MY ROLE
THE TEAM
THE TIMELINE
THE PROBLEM
THE SOLUTION
THE OUTCOMES
Studocu’s strongest asset was hidden
Studocu is a global study platform with more than 50 million student-contributed documents. In 2025, they made the strategic decision of launching Studocu AI, a product that allows students to turn their study materials into summaries, quizzes and mock exams.
The two experiences weren’t properly connected. Students often started an AI project without adding materials, leaving the system without context and themselves without an obvious next step. Meanwhile, Studocu’s existing library frequently contained exactly what they needed.
How might we anticipate students’ needs while keeping them in control of what enters their project?
One sidebar, two ways to add context
We redesigned the library panel to give students two complementary ways to enrich their projects: search when they knew what they needed, and recommendations when they didn’t. Recommendations used the student’s course and existing project materials as the most reliable signals, while leaving students in control of what was added.
Search when students know. Recommend when they don’t.
Only 1.8% of users added documents from Studocu to their projects. Hotjar recordings showed that reaching search required several steps, creating avoidable friction. This first opportunity was straightforward: make searching for documents easier to reach, reducing clicks from 3 to 1.
But what if the student didn’t know what to search for?
Together with the PM and EM, I explored proactive recommendations that could help students start with relevant context. Our hypothesis was: surfacing useful documents would lead to more context-rich projects, and ultimately better study sessions.
We explored a broad set of user and project signals, but prioritised the most reliable context available for the first version: the student’s course and the materials already in their project.
1) One sidebar, two ways to add context
We redesigned the library panel to give students two complementary ways to enrich their projects: search when they knew what they needed, and recommendations when they didn’t. Recommendations used the student’s course and existing project materials as the most reliable signals, while leaving students in control of what was added.
AI helped me explore faster, not decide for me.
2) AI shouldn’t take over
The system could anticipate what might be useful, but it never added content automatically. Recommendations remained suggestions: students reviewed the documents and decided what belonged in their project.
3) When context is missing, ask for it.
More context-rich projects, without harming core behaviour
+37.9%
Studocu documents added to AI projects
21.8K treatment / 15.81K control
+27.4%
Unique users adding Studocu documents
8,580 treatment / 6,736 control
+14.7%
Studocu documents added through Search
17.43K treatment / 15.19K control
No statistically significant change in retention, activation, Core usage, trials or payments.
The test demonstrated stronger library discovery and project enrichment. Whether richer projects subsequently improve learning outcomes, retention or paid conversion remains the next question…
What's next?
While the initial results are successful, I would still want to explore:
Does adding more relevant context lead to better study sessions and repeat usage?
How should recommendations evolve as the system learns more about the student and their project?
There’s more behind the final flow: discarded directions, recommendation logic and technical trade-offs.
I’d be happy to unpack them in conversation
![](data:image/svg+xml,<svg display="block" id="548940053" role="presentation" viewBox="0 0 50 22" xmlns="http://www.w3.org/2000/svg"><path d="M 0.08 6.912 C -0.21 7.464 0.344 8.125 0.808 8.467 C 6.44 12.63 13.535 14.124 20.01 13.295 C 25.114 12.642 29.277 10.139 33.379 6.909 C 33.976 6.439 34.611 5.889 35.23 5.352 C 35.646 4.993 36.054 4.638 36.44 4.319 C 37.456 3.477 38.448 2.748 39.473 2.274 C 39.984 2.038 40.238 1.355 40.041 0.75 C 39.845 0.144 39.271 -0.155 38.761 0.081 C 37.513 0.658 36.367 1.515 35.31 2.391 C 34.842 2.779 34.411 3.153 33.994 3.516 C 33.435 4.009 32.866 4.487 32.288 4.949 C 28.306 8.084 24.456 10.361 19.806 10.955 C 13.699 11.736 6.688 10.113 1.572 6.332 C 0.894 5.831 0.371 6.359 0.08 6.912 Z" fill="rgb(239, 196, 107)" height="13.514905960545759px" id="hqoyr1Nz6" transform="translate(4.849 6.126) rotate(3 20.054 6.757)" width="40.10788288257481px"/><path d="M 1.241 1.664 C 1.241 1.659 1.527 1.664 2.299 1.481 L 8.39 0.032 C 8.781 -0.065 9.193 0.062 9.47 0.364 C 9.746 0.667 9.844 1.099 9.727 1.495 C 9.609 1.892 9.293 2.192 8.9 2.281 L 3.454 3.576 C 4.864 5.554 5.553 7.669 6.088 10.025 C 6.229 10.646 5.856 11.266 5.254 11.409 C 4.651 11.553 4.049 11.166 3.908 10.545 C 3.301 7.874 2.55 5.829 0.84 3.928 L 0.836 3.927 C 0.733 3.899 0.635 3.856 0.543 3.8 L 0.541 3.799 C 0.381 3.7 0.248 3.562 0.153 3.396 C 0.063 3.178 0.012 2.946 0 2.709 C 0.053 2.469 0.149 2.242 0.282 2.037 C 0.364 1.953 0.522 1.83 0.593 1.788 C 0.653 1.758 0.715 1.732 0.778 1.709 C 0.866 1.681 0.957 1.664 1.049 1.659 L 1.051 1.659 C 1.119 1.654 1.173 1.658 1.188 1.659 Z" fill="rgb(239, 196, 107)" height="11.439201646916565px" id="ulurXoEpt" transform="translate(36.662 3.217) rotate(113 4.888 5.72)" width="9.775480302533598px"/></svg>)