CK-12 Teacher Analytics
Separate prediction from diagnosis. Make uncertainty visible. Help teachers decide where to look next.
- A predicted skill level does not tell a teacher what to do next, and a single score is easy to overread.
- I led the experience architecture, interaction model, and visual language: predicted ranges instead of a single score, a clear no-data state, and Insights, which separates skill from engagement so a teacher can check one student before acting.
- In CK-12’s published reviews with ten teachers each, 90% found each tool’s menu easy to interpret and 90% or more expected it to save time. Half struggled with each tool’s main chart. We answered with Recommendations, which shipped: one action button per insight, so the chart became supporting detail.
On this page
Before an assignment, a teacher needs to know who may need support. After it, what to do next. One score cannot answer both.
In CK-12's focus groups(opens in a new tab), most of the eight teachers asked for class trends on upcoming content. Six cited a lack of transparency in AI assessment as a reason for distrust.
My goal: make the model's signals useful without making them look more certain than they were.
As Lead UX Designer, I owned the experience architecture, interaction model, prototypes, and visual language for both tools. I started on Insights in 2020 and on Foresights in 2021. Both were live for teachers by May 2023, when CK-12 published its introduction video(opens in a new tab).
I worked with four groups, and each owned a different question. Educators told us what the data meant in a classroom. Product set scope and priorities. Data Science decided what the model could responsibly claim. Engineering owned feasibility and delivery. I turned classroom needs and model limits into designs we could test, then worked with Engineering through delivery.
I framed the work around three questions a teacher moves through: what might happen, what happened, and what to do next. Three rules decided close calls: lead with the takeaway and show model detail on demand; show why a signal exists and never present it as fact; suggest a next step and let the teacher decide.
I did not run the research. CK-12's research team ran the focus groups and the two teacher studies below.
One performance score is easy to scan but blurs two jobs: planning and investigation. I kept Foresights and Insights apart so a prediction does not pose as an explanation.
Two moments, connected
- Before: anticipate supportForesights
- Students practice
- After: investigate evidenceInsights
- Teacher chooses the next action
A single score was simple but misleading. A raw probability was honest but hard to read between classes. I chose the skill levels teachers already use, from Beginning to Mastery, and showed each student as a range on that scale (see the demo-class chart at the top). But teachers still need to know what a range lets them conclude.
Low performance can come with different patterns of effort, and one score hides that. A student who works hard but is stuck needs a different move than one who understands the concept with little activity. Insights separates skill from engagement, so a teacher can spot a class pattern, then check one student before acting. Selecting a student opens the evidence behind the dot and ends in a suggested next step, which the teacher accepts or ignores.
Evidence is easier to act on when it maps to one concept. Assignments that span several concepts blur the result. Chart styling cannot fix that. CK-12 later moved to one concept per assignment and named better Foresights and Insights as a benefit. That platform decision was not mine alone.
From content to a decision
- Concept
- Assignment
- Evidence
- Teacher interpretation
- Next action
Each CK-12 study used ten teachers and a demo class. They measured perceived usefulness and comprehension, not learning gains or time saved.
| Reported measure | Foresights study(opens in a new tab) | Insights study(opens in a new tab) |
|---|---|---|
| Found the menu easy to interpret | 90% | 90% |
| Found it meaningful | 70% | 90% |
| Anticipated time savings | 90% | 100% |
| Would recommend it to a colleague | 70% | 90% |
| Mean comprehension score | 75% | 73% |
| Had difficulty with the key graph | 50% | 50% |
Graph difficulty offsets the usefulness ratings. A valuable concept can still demand too much interpretation.
Half the teachers struggled with the main skill chart. We did not answer that with a better chart. We used the findings to build Recommendations, and it shipped to teachers: one action button for each insight, with a next step the system recommends. For example: "8 of your students struggled on this part. We recommend assigning this remedial content."
The teacher can act in one click. The skill chart becomes supporting detail, not the thing a teacher has to decode first.
For the student side, see Flexi: helping students get unstuck.