From Data Overload to Action-driven Educator Dashboard
Teachers had a dashboard full of analytics that never told them what to do next. Across five educator interviews the same gap kept surfacing: the charts showed student performance, not what to do about it.
I led UX on the 4-week client build, then took it back solo and reframed it into one decision flow: spot the students falling behind, then send them support without leaving the screen.
Built to visualize data, not to act on it
NOTE: To comply with the NDA, the company name and specific confidential details have been omitted from this case study. Please feel free to reach out and I'd be more than happy to chat about it privately.
The client is an AI learning platform that turns scattered study materials, such as notes, slides, PDFs, and AI chats, into structured study plans in seconds. Our team was brought in on a 4-week engagement to design its educator-facing dashboard: the screen where teachers see how students are actually doing. I led UX; another designer led UI.
The brief was a feature list and a PRD. No user stories, no personas, no success metrics, so every UI decision carried more guesswork than it should have, and we shipped to spec anyway. Then usability testing started returning more questions than answers, the sharpest one from a teacher mid-interview: “What am I supposed to do after seeing this?” Once the engagement wrapped, I took the redesign back solo.
Here's a quick look at the dashboard home, before and after the redesign.
Charts ended where decisions began
Every widget reported performance. Not one of them named the next step, and reading them cost time teachers didn't have.
Grades got the spotlight, growth stayed invisible
Snapshots emphasized final scores over trajectory, so the story of how a student got there never surfaced.
Signals arrived too late to act on
Struggling students surfaced only after the exam, with the context scattered across other platforms.
How these problems appeared in the product:
The gap
Teachers were not short on data. They were short on clarity, and every hour spent decoding a chart is an hour not spent on the student the chart is about.
Prioritizing actions over data visualization could go a long way, for educators and students both.
Surfacing action suggestions teachers can take in one workflow would save time, and create better moments of feedback for the students who need them.
This framing intentionally constrained the solution
We were not trying to add AI for its own sake, but to validate whether AI-supported feedback, ending in remedial work a student can actually do, could support teachers without adding friction or time to their workflow. Action-driven, not another layer of data visualization.
Deciding what ships
I dug back into our interviews with 5 educators: one elementary teacher, one high-school teacher, and three professors, and re-sorted all 14 interview observations until 4 patterns held, and each pattern re-drew a piece of what the dashboard should even be.
The mental load isn’t the teaching. It’s moving between tools that don’t talk to each other.
Time-on-app says nothing about learning. Teachers read sloppiness, skipped details, and the quiet middle.
What teachers put into grading and what students take out of it don’t match.
The fear isn’t AI. It’s AI skipping the struggle that learning needs.
One insight reframed the problem
Four patterns, one shape underneath: every one of them is a teacher doing manual work, spending brain power that could've gone into taking action for students. Teachers weren't lacking data, they were lacking clarity. So the problem was never “teachers need better charts.” Teachers need one connected workflow that closes the distance between noticing a struggling student and acting on it.
I mapped 4 teacher scenarios against the feature list, expecting to find priorities. Instead the map kept closing on itself: every scenario ended where another began. A teacher creates content, assigns it, tracks progress, identifies gaps, and acts.
Create → Assign → Track → Identify → Act
The feature list was one loop, and the dashboard owns the Identify → Act handoff, the moment a teacher goes from “I see a problem” to “I’ve responded.”
That handoff carries the two problems educators most hoped the dashboard would solve:
Spot who's slipping, early
Surface the students who need help while there is still time to act, not after the exam.
Act without leaving the screen
Turn a noticed gap into sent support in one flow, instead of stitching tools together.
Features like exporting, parent reports, and operational utilities were deliberately set aside. That choice has a real cost: a teacher still leaves the dashboard to build a parent report. I accepted it to keep the loop whole.
Performance filtering, by need
Filter student performance by the question being asked: a class view for “how are my classes doing,” an assignment view for “what happened to the thing I just sent.”
Action-driven AI feedback
AI drafts the outreach, from nudge messages in a friendly or formal tone to suggested remedial content bundled into a study plan, and the teacher reviews before anything sends.
Feedback they can act on
Remedial work arrives as a sequenced study plan with time estimates: something to do, not a comment to skim.
Reached while there's still time
Nudges arrive before the deadline passes, with the assignment context right in the message, so course-correcting is still possible.
This framing pushed us to design a model
One where “feedback on student performance, and the action that follows” is an object with its own lifecycle (Create → Assign → Track → Identify → Act), instead of a time-consuming gap between receiving an assignment and giving feedback. Narrower, but easier for instructors to audit and trust.
One connected decision flow
The redesign puts the whole teaching loop on one spine, and the dashboard owns its last three steps.
Three decisions carried the redesign. Each one pairs the reasoning with the screen it produced, before and after, and names the cost I accepted along the way.
Decision 1: two-step filtering tabs
Teachers check two things: how is my class doing, and what happened to the thing I just sent? One tab each. Nothing else earned a tab. It's David's ask made literal: automation that fits the workflow he already runs.
- Filter by Class: every assignment tied to a class, with cohort actions like exporting progress reports.
- Filter by Assignment: an instant who-what-when snapshot of a just-sent task across every class.
Exporting, parent reports, and operational utilities were deliberately set aside; a teacher still leaves the dashboard to build a parent report. I accepted that cost to keep the loop whole.
Decision 2: a student list that acts
Teachers send reminders to groups like “have not yet submitted” or “absent.” Recipients select themselves, so a nudge costs one tap instead of a scroll through the roster. For Maggie, with 150 students and 15 hours of weekly feedback debt, the unit of value is minutes.
The per-assignment student list
The Problem
A passive status report that forces teachers to act on each student one by one, slow, fragmented, and context-free.
Solution
Checkboxes, filters, and per-row controls let teachers group outliers and send a nudge or study plan without ever leaving the list.
Self-selecting groups only see what the system tracks. A student who submitted on time but is quietly struggling never lands in a nudge group, so the student list keeps per-row controls for exactly those cases.
Decision 3: remedial work you can steer
The system suggests next steps, like remedial work for students across classes who struggled with the same lesson, and teachers steer the order and delivery before anything is sent. It's David's other interview line made real: AI that prompts deeper thinking without handing students finished answers.
The Create Study Plan flow
The Problem
A general text-based remedial suggestion that teachers had to approve with little to review.
Solution
Sequenced activity types and time estimates are visible before anything is sent, so teachers can review and steer the content rather than approving it blindly.
The automation deliberately stops short of auto-send. Reviewing a plan is slower than one-click approval, but it keeps the teacher's judgment in the loop, which is the one thing David said he wouldn't trade.
Everyone is adding AI to the classroom. Almost no one closes the loop.
I mapped five products circling the same teachers. Tutors, content tools, and tool libraries are everywhere, and features like these are easy for LMS vendors and big tech to copy. The weaknesses read differently: signals live in one place, actions in another.
five products, one shared blind spot: the distance between noticing a student and acting on it.
Where this redesign fits
Every opportunity note on my board points the same direction: deeper analytics, class-level views, intervention workflows. Features get copied; a workflow that closes the notice-to-act gap is harder to bolt on. That connected loop is the position this dashboard stakes out.
What I'd build next, and what stays with me
what could be improved
- Let teachers customize how classes and courses are grouped, so the dashboard fits their own structure instead of a fixed one.
- A future iteration would replace manual content edits with a lightweight axis UI: adjust intensity (difficulty level, depth) and learning dimension (focus areas, topic scope), and the AI regenerates accordingly. Less authoring, more calibrating.
what i learnt
- The best education tools don't do more, they surface the right thing at the right moment. At classroom scale that is a genuinely hard problem, and this reframe was one structural attempt at it: not a change to how students learn, but to when feedback reaches them.
- Visibility matters more than volume. Hierarchy is a decision-making tool before it is a visual one, and choosing what a teacher sees first is the actual work.
That's the end of this story. If you're still curious, here are two more.