PRODUCT CONCEPT · MINI PRD · AI
“Tonight”: AI that ends the endless scroll
A self-initiated concept for Disney+ with Hulu. Households open the app, scroll, and give up. Tonight turns “what should we watch?” into two taps and three picks, each with a plain-language reason, built for everyone on the couch.
Product sense
AI recommendations
PRD
Success metrics
Streaming
MY ROLE
Product strategy, PRD, user stories, success metrics, UX concept
PRODUCT
Disney+ with Hulu, TV and mobile
TYPE
Self-initiated concept. Not affiliated with or endorsed by Disney
FOCUS
Decision time, trust, and household viewing
TONIGHT
Three picks for Mia, Dan, and Leo
👪 3 watching
⏱ About 1h 45m
Something light
BEST MATCH
The Long Harbor
Adventure · 1h 52m · PG
✦
All three of you finished a family adventure last week.
QUICK WATCH
Night Shift Kitchen
Comedy series · 28m episodes
✦
Light, short, and new to everyone. Fits before Leo’s bedtime.
WILDCARD
Paper Moons
Animated · 1h 38m · PG
✦
A wildcard: outside your usual picks, loved by similar households.
▶ Play The Long Harbor
Not tonight: show 3 more
Titles and artwork are placeholders
THE PROBLEM
More to watch has made choosing harder.
Bringing Hulu into the Disney+ app gives households a much bigger catalog in one place. That is the value of the bundle, and it also creates a risk: rows of tiles leave all the choosing to the viewer. You know the result: scrolling, trailers, a debate on the couch, and then a rewatch or closing the app.
WHAT THE DATA SAYS
12 min
Average time U.S. viewers spend searching for something to watch, up from 10.5 minutes in 2023.
19%
Abandon a viewing session when they can’t find something to watch.
49%
Would cancel a service because it’s hard to find something to watch.
Source: Gracenote (Nielsen), 2025 State of Play, a survey of 3,000 streaming consumers in six countries. The 12-minute figure is U.S. only. The percentages cover all six countries. Next step: validate with a survey of Disney+ households.
HYPOTHESIS
My hypothesis: the bottleneck is decision confidence, especially when more than one person is on the couch.
I saw the same pattern in Taste Test research. When options multiply, people start worrying about choosing wrong.
WHO IT’S FOR
Three moments when choosing breaks down.
SEGMENT 01
The mixed-age household
A parent choosing for kids and adults at once, inside parental controls, with a bedtime deadline.
SEGMENT 02
Two tastes, one screen
A couple or roommates with different profiles who spend more time negotiating than watching.
SEGMENT 03
The end-of-day solo viewer
Low energy, short window, wants something good fast without browsing.
THE SOLUTION
Two taps, three picks, and a reason for each.
Tonight is an entry point on the home screen. Viewers confirm who’s watching and roughly how much time they have, then get three picks across Disney+ and Hulu. Each pick comes with a one-line reason. “Not tonight” shows three more and teaches the model. Every pick respects each viewer’s profile and parental controls.
MATCHING THE PROBLEM TO THE RIGHT AI CAPABILITY
Recommendations
Rank candidates for the whole group by blending each viewer’s history with runtime and rating limits.
Generative AI, grounded
Write the one-line reason using only catalog metadata and viewing signals, so it can’t invent plot details.
Learning from “no”
Treat “Not tonight” as a strong signal to reshape the next set and future evenings.
LIVE PICK
Your team, tonight
✦ Starts in 20 minutes.
Ends before bedtime.
WHERE ESPN FITS
A live game can be one of the three.
For households with ESPN in the app, a game that starts soon competes with movies and shows on the same terms: who is watching, how much time they have, and a clear reason. Since games start at a set time, a live pick matters most just before it begins.
DECISION 01
Show three picks.
Fewer options is the product. Three is enough for a real choice and small enough to decide on a TV remote.
TRADEOFF
Some viewers want to browse, so browse stays one click away. Tonight sits alongside it.
DECISION 02
Lead with reasons.
A reason like “all three of you finished a family adventure last week” builds more trust than a better thumbnail.
TRADEOFF
Each reason has to be accurate, so generation only uses catalog metadata and falls back to a template.
DECISION 03
Use chips in V1 and save chat for later.
Typing on a TV remote is painful. Chips for who, time, and mood cover most of the need with almost no effort.
TRADEOFF
We give up open-ended requests like “something like that show from last summer.” I’d test those in V2 on mobile.
USER STORIES
User stories.
AS A…
I WANT TO…
SO THAT…
PRIORITY
Parent in a mixed-age household
see picks that fit every viewer’s parental controls
I never have to screen a suggestion before the kids see it
P0
Couple sharing a screen
get picks that balance both of our profiles
we spend the evening watching instead of negotiating
P0
Tired solo viewer
start something good in under a minute
the app feels effortless at the end of the day
P0
Any viewer
see why each title was picked for me
I trust the pick enough to press play
P0
Viewer who says no
dismiss a set and get three different picks
the next set feels smarter and doesn’t repeat
P1
Viewer in a hurry
say how much time I have
nothing I start runs past bedtime
P1
SUCCESS METRICS
North star: time from opening the app to pressing play.
Targets get set after we measure a baseline. Defining them before launch is a lesson from my Learner Mobile work, where the best result was discovered after the fact.
PRIMARY
• Time to first play
• Sessions that end with no playback
• Share of Tonight sets where a pick gets played
QUALITY
• Chosen title watched past 50%
• “Not tonight” rate per set
• Return use of Tonight within 7 days
GUARDRAILS
• Total viewing time holds steady
• Zero parental control violations
• Picks load in under 1 second
OKRS
One objective, three key results.
Each key result maps to a metric above. Targets are proposals until Phase 0 sets a baseline.
OBJECTIVE
Make choosing what to watch together feel fast and trustworthy.
KR 1
Cut median time from opening the app to pressing play by 25% for households using Tonight, compared with browse.
Primary
KR 2
At least 40% of Tonight sets end with a pick being played.
Primary
KR 3
Hold total viewing time and title variety steady, with zero parental control violations.
Guardrail
COMPETITIVE LANDSCAPE
How streaming apps help people choose today.
Most services mix a few patterns. Each solves part of the problem. None is built for a group making a fast decision together.
APPROACH
GOOD AT
FALLS SHORT
Rows and carousels
Breadth. Shows off a big catalog.
Leaves all the choosing to the viewer.
One-tap “play something”
Speed.
No reason to trust the pick, so people skip it.
Conversational AI search
Flexible, open-ended requests.
Typing on a TV remote is effort. Built for one person.
Tonight
Speed plus a reason, for the whole household.
Narrow by design. Browse still exists for exploring.
A concept-level scan based on common industry patterns. I’d validate it with a hands-on audit of current apps before any real review.
RISKS AND MITIGATIONS
What could go wrong, and what I’d do about it.
Privacy
Household picks could reveal what one person watches on their own profile.
Mitigation: Use only signals from profiles marked as watching. Never surface one viewer’s private history in a reason shown to others. Kids’ profiles use conservative defaults.
Latency
Group ranking plus generated text could make the screen slow on a TV.
Mitigation: Precompute candidate sets per household, cache reasons, and fall back to templates if generation is slow. Target: picks in under 1 second.
Wrong or invented reasons
A generated reason could state something untrue about a title.
Mitigation: Generate only from catalog metadata and real viewing events, and use a template when confidence is low.
Filter bubble
Three safe picks every night gets repetitive.
Mitigation: One of three picks is always a labeled wildcard.
Pulling time from browse
Tonight could cut exploration that matters for new releases.
Mitigation: Guardrail on total viewing and title diversity. Browse stays one click away.
ROLLOUT PLAN
Test the riskiest assumption first.
I think the riskiest assumption is whether a reason makes people press play, more than the ranking itself. The first experiment tests that on its own.
PHASE 0
Discover
Diary study with mixed-age households, plus usability tests of this prototype on a TV.
PHASE 1
Isolate the reason
Small A/B test: three picks with reasons vs. three picks without. Measure time to play and pick acceptance.
PHASE 2
Expand
Roll out on TV and mobile to more households if guardrails hold. Add mood chips.
PHASE 3
Open it up
Test natural-language requests on mobile, where typing is easy.
WHY THIS CONCEPT
Across these projects I kept finding the same thing: when choice multiplies, people need a reason to trust what’s in front of them.
Taste Test showed me trust beats discovery. Twofer showed me a live, limited set drives action. Learner Mobile showed me adoption comes before depth. Tonight applies all three to the moment a family sits down to watch.
LET’S TALK
Building for real people
at scale?
I’m looking for an Associate Product Manager role on a team like yours. LinkedIn is the fastest way to reach me.
NEXT CASE STUDY
Learner Mobile →
Back to my shipped, award-winning work.
PRODUCT CONCEPT · MINI PRD · AI
“Tonight”: AI that ends the endless scroll
A self-initiated concept for Disney+ with Hulu. Households open the app, scroll, and give up. Tonight turns “what should we watch?” into two taps and three picks, each with a plain-language reason, built for everyone on the couch.
Product sense
AI recommendations
PRD
Success metrics
Streaming
MY ROLE
Product strategy, PRD, user stories, success metrics, UX concept
PRODUCT
Disney+ with Hulu, TV and mobile
TYPE
Self-initiated concept. Not affiliated with or endorsed by Disney
FOCUS
Decision time, trust, and household viewing
TONIGHT
Three picks for Mia, Dan, and Leo
👪 3 watching
⏱ About 1h 45m
Something light
BEST MATCH
The Long Harbor
Adventure · 1h 52m · PG
✦
All three of you finished a family adventure last week.
QUICK WATCH
Night Shift Kitchen
Comedy series · 28m episodes
✦
Light, short, and new to everyone. Fits before Leo’s bedtime.
WILDCARD
Paper Moons
Animated · 1h 38m · PG
✦
A wildcard: outside your usual picks, loved by similar households.
▶ Play The Long Harbor
Not tonight: show 3 more
Titles and artwork are placeholders
THE PROBLEM
More to watch has made choosing harder.
Bringing Hulu into the Disney+ app gives households a much bigger catalog in one place. That is the value of the bundle, and it also creates a risk: rows of tiles leave all the choosing to the viewer. You know the result: scrolling, trailers, a debate on the couch, and then a rewatch or closing the app.
WHAT THE DATA SAYS
12 min
Average time U.S. viewers spend searching for something to watch, up from 10.5 minutes in 2023.
19%
Abandon a viewing session when they can’t find something to watch.
49%
Would cancel a service because it’s hard to find something to watch.
Source: Gracenote (Nielsen), 2025 State of Play, a survey of 3,000 streaming consumers in six countries. The 12-minute figure is U.S. only. The percentages cover all six countries. Next step: validate with a survey of Disney+ households.
HYPOTHESIS
My hypothesis: the bottleneck is decision confidence, especially when more than one person is on the couch.
I saw the same pattern in Taste Test research. When options multiply, people start worrying about choosing wrong.
WHO IT’S FOR
Three moments when choosing breaks down.
SEGMENT 01
The mixed-age household
A parent choosing for kids and adults at once, inside parental controls, with a bedtime deadline.
SEGMENT 02
Two tastes, one screen
A couple or roommates with different profiles who spend more time negotiating than watching.
SEGMENT 03
The end-of-day solo viewer
Low energy, short window, wants something good fast without browsing.
THE SOLUTION
Two taps, three picks, and a reason for each.
Tonight is an entry point on the home screen. Viewers confirm who’s watching and roughly how much time they have, then get three picks across Disney+ and Hulu. Each pick comes with a one-line reason. “Not tonight” shows three more and teaches the model. Every pick respects each viewer’s profile and parental controls.
MATCHING THE PROBLEM TO THE RIGHT AI CAPABILITY
Recommendations
Rank candidates for the whole group by blending each viewer’s history with runtime and rating limits.
Generative AI, grounded
Write the one-line reason using only catalog metadata and viewing signals, so it can’t invent plot details.
Learning from “no”
Treat “Not tonight” as a strong signal to reshape the next set and future evenings.
LIVE PICK
Your team, tonight
✦ Starts in 20 minutes.
Ends before bedtime.
WHERE ESPN FITS
A live game can be one of the three.
For households with ESPN in the app, a game that starts soon competes with movies and shows on the same terms: who is watching, how much time they have, and a clear reason. Since games start at a set time, a live pick matters most just before it begins.
DECISION 01
Show three picks.
Fewer options is the product. Three is enough for a real choice and small enough to decide on a TV remote.
TRADEOFF
Some viewers want to browse, so browse stays one click away. Tonight sits alongside it.
DECISION 02
Lead with reasons.
A reason like “all three of you finished a family adventure last week” builds more trust than a better thumbnail.
TRADEOFF
Each reason has to be accurate, so generation only uses catalog metadata and falls back to a template.
DECISION 03
Use chips in V1 and save chat for later.
Typing on a TV remote is painful. Chips for who, time, and mood cover most of the need with almost no effort.
TRADEOFF
We give up open-ended requests like “something like that show from last summer.” I’d test those in V2 on mobile.
USER STORIES
User stories.
AS A…
I WANT TO…
SO THAT…
PRIORITY
Parent in a mixed-age household
see picks that fit every viewer’s parental controls
I never have to screen a suggestion before the kids see it
P0
Couple sharing a screen
get picks that balance both of our profiles
we spend the evening watching instead of negotiating
P0
Tired solo viewer
start something good in under a minute
the app feels effortless at the end of the day
P0
Any viewer
see why each title was picked for me
I trust the pick enough to press play
P0
Viewer who says no
dismiss a set and get three different picks
the next set feels smarter and doesn’t repeat
P1
Viewer in a hurry
say how much time I have
nothing I start runs past bedtime
P1
SUCCESS METRICS
North star: time from opening the app to pressing play.
Targets get set after we measure a baseline. Defining them before launch is a lesson from my Learner Mobile work, where the best result was discovered after the fact.
PRIMARY
• Time to first play
• Sessions that end with no playback
• Share of Tonight sets where a pick gets played
QUALITY
• Chosen title watched past 50%
• “Not tonight” rate per set
• Return use of Tonight within 7 days
GUARDRAILS
• Total viewing time holds steady
• Zero parental control violations
• Picks load in under 1 second
OKRS
One objective, three key results.
Each key result maps to a metric above. Targets are proposals until Phase 0 sets a baseline.
OBJECTIVE
Make choosing what to watch together feel fast and trustworthy.
KR 1
Cut median time from opening the app to pressing play by 25% for households using Tonight, compared with browse.
Primary
KR 2
At least 40% of Tonight sets end with a pick being played.
Primary
KR 3
Hold total viewing time and title variety steady, with zero parental control violations.
Guardrail
COMPETITIVE LANDSCAPE
How streaming apps help people choose today.
Most services mix a few patterns. Each solves part of the problem. None is built for a group making a fast decision together.
APPROACH
GOOD AT
FALLS SHORT
Rows and carousels
Breadth. Shows off a big catalog.
Leaves all the choosing to the viewer.
One-tap “play something”
Speed.
No reason to trust the pick, so people skip it.
Conversational AI search
Flexible, open-ended requests.
Typing on a TV remote is effort. Built for one person.
Tonight
Speed plus a reason, for the whole household.
Narrow by design. Browse still exists for exploring.
A concept-level scan based on common industry patterns. I’d validate it with a hands-on audit of current apps before any real review.
RISKS AND MITIGATIONS
What could go wrong, and what I’d do about it.
Privacy
Household picks could reveal what one person watches on their own profile.
Mitigation: Use only signals from profiles marked as watching. Never surface one viewer’s private history in a reason shown to others. Kids’ profiles use conservative defaults.
Latency
Group ranking plus generated text could make the screen slow on a TV.
Mitigation: Precompute candidate sets per household, cache reasons, and fall back to templates if generation is slow. Target: picks in under 1 second.
Wrong or invented reasons
A generated reason could state something untrue about a title.
Mitigation: Generate only from catalog metadata and real viewing events, and use a template when confidence is low.
Filter bubble
Three safe picks every night gets repetitive.
Mitigation: One of three picks is always a labeled wildcard.
Pulling time from browse
Tonight could cut exploration that matters for new releases.
Mitigation: Guardrail on total viewing and title diversity. Browse stays one click away.
ROLLOUT PLAN
Test the riskiest assumption first.
I think the riskiest assumption is whether a reason makes people press play, more than the ranking itself. The first experiment tests that on its own.
PHASE 0
Discover
Diary study with mixed-age households, plus usability tests of this prototype on a TV.
PHASE 1
Isolate the reason
Small A/B test: three picks with reasons vs. three picks without. Measure time to play and pick acceptance.
PHASE 2
Expand
Roll out on TV and mobile to more households if guardrails hold. Add mood chips.
PHASE 3
Open it up
Test natural-language requests on mobile, where typing is easy.
WHY THIS CONCEPT
Across these projects I kept finding the same thing: when choice multiplies, people need a reason to trust what’s in front of them.
Taste Test showed me trust beats discovery. Twofer showed me a live, limited set drives action. Learner Mobile showed me adoption comes before depth. Tonight applies all three to the moment a family sits down to watch.
LET’S TALK
Building for real people
at scale?
I’m looking for an Associate Product Manager role on a team like yours. LinkedIn is the fastest way to reach me.
NEXT CASE STUDY
Learner Mobile →
Back to my shipped, award-winning work.
PRODUCT CONCEPT · MINI PRD · AI
“Tonight”: AI that ends
the endless scroll
A self-initiated concept for Disney+ with Hulu. Households open the app, scroll, and give up. Tonight turns “what should we watch?” into two taps and three picks, each with a plain-language reason, built for everyone on the couch.
Product sense
AI recommendations
PRD
Success metrics
Streaming
MY ROLE
Product strategy, PRD, user stories, success metrics, UX concept
PRODUCT
Disney+ with Hulu, TV and mobile
TYPE
Self-initiated concept. Not affiliated with or endorsed by Disney
FOCUS
Decision time, trust, and household viewing
TONIGHT
Three picks for Mia, Dan, and Leo
👪 3 watching
⏱ About 1h 45m
Something light
BEST MATCH
The Long Harbor
Adventure · 1h 52m · PG
✦
All three of you finished a family adventure last week.
QUICK WATCH
Night Shift Kitchen
Comedy series · 28m episodes
✦
Light, short, and new to everyone. Fits before Leo’s bedtime.
WILDCARD
Paper Moons
Animated · 1h 38m · PG
✦
A wildcard: outside your usual picks, loved by similar households.
▶ Play The Long Harbor
Not tonight: show 3 more
Titles and artwork are placeholders
THE PROBLEM
More to watch has made choosing harder.
Bringing Hulu into the Disney+ app gives households a much bigger catalog in one place. That is the value of the bundle, and it also creates a risk: rows of tiles leave all the choosing to the viewer. You know the result: scrolling, trailers, a debate on the couch, and then a rewatch or closing the app.
WHAT THE DATA SAYS
12 min
Average time U.S. viewers spend searching for something to watch, up from 10.5 minutes in 2023.
19%
Abandon a viewing session when they can’t find something to watch.
49%
Would cancel a service because it’s hard to find something to watch.
Source: Gracenote (Nielsen), 2025 State of Play, a survey of 3,000 streaming consumers in six countries. The 12-minute figure is U.S. only. The percentages cover all six countries. Next step: validate with a survey of Disney+ households.
HYPOTHESIS
My hypothesis: the bottleneck is decision confidence, especially when more than one person is on the couch.
I saw the same pattern in Taste Test research. When options multiply, people start worrying about choosing wrong.
WHO IT’S FOR
Three moments when choosing breaks down.
SEGMENT 01
The mixed-age household
A parent choosing for kids and adults at once, inside parental controls, with a bedtime deadline.
SEGMENT 02
Two tastes, one screen
A couple or roommates with different profiles who spend more time negotiating than watching.
SEGMENT 03
The end-of-day solo viewer
Low energy, short window, wants something good fast without browsing.
THE SOLUTION
Two taps, three picks, and a reason for each.
Tonight is an entry point on the home screen. Viewers confirm who’s watching and roughly how much time they have, then get three picks across Disney+ and Hulu. Each pick comes with a one-line reason. “Not tonight” shows three more and teaches the model. Every pick respects each viewer’s profile and parental controls.
MATCHING THE PROBLEM TO THE RIGHT AI CAPABILITY
Recommendations
Rank candidates for the whole group by blending each viewer’s history with runtime and rating limits.
Generative AI, grounded
Write the one-line reason using only catalog metadata and viewing signals, so it can’t invent plot details.
Learning from “no”
Treat “Not tonight” as a strong signal to reshape the next set and future evenings.
LIVE PICK
Your team, tonight
✦ Starts in 20 minutes.
Ends before bedtime.
WHERE ESPN FITS
A live game can be one of the three.
For households with ESPN in the app, a game that starts soon competes with movies and shows on the same terms: who is watching, how much time they have, and a clear reason. Since games start at a set time, a live pick matters most just before it begins.
DECISION 01
Show three picks.
Fewer options is the product. Three is enough for a real choice and small enough to decide on a TV remote.
TRADEOFF
Some viewers want to browse, so browse stays one click away. Tonight sits alongside it.
DECISION 02
Lead with reasons.
A reason like “all three of you finished a family adventure last week” builds more trust than a better thumbnail.
TRADEOFF
Each reason has to be accurate, so generation only uses catalog metadata and falls back to a template.
DECISION 03
Use chips in V1 and save chat for later.
Typing on a TV remote is painful. Chips for who, time, and mood cover most of the need with almost no effort.
TRADEOFF
We give up open-ended requests like “something like that show from last summer.” I’d test those in V2 on mobile.
USER STORIES
User stories.
AS A…
I WANT TO…
SO THAT…
PRIORITY
Parent in a mixed-age household
see picks that fit every viewer’s parental controls
I never have to screen a suggestion before the kids see it
P0
Couple sharing a screen
get picks that balance both of our profiles
we spend the evening watching instead of negotiating
P0
Tired solo viewer
start something good in under a minute
the app feels effortless at the end of the day
P0
Any viewer
see why each title was picked for me
I trust the pick enough to press play
P0
Viewer who says no
dismiss a set and get three different picks
the next set feels smarter and doesn’t repeat
P1
Viewer in a hurry
say how much time I have
nothing I start runs past bedtime
P1
SUCCESS METRICS
North star: time from opening the app to pressing play.
Targets get set after we measure a baseline. Defining them before launch is a lesson from my Learner Mobile work, where the best result was discovered after the fact.
PRIMARY
• Time to first play
• Sessions that end with no playback
• Share of Tonight sets where a pick gets played
QUALITY
• Chosen title watched past 50%
• “Not tonight” rate per set
• Return use of Tonight within 7 days
GUARDRAILS
• Total viewing time holds steady
• Zero parental control violations
• Picks load in under 1 second
OKRS
One objective, three key results.
Each key result maps to a metric above. Targets are proposals until Phase 0 sets a baseline.
OBJECTIVE
Make choosing what to watch together feel fast and trustworthy.
KR 1
Cut median time from opening the app to pressing play by 25% for households using Tonight, compared with browse.
Primary
KR 2
At least 40% of Tonight sets end with a pick being played.
Primary
KR 3
Hold total viewing time and title variety steady, with zero parental control violations.
Guardrail
COMPETITIVE LANDSCAPE
How streaming apps help people choose today.
Most services mix a few patterns. Each solves part of the problem. None is built for a group making a fast decision together.
APPROACH
GOOD AT
FALLS SHORT
Rows and carousels
Breadth. Shows off a big catalog.
Leaves all the choosing to the viewer.
One-tap “play something”
Speed.
No reason to trust the pick, so people skip it.
Conversational AI search
Flexible, open-ended requests.
Typing on a TV remote is effort. Built for one person.
Tonight
Speed plus a reason, for the whole household.
Narrow by design. Browse still exists for exploring.
A concept-level scan based on common industry patterns. I’d validate it with a hands-on audit of current apps before any real review.
RISKS AND MITIGATIONS
What could go wrong, and what I’d do about it.
Privacy
Household picks could reveal what one person watches on their own profile.
Mitigation: Use only signals from profiles marked as watching. Never surface one viewer’s private history in a reason shown to others. Kids’ profiles use conservative defaults.
Latency
Group ranking plus generated text could make the screen slow on a TV.
Mitigation: Precompute candidate sets per household, cache reasons, and fall back to templates if generation is slow. Target: picks in under 1 second.
Wrong or invented reasons
A generated reason could state something untrue about a title.
Mitigation: Generate only from catalog metadata and real viewing events, and use a template when confidence is low.
Filter bubble
Three safe picks every night gets repetitive.
Mitigation: One of three picks is always a labeled wildcard.
Pulling time from browse
Tonight could cut exploration that matters for new releases.
Mitigation: Guardrail on total viewing and title diversity. Browse stays one click away.
ROLLOUT PLAN
Test the riskiest assumption first.
I think the riskiest assumption is whether a reason makes people press play, more than the ranking itself. The first experiment tests that on its own.
PHASE 0
Discover
Diary study with mixed-age households, plus usability tests of this prototype on a TV.
PHASE 1
Isolate the reason
Small A/B test: three picks with reasons vs. three picks without. Measure time to play and pick acceptance.
PHASE 2
Expand
Roll out on TV and mobile to more households if guardrails hold. Add mood chips.
PHASE 3
Open it up
Test natural-language requests on mobile, where typing is easy.
WHY THIS CONCEPT
Across these projects I kept finding the same thing: when choice multiplies, people need a reason to trust what’s in front of them.
Taste Test showed me trust beats discovery. Twofer showed me a live, limited set drives action. Learner Mobile showed me adoption comes before depth. Tonight applies all three to the moment a family sits down to watch.
LET’S TALK
Building for real people at scale?
I’m looking for an Associate Product Manager role on a team like yours. The fastest way to reach me is LinkedIn.
NEXT CASE STUDY
Learner Mobile →
Back to my shipped, award-winning work.
PRODUCT CONCEPT · MINI PRD · AI
“Tonight”: AI that ends
the endless scroll
A self-initiated concept for Disney+ with Hulu. Households open the app, scroll, and give up. Tonight turns “what should we watch?” into two taps and three picks, each with a plain-language reason, built for everyone on the couch.
Product sense
AI recommendations
PRD
Success metrics
Streaming
MY ROLE
Product strategy, PRD, user stories, success metrics, UX concept
PRODUCT
Disney+ with Hulu, TV and mobile
TYPE
Self-initiated concept. Not affiliated with or endorsed by Disney
FOCUS
Decision time, trust, and household viewing
TONIGHT
Three picks for Mia, Dan, and Leo
👪 3 watching
⏱ About 1h 45m
Something light
BEST MATCH
The Long Harbor
Adventure · 1h 52m · PG
✦
All three of you finished a family adventure last week.
QUICK WATCH
Night Shift Kitchen
Comedy series · 28m episodes
✦
Light, short, and new to everyone. Fits before Leo’s bedtime.
WILDCARD
Paper Moons
Animated · 1h 38m · PG
✦
A wildcard: outside your usual picks, loved by similar households.
▶ Play The Long Harbor
Not tonight: show 3 more
Titles and artwork are placeholders
ON THIS PAGE
THE PROBLEM
More to watch has made choosing harder.
Bringing Hulu into the Disney+ app gives households a much bigger catalog in one place. That is the value of the bundle, and it also creates a risk: rows of tiles leave all the choosing to the viewer. You know the result: scrolling, trailers, a debate on the couch, and then a rewatch or closing the app.
WHAT THE DATA SAYS
12 min
Average time U.S. viewers spend searching for something to watch, up from 10.5 minutes in 2023.
19%
Abandon a viewing session when they can’t find something to watch.
49%
Would cancel a service because it’s hard to find something to watch.
Source: Gracenote (Nielsen), 2025 State of Play, a survey of 3,000 streaming consumers in six countries. The 12-minute figure is U.S. only. The percentages cover all six countries. Next step: validate with a survey of Disney+ households.
HYPOTHESIS
My hypothesis: the bottleneck is decision confidence, especially when more than one person is on the couch.
I saw the same pattern in Taste Test research. When options multiply, people start worrying about choosing wrong.
WHO IT’S FOR
Three moments when choosing breaks down.
SEGMENT 01
The mixed-age household
A parent choosing for kids and adults at once, inside parental controls, with a bedtime deadline.
SEGMENT 02
Two tastes, one screen
A couple or roommates with different profiles who spend more time negotiating than watching.
SEGMENT 03
The end-of-day solo viewer
Low energy, short window, wants something good fast without browsing.
THE SOLUTION
Two taps, three picks, and a reason for each.
Tonight is an entry point on the home screen. Viewers confirm who’s watching and roughly how much time they have, then get three picks across Disney+ and Hulu. Each pick comes with a one-line reason. “Not tonight” shows three more and teaches the model. Every pick respects each viewer’s profile and parental controls.
MATCHING THE PROBLEM TO THE RIGHT AI CAPABILITY
Recommendations
Rank candidates for the whole group by blending each viewer’s history with runtime and rating limits.
Generative AI, grounded
Write the one-line reason using only catalog metadata and viewing signals, so it can’t invent plot details.
Learning from “no”
Treat “Not tonight” as a strong signal to reshape the next set and future evenings.
LIVE PICK
Your team, tonight
✦ Starts in 20 minutes.
Ends before bedtime.
WHERE ESPN FITS
A live game can be one of the three.
For households with ESPN in the app, a game that starts soon competes with movies and shows on the same terms: who is watching, how much time they have, and a clear reason. Since games start at a set time, a live pick matters most just before it begins.
DECISION 01
Show three picks.
Fewer options is the product. Three is enough for a real choice and small enough to decide on a TV remote.
TRADEOFF
Some viewers want to browse, so browse stays one click away. Tonight sits alongside it.
DECISION 02
Lead with reasons.
A reason like “all three of you finished a family adventure last week” builds more trust than a better thumbnail.
TRADEOFF
Each reason has to be accurate, so generation only uses catalog metadata and falls back to a template.
DECISION 03
Use chips in V1 and save chat for later.
Typing on a TV remote is painful. Chips for who, time, and mood cover most of the need with almost no effort.
TRADEOFF
We give up open-ended requests like “something like that show from last summer.” I’d test those in V2 on mobile.
USER STORIES
User stories.
AS A…
I WANT TO…
SO THAT…
PRIORITY
Parent in a mixed-age household
see picks that fit every viewer’s parental controls
I never have to screen a suggestion before the kids see it
P0
Couple sharing a screen
get picks that balance both of our profiles
we spend the evening watching instead of negotiating
P0
Tired solo viewer
start something good in under a minute
the app feels effortless at the end of the day
P0
Any viewer
see why each title was picked for me
I trust the pick enough to press play
P0
Viewer who says no
dismiss a set and get three different picks
the next set feels smarter and doesn’t repeat
P1
Viewer in a hurry
say how much time I have
nothing I start runs past bedtime
P1
SUCCESS METRICS
North star: time from opening the app to pressing play.
Targets get set after we measure a baseline. Defining them before launch is a lesson from my Learner Mobile work, where the best result was discovered after the fact.
PRIMARY
• Time to first play
• Sessions that end with no playback
• Share of Tonight sets where a pick gets played
QUALITY
• Chosen title watched past 50%
• “Not tonight” rate per set
• Return use of Tonight within 7 days
GUARDRAILS
• Total viewing time holds steady
• Zero parental control violations
• Picks load in under 1 second
OKRS
One objective, three key results.
Each key result maps to a metric above. Targets are proposals until Phase 0 sets a baseline.
OBJECTIVE
Make choosing what to watch together feel fast and trustworthy.
KR 1
Cut median time from opening the app to pressing play by 25% for households using Tonight, compared with browse.
Primary
KR 2
At least 40% of Tonight sets end with a pick being played.
Primary
KR 3
Hold total viewing time and title variety steady, with zero parental control violations.
Guardrail
COMPETITIVE LANDSCAPE
How streaming apps help people choose today.
Most services mix a few patterns. Each solves part of the problem. None is built for a group making a fast decision together.
APPROACH
GOOD AT
FALLS SHORT
Rows and carousels
Breadth. Shows off a big catalog.
Leaves all the choosing to the viewer.
One-tap “play something”
Speed.
No reason to trust the pick, so people skip it.
Conversational AI search
Flexible, open-ended requests.
Typing on a TV remote is effort. Built for one person.
Tonight
Speed plus a reason, for the whole household.
Narrow by design. Browse still exists for exploring.
A concept-level scan based on common industry patterns. I’d validate it with a hands-on audit of current apps before any real review.
RISKS AND MITIGATIONS
What could go wrong, and what I’d do about it.
Privacy
Household picks could reveal what one person watches on their own profile.
Mitigation: Use only signals from profiles marked as watching. Never surface one viewer’s private history in a reason shown to others. Kids’ profiles use conservative defaults.
Latency
Group ranking plus generated text could make the screen slow on a TV.
Mitigation: Precompute candidate sets per household, cache reasons, and fall back to templates if generation is slow. Target: picks in under 1 second.
Wrong or invented reasons
A generated reason could state something untrue about a title.
Mitigation: Generate only from catalog metadata and real viewing events, and use a template when confidence is low.
Filter bubble
Three safe picks every night gets repetitive.
Mitigation: One of three picks is always a labeled wildcard.
Pulling time from browse
Tonight could cut exploration that matters for new releases.
Mitigation: Guardrail on total viewing and title diversity. Browse stays one click away.
ROLLOUT PLAN
Test the riskiest assumption first.
I think the riskiest assumption is whether a reason makes people press play, more than the ranking itself. The first experiment tests that on its own.
PHASE 0
Discover
Diary study with mixed-age households, plus usability tests of this prototype on a TV.
PHASE 1
Isolate the reason
Small A/B test: three picks with reasons vs. three picks without. Measure time to play and pick acceptance.
PHASE 2
Expand
Roll out on TV and mobile to more households if guardrails hold. Add mood chips.
PHASE 3
Open it up
Test natural-language requests on mobile, where typing is easy.
WHY THIS CONCEPT
Across these projects I kept finding the same thing: when choice multiplies, people need a reason to trust what’s in front of them.
Taste Test showed me trust beats discovery. Twofer showed me a live, limited set drives action. Learner Mobile showed me adoption comes before depth. Tonight applies all three to the moment a family sits down to watch.
LET’S TALK
Building for real people at scale?
I’m looking for an Associate Product Manager role on a team like yours. The fastest way to reach me is LinkedIn.
NEXT CASE STUDY
Learner Mobile →
Back to my shipped, award-winning work.