IMPACT

0→4 paying customers!

8%
8%

adoption rate

+14%

+14%

20%
20%

users asked at least 1 query

+26%

+26%

24%
24%

avg. monthly retention rate

+33%

+33%

COMPANY

LearningClues is a B2B AI startup that sells edtech software to universities.

PRODUCT

StudyClues is an AI chat linked to the LMS that responds with citations to course materials.

Citations to exact lecture moments & sections in course documents

Citations to exact lecture moments & sections in course documents

PROBLEM EXPLORATION

Identifying and prioritizing business, stakeholder, and end-user problems

BUSINESS PROBLEM

Inability to convert pilots into paying customers.

Adoption rate

Weekly active students (% of enrolled)

LearningClues needed 18–22% adoption to convert pilots into paying clients.

STAKEHOLDER ADOPTION PROBLEM

Academic Integrity Concerns

Limited feature set,

strong competition

  • 80% dropoff without first action

  • Less than 1% open popular questions

  • Fear of instructor surveillance

END-USER ADOPTION & USABILITY PROBLEM

First-touch friction and privacy concerns.

END-USER ADOPTION & USABILITY PROBLEM

First-touch friction and privacy concerns.

  • 80% dropoff without first action

  • Less than 1% open popular questions

  • Fear of instructor surveillance

FINAL PROBLEM STATEMENT

StudyClues has low end-user adoption, and usability issues; stakeholders are concerned about academic misconduct and limited features — both are preventing LearningClues from converting pilots into paying customers.

CONSTRAINTS

(1/3) Designing for moving targets

Because we were a lean team trying to move really fast within an early-age startup, I was often designing features as they were actively being developed. To overcome this challenge I consistently communicated progress with engineers to ensure we were all on the same page.

(2/3) User Research Blockers

There were a lot of institutional restrictions around data gathering and talking to students directly, so I was unable to conduct any students interviews. To fill this gap, I relied heavily on competitive analysis to guide my design thinking and process.

(3/3) Accessibility was non-negotiable

No feature could be deployed at any client institution until it passed their specific accessibility requirements. So I incorporated accessibility into the design process from the ground up using the Stark plugin in Figma, with Wave, aXeDev Tools and manual keyboard testing used on deployed builds.

(1/3) Designing for moving targets

Because we were a lean team trying to move really fast within an early-age startup, I was often designing features as they were actively being developed. To overcome this challenge I consistently communicated progress with engineers to ensure we were all on the same page.

(2/3) User Research Blockers

There were a lot of institutional restrictions around data gathering and talking to students directly, so I was unable to conduct any students interviews. To fill this gap, I relied heavily on competitive analysis to guide my design thinking and process.

(3/3) Accessibility was non-negotiable

No feature could be deployed at any client institution until it passed their specific accessibility requirements. So I incorporated accessibility into the design process from the ground up using the Stark plugin in Figma, with Wave, aXeDev Tools and manual keyboard testing used on deployed builds.

IDEATION / ADDRESSING STAKEHOLDER ADOPTION PROBLEMS

  1. Academic Integrity: What happens when students put homework questions in the chat?

How do we implement this coaching persona, then?

Can students explicitly enter the coach mode or is it more of an implicit persona that the bot takes on when it detects homework questions?

DESIGN EXPLORATIONS: COMPETITIVE ANALYSIS

Stand-alone "Tutor Me” chats.

Stand-alone "Tutor Me” chats.

easy to find past chats or start a new tutor chat
too disruptive to open new chat in an existing conversation

easy to find past chats or start a new tutor chat
too disruptive to open new chat in an existing conversation

Students can choose a global AI chat persona.

Students can choose a global AI chat persona.

more customization
decision-fatigue
engineering cost

more customization
decision-fatigue
engineering cost

Switch to a different “search” mode within the same conversation.

Switch to a different “search” mode within the same conversation.

preserves chat context
technically feasible
small learning curve

preserves chat context
technically feasible
small learning curve

SOLUTION / ACADEMIC MISCONDUCT

Enter Coach Mode

  1. Auto-activates implicitly in StudyClues responses if a homework question is detected.

  2. Students can explicitly toggle it on/off and use it as a new conversational feature.

Added as a visible chip based on ChatGPT’s already familiar conversation modes

Stakeholder approved!

IDEATION / ADDRESSING STAKEHOLDER ADOPTION PROBLEMS

  1. Limited Search Experience: How can we broaden our search capability?

BROADENING SEARCH: OPTION #1

Web Search

BROADENING SEARCH: OPTION #1

Web Search

BROADENING SEARCH: OPTION #2

File Collections

Students can create folders of course documents (LMS + individual file uploads) and then chat about them with StudyClues.

BROADENING SEARCH: OPTION #2

File Collections

Students can create folders of course documents (LMS + individual file uploads) and then chat about them with StudyClues.

BROADENING SEARCH: OPTION #3

Multi-course Search

BROADENING SEARCH: OPTION #3

Multi-course Search

EXPLORING TRADEOFFS

Web Search

competitive edge with URL collections
already using OpenAI APIs
too costly with no paying customers

File Collections

more personalized
protects university IP
too big of an engineering lift
increased data storage costs

Multi-course Search

easiest technical & cost lift
existing access to student LMS data
competitively unique
not as broad of a search radius

SOLUTION / BROADENING SEARCH

Enter Multi-course Mode

Stakeholder approved!

IDEATION / ADDRESSING END-USER ADOPTION & USABILITY PROBLEMS

  1. First-touch Friction: How can we prompt students to use StudyClues?

ADDRESSING FIRST-TOUCH FRICTION: OPTION #1

Instructor Publicity & Influence

Design StudyClues-specific class activities to help instructors demo use cases of the tool to students.

ADDRESSING FIRST-TOUCH FRICTION: OPTION #1

Instructor Publicity & Influence

Design StudyClues-specific class activities to help instructors demo use cases of the tool to students.

ADDRESSING FIRST-TOUCH FRICTION: OPTION #2

Onboarding Flow

An Interactive walkthrough of StudyClues: Asking a question and interacting with citations.

ADDRESSING FIRST-TOUCH FRICTION: OPTION #3

Canvas Email Reminders

LMS-driven email reminders synced with major course events.

[...]
How well did you understand CBT this week? Ask StudyClues!

Quiz 2 is coming up. Use StudyClues to refresh your knowledge of Laws of UX.
[...]

ADDRESSING FIRST-TOUCH FRICTION: OPTION #3

Canvas Email Reminders

LMS-driven email reminders synced with major course events.

[...]
How well did you understand CBT this week? Ask StudyClues!

Quiz 2 is coming up. Use StudyClues to refresh your knowledge of Laws of UX.
[...]

ADDRESSING FIRST-TOUCH FRICTION: OPTION #4

Dashboard with Multiple Entry Points

ADDRESSING FIRST-TOUCH FRICTION: OPTION #4

Dashboard with Multiple Entry Points

EXPLORING TRADEOFFS

Instructor Publicity

students heavily influenced
not all instructors willing
extra work for instructor

Onboarding

no external involvement from stakeholders req
easiest to ensure students see it
can be used to address privacy
only a one-time thing

LMS Reminders

Market potential use cases
strategically remind students to keep coming back
universities do not allow direct messaging students

Dashboard

more personalized entry points to capture attention and nudge students into action
can be combined with onb
engineering cost is high
instructors denied lecture highlights concept

SOLUTION / FIRST-TOUCH FRICTION

Enter Walkthrough Onboarding Flow

Engineers said NO :(

A walkthrough onboarding was too invasive for our current codebase.

Enter Modal Onboarding Flow (Iteration #2)

Engineering Approved!

Addresses privacy concerns in step 1

A click-through modal slideshow was easier to maintain long-term and used GIFs to mimic the same interactions.

IDEATION / ADDRESSING END-USER ADOPTION & USABILITY PROBLEMS

  1. How can we improve feature discoverability?

DESIGN ALTERNATIVES

Iteration #1

Popular Questions accordion is open by default

Iteration #12

No Accordion. Popular Questions only appear at the start of each conversation.

saves screen real estate
aligns with past usage trends

SOLUTION / ADDRESSING END-USER ADOPTION & USABILITY PROBLEMS

  1. How can we address privacy concerns?

An up-front privacy-disclaimer as the first non-skippable step of the onboarding flow.

FINAL SOLUTION

Bringing everything together…

FINAL DESIGN

Now for a visual refresh…

RESPONSIVENESS & ACCESSIBILITY

Ensuring designs do not break even when zoomed up to 200%

OUTCOME

By driving a 22% adoption rate for StudyClues, I helped LearningClues go from 0→4 customers and achieve multiple new pilot partnerships!

Pictured alongside Perry Samson, CEO & Co-Founder @ LearningClues, presenting my work at EDTECH WEEK 2025 in NYC

Demo-ing my LearningClues work to faculty and investors at the Michigan Invention Celebration, Nov 2024.

REFLECTIONS

What I would do Differently

Leverage AI to explore more alternatives

How can we make StudyClues more agentic?

Suggesting context-aware follow-up questions?

Administer In-app Surveys

Record qualitative feedback from students through platform surveys to work around stakeholder user research barriers.

No direct evidence to confirm if privacy concerns were fully addressed.

Learning to anticipate and design around stakeholder policy restrictions…

Accessibility compliance can be time-consuming and a deal-breaker in B2B Edtech.

It's so important to design AI agents for trust, agency, and transparency.

Thanks for sticking till the end! :)