Empowering educators with AI-powered insights
I designed an AI-powered feature in 2 weeks.
Insights generates in-depth summaries on students based on their data, allowing educators to quickly synthesize the most valuable takeaways to support their students.
Role
Product Designer
Timeline
2 weeks for design, 8 weeks in total to ship
Team
1 Product Manager, 3-4 Engineers, 1 Engineering Manager
Impact & outcomes
Educators were spending time analyzing the data on each student profile, an information-heavy page.
The Student Profile, the most highly used page on Student Success, is loaded with academic, attendance, behavioral, and other data — but it takes effort and time to figure out what it all means.
The Challenge
With so many data points to look at, how do we make it easier for educators to arrive at the big picture?
AI had the opportunity to make this process easier by surfacing insights to educators.
By doing some of the analysis work for educators, they can spend more time supporting students using data-informed insights rather than trying to make meaning from the many data points available.
When, where, and how should insights appear?
I focused on designing according to evidence-based education principles while providing truly valuable information that educators could use to take action.
General insights, above the fold
Insights never would be missed, highlighting the TL;DR every time a user visits the page. This approach acknowledges the student's experience holistically.
But is this intrusive, potentially unwanted? Do educators prefer general summaries over pinpointing specific challenge areas?
Category-specific insights
Granularity into specific sections might allow educators to better understand a specific area of concern, such as attendance patterns.
Would the information feel disparate if looked at separately? Would the discoverability be a challenge? How do we offer specificity without brushing over how different areas may be connected?
Auto-generated vs. driven by user consent
Since this was the first time Panorama was introducing AI-powered features (beta release, if I might add) into Student Success, I wanted to include a level of consent and comfort for end users.
From a technical perspective, it would also allow the team to process only the prompts that users wanted — saving costs and technical effort.
How should prompting drive the surfaced insights?
After understanding Engineering capabilities and context, I suggested several directions to drive prompting. Engineering, PM, Data Science, and Design iterated on the prompt to ensure that it met the needs and perspectives from all disciplines.
High level sections
Adding subheaders for each bullet point allowed the information to be more skimmable, reducing cognitive load and ultimately focus on key topics.
Avoid binaries & forced categorization
Requiring strengths and challenges to be called out for every summary implies binary perspectives on students, straying away from the whole-child philosophy we had at Panorama.
Why not let AI surface the most compelling points about a student's profile that most represents their story, encouraging their strengths to be highlighted?
Limit the number of insights
Overwhelming educators with too much information was the problem we were trying to solve, not perpetuate.
I decided to feature 3 key insights so that the student profile summary is able to retain its usefuness without scattering focus.
Designing with speed forced me to keep things simple, yet keep the user need at the core of my process.
Designing live alongside Engineering allowed for collaborative innovation
In ambiguous territory, my cross functional partners and I leaned on each other to share ideas, perspectives, and limitations to arrive at the best approach.
Instead of doing detailed mockups before discussing with Engineering, we opted to prototype live together — bypassing extra steps.
Know when to follow AI UI patterns and when to diverge slightly
While AI summaries tend to be surfaced without user action, the beta release of this feature required more intention and slow introduction of AI to our users.
Treating popular patterns as a suggestion and not a rule allowed me to design solutions that fit the nuance and sensitivity of student data.