclude, using familiar interaction patterns from chat bots helped users navigate this new feature and offers scalability by allowing to write custom prompts. Initial worries about data security are addressed by allowing administrators to enable certain functionalities separately to better control sections that may expose sensitive data.

Further iterations need to now address discoverability of language and text output options. In the near future, I would define metrics to better understand interaction and usage patterns.

Visit Astrato