← All projects

Case studyArtificial Intelligence3 days

Eklipse.gg: AI Highlights Enhancement

An enhancement plan for Eklipse’s AI highlight feature: more relevant clips, better meme timing, and faster processing for creators.

Key skills
Product Management
  • Product Roadmapping
  • Product Strategy
User & Market Research
  • User Research
  • Product Research
Tools
  • Google Docs
  • Google Sheets
Illustration of a video creator editing footage on a professional setup

Product Management Case Study: Feature Enhancement Plan for Eklipse.gg


1. Feature Selection: AI Highlights

1.1 Reason for Selection

Although themes like App Stability (10), Monetization (10), and Platform Integration (8) received a higher number of user complaints compared to AI Highlights (7), these themes primarily reflect system-level or policy-related limitations. Issues like crashes, paywalls, or third-party account connectivity often require long-term infrastructure or business model adjustments beyond the scope of a single feature enhancement. In contrast, AI Highlights is a core, user-facing feature that can be directly improved by the product and engineering teams, with a tangible effect on user experience and retention.

Based on user reviews on Play Store over the past 3 months, AI Highlights is among the most criticized features (see Appendix A for full theme breakdown). A total of 7 negative reviews cited issues such as:

  • Inaccurate moment detection (e.g., missed jump scares, irrelevant cuts)
  • Irrelevant or mistimed meme overlays
  • Excessive processing times
  • Final clips often deemed irrelevant or unusable

This consistent pattern suggests a clear gap between user expectations and actual performance, especially for streamers looking for faster highlight automation. Addressing this gap presents a practical opportunity to improve the core product experience.

1.2 Why It Matters

AI Highlights plays a central role in Eklipse.gg’s value proposition — automating the time-consuming process of scrubbing hours of footage to find share-worthy content. For many streamers, especially solo creators and small channels, this feature is the key reason they try the platform.

When these shortcomings occur, it directly erodes user trust and satisfaction. Negative feedback on Play Store shows that poor highlight quality not only frustrates users but also discourages them from upgrading to premium or continuing to use the product.

Improving this feature directly impacts:

  • User retention, by delivering more usable clips
  • Perceived value of premium plans, which depend on highlight quality
  • Shareability and growth, as better clips lead to more exposure across platforms like TikTok, YouTube Shorts, and Instagram Reels

In short, enhancing AI Highlights strengthens the core utility of Eklipse.gg and reinforces its position as a reliable, smart assistant for streamers.


2. Metrics Analysis

To evaluate the performance and user experience of the AI Highlights feature, I would focus on a set of quantitative metrics that reflect both system effectiveness and user behavior. These metrics directly align with Eklipse’s business goals of increasing satisfaction, retention, and growth, and are feasible based on the available engagement, feedback, and performance data.

2.1 Key Performance Indicators

Metric What It Measures Why It Matters
Highlight Generation Success Rate The % of highlight requests completed without errors or crashes Indicates technical reliability of the feature. A low rate suggests backend or AI logic failures
Average Processing Time Time from user request to clip delivery Long waits reduce perceived speed and responsiveness, leading to frustration or drop-off
Clip Deletion Rate % of AI-generated clips deleted without being saved or exported High deletion suggests dissatisfaction with content accuracy or usefulness
Clip Engagement Rate % of AI-generated clips that are viewed, downloaded, or shared Helps evaluate the relevance, appeal, and utility of the AI-selected moments
User Retention After Highlight Use % of users who return within 7 and 30 days after using AI Highlights Directly reflects whether the feature drives repeated usage and lasting product value to the user

2.2 Supplementary User Signals

While not quantitative metrics, user feedback can provide valuable context to supplement the knowledge received from the existing metrics:

  • Play Store and App Store reviews that reference AI Highlights-related terms, such as “AI”, “highlights”, or “meme timing”
  • Support tickets and survey responses mapped to known AI Highlight pain points such as inaccurate clips, slow processing, or meme misplacement.

These signals help validate hypotheses generated from behavioral data, but are not used as primary success indicators.

2.3 Application of Metrics

This combination of usage, retention, and performance metrics allows the team to:

  • Identify failure points in the AI highlight generation process
  • Quantify user satisfaction through behavioral data, rather than relying solely on opinions
  • Prioritize improvements that directly impact user retention and clip value
  • Validate post-enhancement success through measurable improvements in processing speed, engagement, and return usage

3. Proposed Enhancement

To address the recurring issues found in recent user reviews and product metrics, I propose a targeted enhancement plan focused on four key areas: moment detection, meme overlay logic, processing time, and content preview.

3.1 Targeted Enhancements

  1. Improve Moment Detection Accuracy

Problem: The AI frequently fails to capture emotionally significant moments such as jump scares, kill streaks, or humorous interactions — especially in horror or narrative-driven games. This leads to low-quality clips that don’t resonate with users.

Enhancement:

  • Fine-tune detection models using genre-specific training data (e.g., detect loud audio spikes in horror, rapid kill sequences in FPS).
  • Incorporate contextual metadata such as game titles, in-game audio intensity, and viewer chat spikes to improve the AI’s ability to recognize highlight-worthy events.
  1. Smarter Meme Overlay & Sound Logic

Problem: Many users report that meme overlays and sound effects are misplaced or irrelevant to the content, often disrupting the emotional tone of the clip.

Enhancement:

  • Train a separate tagging model to detect the emotional or contextual tone of a moment and pair it with fitting meme/sound overlays.
  • Allow users to customize the presence and intensity of memes through a toggle (e.g., None / Minimal / Full) before finalizing the clip.
  1. Processing Time Optimization

Problem: Clip generation can take over 12–24 hours, particularly for long-form streams, making the process slower than manual editing — defeating the core promise of automation.

Enhancement:

  • Prioritize processing based on user type (e.g., premium users get faster turnaround), stream length, and user activity levels.
  • Use adaptive frame sampling to speed up analysis by skipping low-activity segments and focusing only on parts of the stream with significant in-game action or viewer interaction.
  • Introduce early-exit logic to stop processing once a sufficient number of highlight-worthy moments are found to avoid unnecessary computation.
  1. Preview & Feedback Loop

Problem: Many users discard AI-generated clips without using them, indicating dissatisfaction and lack of control over final output quality.

Enhancement:

  • Introduce a lightweight clip rating system (e.g., thumbs up/down, report issue) to gather structured qualitative feedback.
  • Use feedback signals to iteratively retrain the models and personalize future clips based on user preferences.
  • Add a preview screen showing AI-chosen moments side-by-side, allowing users to compare and refine clips before export.

3.2 Problem-Solution Fit

The enhancements directly align with the top issues raised in recent user reviews (see Appendix B). These include poor moment detection, mistimed meme overlays, long processing times, and unusable clips. By targeting these pain points, the solution ensures strong alignment with user needs and delivers improvements that are both impactful and relevant to the Eklipse user base.

3.3 Scope

The proposed improvements are designed to be realistic and achievable within 3 months. They make use of systems that already exist, so the team does not need to rebuild the product from scratch. Each change is meant to create a clear benefit for users without requiring major technical changes behind the scenes.

Specifically, the scope is limited to:

  • Model fine-tuning, not full retraining — leveraging existing detection pipelines with improved training data and tagging strategies.
  • UI-level additions, such as preview screens and meme toggle controls, which can be incrementally added without disrupting core workflows.
  • Backend adjustments like adaptive frame sampling and early-exit logic, applied at the feature layer without reworking the core AI architecture.
  • Lightweight feedback mechanisms, such as in-product clip ratings, that require minimal engineering but provide high learning value.

By focusing on these incremental yet impactful changes, the enhancement plan remains realistic in terms of effort, while still driving clear improvements to user experience and perceived product value.


4. Implementation Plan

To ensure the successful rollout of the proposed enhancements for the AI Highlights feature, the following implementation plan outlines the key steps, timeline, required resources, and mitigation strategies.

4.1 Key Steps and Timeline

The enhancement plan is structured across 6 sprints (approx. 12 weeks), aligning with standard 2-week agile cycles. Each sprint is scoped to balance progress in engineering, design, and validation while avoiding overload.

Sprint Duration Milestones
Sprint 1 Weeks 1–2 • Finalize feature scope and define success metrics
• Collect training datasets for model refinement
• Gather targeted user feedback related to AI Highlights behavior
Sprint 2 Weeks 3–4 • Fine-tune the AI detection model using genre-specific logic
• Begin implementing adaptive frame sampling to reduce processing time
• Implement processing prioritization based on user type (e.g., premium), stream length, and user activity level
Sprint 3 Weeks 5–6 • Develop meme overlay logic based on contextual tagging
• Build user-facing controls for meme customization using a toggle interface
Sprint 4 Weeks 7–8 • Implement the clip rating system for structured user feedback
• Design and build the preview interface for side-by-side clip comparison
• Integrate early-exit logic into the clip generation process
Sprint 5 Weeks 9–10 • Conduct internal QA and resolve high-priority bugs
• Soft-launch to a limited user group
• Monitor real-world performance metrics like processing time and engagement
Sprint 6 Weeks 11–12 • Analyze user feedback and retrain models as needed
• Improve UX polish and finalize refinements
• Roll out the enhanced AI Highlights feature to all users

4.2 Resource Allocation

To deliver the enhancements efficiently, the following cross-functional team and resources are proposed:

Team/Role Responsibilities
Product Manager (1) Define feature scope and success metrics, coordinate sprints, prioritize tasks, and ensure timely delivery across design, engineering, and QA.
AI/ML Engineer (2) Fine-tune detection and meme-tagging models using genre-specific training data; implement adaptive sampling, retraining, and feedback integration logic.
Backend Engineer (1) Implement adaptive frame sampling, early-exit logic, and upgrade the queueing system to support processing prioritization based on user type and stream length.
Frontend Engineer (1) Build the meme toggle UI, preview interface, and feedback rating system. Ensure a smooth user-facing experience throughout the enhancement rollout.
UX/UI Designer (1) Design wireframes for meme control toggles, clip preview screens, and in-product feedback components, ensuring clarity and minimal workflow friction.
QA Tester (1) Test the new AI logic, preview UI, and meme settings across multiple use cases. Coordinate with engineers to identify and report issues during each sprint.
Customer Support (shared) Monitor user reports during the soft-launch phase. Collect qualitative feedback and escalate recurring or critical issues to the product and engineering teams.

Additional tools and resources:

  • GPU resources: Required for AI model training, testing, and deployment.
  • Analytics tools: For tracking clip deletion rates, engagement metrics, highlight success rate, and user retention (e.g., Mixpanel).
  • Survey platform: To collect structured post-launch feedback from soft-launch and public users (e.g., Google Forms).
  • Error monitoring tools: Capture and track runtime errors during clip generation, meme application, or preview rendering (e.g., Firebase Crashlytics).

4.3 Risk Assessment and Mitigation Strategies

Several risks may arise during implementation. Here is a risk assessment with proposed mitigation strategies to ensure smooth delivery and alignment with user expectations.

Risk Likelihood Impact Mitigation
Model retraining delays Medium High Begin model preparation in Sprint 1 and leverage transfer learning to minimize development time
User dissatisfaction with meme toggle UX Medium Medium Conduct lightweight usability testing in Sprint 3–4 to validate design
Longer than expected processing even after optimization Medium High Use Sprint 5 soft-launch data to dynamically adjust queue priority and sampling rates
Preview screen might confuse users or make the process feel harder Low Medium Make the preview optional for users. Track how often the feature is used and improve the design based on user behavior.
Low adoption of feedback/rating system Medium Medium Keep rating system lightweight and tie to small user incentives (e.g., unlock themes)
Misalignment with actual streamer needs Low High Base enhancements on real user reviews and soft-launch validation feedback

5. Suggestions

To ensure a smooth rollout and strong adoption of the enhanced AI Highlights feature, the following approaches are recommended:

5.1 Help Users Discover What’s New

Include in-product guidance to explain the updated features. Tooltips, welcome messages, or short banners can briefly introduce the meme customization toggle, the clip preview interface, and the new clip rating system. These simple cues help users understand the new tools without overwhelming them.

5.2 Make Feedback Rewarding

Encourage users to rate the clips they receive by offering small incentives. These could include badges, unlockable themes, or fun summary stats showing their impact. This turns feedback into something users enjoy doing and builds a sense of contribution to improving the product.

5.3 Communicate the Update Clearly

Announce the improvements with messaging that focuses on user benefits. Use clear phrases like “Better moment detection, faster processing, more editing control.” Show visual comparisons between the old and new versions of AI Highlights. Include quotes or examples from early users to build trust. Share the message through emails, the website, and community channels.

5.4 Keep Improving After Launch

After releasing the feature publicly, continue to monitor how users interact with it. Track metrics like clip deletion rates, engagement levels, and time to process. Run interface tests to explore which layout or controls perform best. Set up monthly team reviews to evaluate user feedback and make model adjustments or UX refinements when needed.

Appendix A – Summary of User Complaints by Theme

No Theme Positive Neutral Negative Grand Total
1 App Stability 0 0 10 10
2 Monetization 0 0 10 10
3 Platform Integration 0 0 8 8
4 AI Highlights 0 0 7 7
5 Editing Quality 0 0 3 3
6 Performance 0 0 3 3
7 App UX 0 1 0 1
8 Content Outcome 1 0 0 1
9 Editing Simplicity 1 0 0 1
10 Processing Time 0 0 1 1
11 Search Functionality 0 0 1 1

Appendix B – AI Highlights Negative Reviews

No User Review Summary
1 Rylan Hobbs Low-quality AI, poor timing of meme overlays
2 xMikaelax Fails to capture funny/jump scare moments, weak highlight detection
3 Andre Whack Too much content in clips, needs re-editing
4 Chris Troutman AI fails to find clips in long videos, slow delivery (12 hrs), poor support
5 Timerra Lamoureux Low quality AI, not worth subscription
6 Luis Menchaca Random cuts, slow processing, unusable clips
7 Marcella Quinones Irrelevant memes, frequent crashes, forced to manually edit