Mapping the Spectrum of Engagement
The most consequential users on a platform are not always the most visible ones. The more consequential question is not who they are. It is why the strategy being used to reach them hasn't been built to find them.
TL;DR
Overview
Watch Streaks was a Twitch retention initiative built around extending engagement recognition to viewers who were already returning but had never had a feature designed around how they participated. On this project I owned the research function, defining the research questions, the experiment design, and the methodological limits before the data came in. The existing retention strategy was optimized for visible participation, and I recognized early that a successful launch built around that assumption would produce a short-term lift, not a retention outcome.
My accessibility background had trained me to ask who a system was not designed for before asking what it should do next. On this platform, that question pointed to lurkers, the second highest segment by return rate and the one the existing retention strategy had no design language for.
Business Problem
Retention had surged during the pandemic and was declining as behavior normalized, and engagement was one of several levers tied to revenue. The existing strategy was concentrated on active, visible participants, the segment most likely to produce a measurable short-term lift. The data being used to address the retention problem tracked visible behavior only, which made an entire class of returning behavior structurally invisible to the platform's measurement model. A successful launch and a meaningful retention outcome were not the same thing.
Research Approach
I proposed this study, which was not on the roadmap. This was a behavioral choice experiment run as a live in-product test, designed to surface what people did while leaving the question of why open for qualitative follow-on. Before the data came in, I named the limit to the team directly. Behavioral data could surface what people chose, not why they chose it. The case I made connected lurker behavior to engagement and micro-transactions, positioning the study as a platform opportunity, not a feature follow-on. I got the qualitative study prioritized on the roadmap.
What the Findings Surfaced
The behavioral data gave the lurker segment a measurable profile for the first time.
- Lurkers drove 54% of milestone activity, outperforming the segment the platform had been designing for
- 1.75% Five-Minute Play (the share of viewers who stayed in a stream for five or more minutes) increase within the experiment cohort
- 94% overall engagement increase on initial launch, measured across the first 2-3 months post-launch
The more structural finding was in the measurement model itself. The platform's existing engagement metrics tracked visible behavior only, which made ambient participation absent from the data being used to address the retention problem. The lurker segment was not absent from the platform. It was absent from the measurement model.
Impact and Forward Motion
Before this work, Watch Streaks was built for one type of viewer, the one willing to participate publicly. The private sharing flow exists because the research pitch named a behavioral gap the original scope had no mechanism to detect. The team's frame shifted from engagement as a single visible mode to a spectrum of participation, including the ambient behavior the measurement model had not previously been built to surface. Before this work, lurker behavior had no dedicated research investment on the platform. The qualitative study now on the roadmap marks the first time the platform has resourced ambient participation as a platform-scope question in its own right.
“Expanding perspectives into deliberate insight for bolder moves”
Made and designed by nayeri
© 2026 All Rights Reserved
“Expanding perspectives into deliberate insight for bolder moves”
Made and designed by nayeri
© 2026 All Rights Reserved
Mapping the Spectrum of Engagement
The most consequential users on a platform are not always the most visible ones. The more consequential question is not who they are. It is why the strategy being used to reach them hasn't been built to find them.
TL;DR
Overview
Watch Streaks was a Twitch retention initiative built around extending engagement recognition to viewers who were already returning but had never had a feature designed around how they participated. On this project I owned the research function, defining the research questions, the experiment design, and the methodological limits before the data came in. The existing retention strategy was optimized for visible participation, and I recognized early that a successful launch built around that assumption would produce a short-term lift, not a retention outcome.
My accessibility background had trained me to ask who a system was not designed for before asking what it should do next. On this platform, that question pointed to lurkers, the second highest segment by return rate and the one the existing retention strategy had no design language for.
Business Problem
Retention had surged during the pandemic and was declining as behavior normalized, and engagement was one of several levers tied to revenue. The existing strategy was concentrated on active, visible participants, the segment most likely to produce a measurable short-term lift. The data being used to address the retention problem tracked visible behavior only, which made an entire class of returning behavior structurally invisible to the platform's measurement model. A successful launch and a meaningful retention outcome were not the same thing.
Research Approach
I proposed this study, which was not on the roadmap. This was a behavioral choice experiment run as a live in-product test, designed to surface what people did while leaving the question of why open for qualitative follow-on. Before the data came in, I named the limit to the team directly. Behavioral data could surface what people chose, not why they chose it. The case I made connected lurker behavior to engagement and micro-transactions, positioning the study as a platform opportunity, not a feature follow-on. I got the qualitative study prioritized on the roadmap.
What the Findings Surfaced
The behavioral data gave the lurker segment a measurable profile for the first time.
- Lurkers drove 54% of milestone activity, outperforming the segment the platform had been designing for
- 1.75% Five-Minute Play (the share of viewers who stayed in a stream for five or more minutes) increase within the experiment cohort
- 94% overall engagement increase on initial launch, measured across the first 2-3 months post-launch
The more structural finding was in the measurement model itself. The platform's existing engagement metrics tracked visible behavior only, which made ambient participation absent from the data being used to address the retention problem. The lurker segment was not absent from the platform. It was absent from the measurement model.
Impact and Forward Motion
Before this work, Watch Streaks was built for one type of viewer, the one willing to participate publicly. The private sharing flow exists because the research pitch named a behavioral gap the original scope had no mechanism to detect. The team's frame shifted from engagement as a single visible mode to a spectrum of participation, including the ambient behavior the measurement model had not previously been built to surface. Before this work, lurker behavior had no dedicated research investment on the platform. The qualitative study now on the roadmap marks the first time the platform has resourced ambient participation as a platform-scope question in its own right.
“Expanding perspectives into deliberate insight for bolder moves”
Made and designed by nayeri
© 2026 All Rights Reserved
Mapping the Spectrum of Engagement
The most consequential users on a platform are not always the most visible ones. The more consequential question is not who they are. It is why the strategy being used to reach them hasn't been built to find them.
TL;DR
Overview
Watch Streaks was a Twitch retention initiative built around extending engagement recognition to viewers who were already returning but had never had a feature designed around how they participated. On this project I owned the research function, defining the research questions, the experiment design, and the methodological limits before the data came in. The existing retention strategy was optimized for visible participation, and I recognized early that a successful launch built around that assumption would produce a short-term lift, not a retention outcome.
My accessibility background had trained me to ask who a system was not designed for before asking what it should do next. On this platform, that question pointed to lurkers, the second highest segment by return rate and the one the existing retention strategy had no design language for.
Business Problem
Retention had surged during the pandemic and was declining as behavior normalized, and engagement was one of several levers tied to revenue. The existing strategy was concentrated on active, visible participants, the segment most likely to produce a measurable short-term lift. The data being used to address the retention problem tracked visible behavior only, which made an entire class of returning behavior structurally invisible to the platform's measurement model. A successful launch and a meaningful retention outcome were not the same thing.
Research Approach
I proposed this study, which was not on the roadmap. This was a behavioral choice experiment run as a live in-product test, designed to surface what people did while leaving the question of why open for qualitative follow-on. Before the data came in, I named the limit to the team directly. Behavioral data could surface what people chose, not why they chose it. The case I made connected lurker behavior to engagement and micro-transactions, positioning the study as a platform opportunity, not a feature follow-on. I got the qualitative study prioritized on the roadmap.
What the Findings Surfaced
The behavioral data gave the lurker segment a measurable profile for the first time.
- Lurkers drove 54% of milestone activity, outperforming the segment the platform had been designing for
- 1.75% Five-Minute Play (the share of viewers who stayed in a stream for five or more minutes) increase within the experiment cohort
- 94% overall engagement increase on initial launch, measured across the first 2-3 months post-launch
The more structural finding was in the measurement model itself. The platform's existing engagement metrics tracked visible behavior only, which made ambient participation absent from the data being used to address the retention problem. The lurker segment was not absent from the platform. It was absent from the measurement model.
Impact and Forward Motion
Before this work, Watch Streaks was built for one type of viewer, the one willing to participate publicly. The private sharing flow exists because the research pitch named a behavioral gap the original scope had no mechanism to detect. The team's frame shifted from engagement as a single visible mode to a spectrum of participation, including the ambient behavior the measurement model had not previously been built to surface. Before this work, lurker behavior had no dedicated research investment on the platform. The qualitative study now on the roadmap marks the first time the platform has resourced ambient participation as a platform-scope question in its own right.
“Expanding perspectives into deliberate insight for bolder moves”
Made and designed by nayeri
© 2026 All Rights Reserved