Providing Autonomy in Slide Creation by Rethinking Human AI Interactions
Sep 2023 - Dec 2023
Dr.Lambda is an AI-driven platform that enhances content creation with advanced multi-modal slide-building tools.
Product design intern working with 2 designers and product team.
I end-to-end designed the new AI-powered features, prioritized features with stakeholders, created wireframes and interactive prototypes, and prepared detailed documentation for seamless handoff to the development team.
Impact
+19%
+40%
User problem
User wants to spend as little in time, effort, and money to create better slides.
Business Goal
The product faces challenges in converting premium users
Solution
Customize your slide deck in seconds with AI
2.1 Elemental Customization
2.2 Mass Customization
Grammarly but for you presentation slides
2.3 AI Review Agent
So, how did I get here? It all started with...
Research — Summary
Crafting aesthetic slides is extremely time consuming
Research — Problem Statement
How might we convert users to the premium version by saving them time in aesthetics and detail editing?
Design Principles
User are more likely to pay for productivity product that is
🤔
Intuitive to use
Designed to kept users' work at the center and prepare for rapid iteration without disrupting established patterns.
🚀
Empowering to use
Designed to empower users to interact with AI tools effectively, aiming to increasing quality and reducing time.
🥳
Rewarding to use
Designed to make users feel they've found the right tool, fostering loyalty, leading to increased conversions.
Ideation — Approaches and goals
Problem
Approach
Success Metrics
Rapid Generation
Choosing the right form of visual communication media takes too much time and disrupts the user's workflow.
AI function that generates and modifies visual content based on the presentation context and user commands.
Reduced Time Spent
User Satisfaction Generated Content
Workflow Continuity
AI Mass Edits
Repetitive edits are required when making changes across different pages wasting time and causes emotional distress
Mass editing AI that responds to textual commands to apply changes across multiple elements.
Reduced Repetitive Edits
Efficiency of Editing Commands
User Satisfaction
Ideation — Feature implementation
User are more likely to pay for productivity product that is
Iteration 1: Convenience — Crowded & Outdated WorkSpace
User needs more horizontal screen real-estate for convenience.
As 16:9 decks gain popularity, users are working with wider, shorter slides. Current interfaces, designed for the 4:3 format, feel unbalanced—featuring overly complex toolbars for everyday users and inefficient use of space on the sides and bottom.
Iteration 1: Convenience — Decluttering the interface
Clearing out the stage to let your work shines for convenience.
To enhance slide focus and minimize clutter, we optimized for 16:9 by clearing horizontal space. Non-editing commands were moved to the top, while the dense, frequently used editing toolbar was shifted to the bottom. This streamlined layout prioritizes key tools and improves user workflow.
Iteration 1: Convenience — Designed to scale
Designing for future scalability without disrupting user habitt.
To minimize bottom toolbar content, we moved editing actions to a dynamic toolkit that appears with element selection, cleaning up the interface while maintaining ease of use. This approach also supports scalability, allowing the toolbar to adapt as new features are added.
😫
“This tool is convenient and useful, but the AI output is not ideal"
Iteration 2: Autonomy — Back to research
Users seem to believe AI is magic that fixes everything.
Ideal Scenario — Rapid iteration with generative AI
Unfortunately, generative AI makes mistakes all the time.
Reality — Repeated re-generation loop
Which leads to
😫
User Frustration
Users became extremely upset when unable to generate content, given their high expectations. This quickly led to a loss of trust in the product.
⌛
Long Generation Time
Visual content took significantly longer to generate compared to text content. Users were unwilling to wait more than 1-2 seconds, leading to dissatisfaction.
💸
Loss for Business
Each content generation incurs a cost for the business. In our subscription-based model, repeated generation attempts become extremely costly.
Iteration 2: Autonomy — Enhanced control via additional input
User needs more control over the generated content.
Adding one extra step in generation workflow.
😊
Improve Satisfaction
With refined instructions, AI is able to better understands user needs, resulting in higher satisfaction with the generated content within the first 3 tries.
🕰️
Willingness to Wait
Despite users spending more time they receiving better results. The perceived wait time also appear to be shorter for them, and willing to wait longer.
📉
Decrease Operating Cost
Increased user satisfaction has led to fewer generation attempts avoiding the regeneration loop . This decreased operating costs in by over 50%.
Iteration 2: Autonomy — Balanced control and automation
Users needs suggested choices to make fast decisions.
Balancing user autonomy and AI automation with AI suggestions.
Testing revealed that task speed improves when AI provides snippets, especially for visuals, while users prefer final control. In the mass editing agent, users give text commands, and AI suggests 2-4 options with hover previews. This enhances autonomy, speeds tasks, and reduces repetitive AI use for better outcomes.
😵💫
“ I felt I saved time on making slides, but it just doesn't look as nice as I wanted to. I don't know where it went wrong!"
Iteration 3: Empowerment — Back to research
User tries their best, but still creates mediocre slides.
Iteration 3: Empowerment — Ideation
Problem
Approach
Success Metrics
Rapid Generation
Users cannot achieve the ideal visual quality due to a gap between their skills and expectations.
A "Grammarly" for slide decks that identifies areas for improvement and proposes solutions.
User Confidence in Visual Presentation
Reduced Time Spent on Polishing
Let an expert AI assist in reviewing your slides.
Iteration 3: Empowerment — Solution
Smart review, and smarter edits.
📘
Closing Knowledge Gaps
AI agent offers personalized suggestions and , helping users improve slide design beyond their current knowledge to improve overall productivity.
⚡
Accelerating Workflow
AI agent provides targeted feedback which reduces the need for trial and error, significantly cutting down the time spent on slide adjustments.
💡
Enhanced Confidence
User often felt more confident about their slides post usage, leading to more informed decisions, reduced hesitation, and overall improvement to qulaity.
Iteration 3: Rewarding to use
Rewarding product leads to returning long-term subscribers.
Engaging user for longterm conversion
In a competitive market, attracting users is just the start—retention is key. To convert first-time users into loyal subscribers, the product must be engaging and rewarding. In the mass editing agent design, users input text commands while AI suggests 2-4 modification options with hover previews. This enhances user autonomy, speeds up tasks, and reduces repetitive AI usage, leading to a more satisfying experience that encourages users to return.
Retrospectives
Embracing Chaos in the Design Process
While I started with the Double Diamond and Design Thinking frameworks, I quickly realized that real-world design processes are often more chaotic. Research, ideation, and prototyping happen throughout the project, not just at the beginning. Being flexible and adaptive is crucial to avoiding wasted resources and staying effective in a constantly changing environment.
User Testing & Behavioral Observation : Unlocking Better Insights
Working under tight timelines often meant relying on guesswork, making user testing essential. Observing both verbal feedback and subtle behavioral cues—like users triple-clicking out of frustration—revealed critical insights into the emotional journey of users. These observations helped me design experiences that truly resonate with users on a deeper level.
Design could be the true differentiator in AI-driven products' market adoption
When I first entered the industry, I underestimated how much design could influence a project. Faced with challenges like slow and sometimes inaccurate AI model generation, I discovered that design could improve user interaction and satisfaction, even in areas traditionally seen as engineering problems. This experience taught me to always seek creative ways to tackle seemingly insurmountable issues.
