A Second Start for Streaming
Redesigning how viewers choose and watch videos on Bazaar, with limited data and challenges left by a fast launch.
Translation note: All original project content was in Persian and has been translated into English for this case study.
Initial product launch
In 2024, I designed the relaunch of Bazaar Video, the movie and TV service within Cafe Bazaar, an Android app store in Iran. To meet a tight deadline, we reused designs, code, and components from the service five years earlier. With product, I limited the scope to browsing, choosing a title, and playback, including season and episode selection. I designed the end-to-end journey across Discovery, Detail, and the player.
Reuse made launch possible, but mixing old and new parts left Figma and code without a coherent structure for further development. The product also lacked alignment with Bazaar’s new design system. The decision to run the web version (PWA) inside the native app created an opportunity to restructure the design and adopt the new system.
Player redesign under a tight deadline
We started with the player in mid-2025. Without time for dedicated research, I relied on benchmarking major platforms and existing heatmaps. With the product manager and frontend engineer, I standardised the full player flow, made it responsive, and aligned it with the design system.
On mobile, I kept large, central controls for easy tapping. On desktop, I placed controls at the bottom and settings in menus for mouse use. The designs were handed off and shipped, but the player’s effect on user behaviour was not measured separately.
Video Detail redesign with limited data
Next came the page for choosing movies and series. I had more time to investigate, but the incomplete initial implementation had left gaps in data. I conducted usability testing and reviewed the findings alongside heatmaps, benchmarking, and a heuristic review.
If I could revisit this project with the necessary tracking in place, I would measure time to start watching, successful playback start rate from Detail, and exit rate without watching to establish a baseline before the redesign.
Findings and design decisions
1. Dead clicks on the cover highlighted the role of imagery
Of the page’s two covers, the portrait attracted many dead clicks. In the similar interface of Myket (a competing Iranian Android app store), that cover started playback. I had two hypotheses: viewers expected playback based on learned behaviour, or wanted to inspect the image.
In testing, five of six participants used posters or actors to choose, and four read descriptions, echoing earlier interviews with active viewers. This supported the importance of imagery, but did not establish Myket as the source of the clicks. I made imagery central to my explorations and designed the header to display it prominently across screen sizes.
2. I prioritised the scattered information
The heuristic review found information scattered below the title, in tiles, and at the bottom. The tiles also took up space and attracted dead clicks by looking interactive.
Drawing on benchmarking and usability findings, we placed essential decision-making information at the top and moved the rest to More Details. I replaced the tiles with simple, non-interactive tags and handed off the responsive page. Trailers and genre exploration remained out of scope.
Improving Continue Watching
The first release followed familiar streaming patterns but represented viewing state poorly: missing or duplicate titles, completed films still visible, and unclear time information and entry rules.
More resumes, without a significant increase in watch time
The first experiment showed resume rates of 33% with Continue Watching and 26% in the control group, without a statistically significant difference in daily watch time. Viewers without the row also resumed through Search or Discovery. Multi-session viewing accounted for 44% of sessions and half of watch time: the need existed, but unfinished viewing needed a more reliable representation.
The Solutions
With product, data, and engineering, I defined the second release’s rules:
- Entry into Continue Watching: actual playback of 5% of the content; seeking forward did not qualify.
- Exit from Continue Watching: reaching 95% of the timeline, 90 days of inactivity, manual removal, or unavailable content.
- State across the product: progress indicators and actions such as Continue S2 E3 across Discovery, Detail, and the player.
- Reliable labels: series name with season and episode numbers instead of unreliable episode titles; less descriptive, but more dependable.
Without reliable intro and credits timing, the 5% and 95% thresholds were provisional, implementable rules to revisit with better data.
Project outcome
The player and Continue Watching shipped. Detail reached handoff, but changes in team structure and company requirements prevented launch; its user and business impact remained unmeasured.
After the second Continue Watching release, more viewers picked up where they left off, more of them finished what they resumed, and slightly more came back to VOD within a week.
- Successful resume rate37%from 33%
- Completion of resumed titles43%from 40.5%
- Seven-day VOD return43.2%from 42.6%
What I learned
I learned that a missing baseline should not stop decisions: use the evidence available, test with users, and revise, while distinguishing observations, hypotheses, and measured outcomes. If I returned, I would propose product experiments earlier. Had I stayed and launch become possible, I would have evaluated Detail through an A/B test using playback starts, time to playback, and exits without watching.