Netflix · PRD
Watch something that fits the time you have tonight
Maayan Nakav & Tal Shami · Planning Products · Final project · September 2026
Half of Netflix viewing happens in the evening. For people with a fixed window, it goes wrong in two ways.
Too long to choose
14min
to choose what to watch
1 in 5
give up and do something else
Too long to finish
A 2.5-hour film started at 21:15 ends at 23:45, long after bedtime.
A "Movies under 90 minutes" row. Films only, and it never says when anything ends.
People with little time in the evening start watching sooner and finish what they start.
Primary metric
Time to start watching
Goal: at least 2% faster
Measured on everyone in the test group, not only people who use the feature
Secondary
Finish the plan
Watched at least 90% of what they planned
Secondary
Use it again
A new feature, so there is no baseline yet
A/B test, 4 weeks, 50/50. Evening viewing and retention must not drop.
Tell us how much time you have. We show what fits, and when it ends.
We never choose for the member.
Netflix removed its shuffle button, Play Something, after low use. So the member sets the limit, and the member picks.
A normal evening, and a short one.
MMaya
Tuesday, 21:00. Two hours before bed.
✓Playing within a minute. Done by 22:55.
OOmer
22:20. Forty minutes. His next episode is 55.
✓Nothing fits exactly. He picks a 25-minute episode instead.
19 requirements, prioritized with MoSCoW.
Four quality targets, and no change to the price.
2sec
Options on screen, 19 times out of 20
3clicks
From Home to playing
±5min
End time accuracy, 19 times out of 20
0
New personal data collected
If it's slowAfter 3 seconds, a simple list by length and popularity. Never a blank screen.
PriceNo change. Included in every ad-free plan.
The TV app, in the US and Canada, on ad-free plans. Ten ideas, ranked by RICE.
| Idea | Reach | Impact | Conf. | Effort | Score | Decision |
|---|---|---|---|---|---|---|
| End time on each option | 4.7M | 1 | 80% | 0.5 | 7.52 | Launch |
| Episode runs | 4.7M | 2 | 70% | 3 | 2.19 | Launch |
| Film options | 4.7M | 1 | 80% | 2 | 1.88 | Launch |
| The "nothing fits" screen | 1.0M | 1 | 80% | 1 | 0.80 | Launch |
Launch line · in: 0.80 and up · out: 0.35 and below | ||||||
| Ad-supported plan | 4.7M | 2 | 30% | 8 | 0.35 | Phase 3 |
| Mobile and web apps | 1.5M | 1 | 50% | 4 | 0.19 | Phase 2 |
| Re-plan after pauses | 2.0M | 0.5 | 50% | 3 | 0.17 | Phase 2 |
| Joint picks for 2+ profiles | 2.0M | 1 | 30% | 8 | 0.08 | Phase 3 |
| A film and episodes in one evening | 1.0M | 0.5 | 40% | 5 | 0.04 | Phase 3 |
| Picks by age group | 4.7M | 0.25 | 20% | 6 | 0.04 | Not planned |
The math · click any row
Reach × Impact × Confidence ÷ Effort
End time on each option
4.7M × 1 × 80% ÷ 0.5
7.52
In the first release. High reach, very low effort. The cheapest way to deliver the core promise.
Reach = members expected to try it in a quarter
Three phases. Each one starts only when the previous gate is passed.
Phase 1
Launch
Phase 2
Expand
Phase 3
Next
What could go wrong, and what we do about it.
People don't use it, like Play Something
The banner appears at the moment of choice. If under 10% use it again after the test, we pause.
End times are wrong
Count to the end credits, round up, and skip titles without reliable runtime data.
Weak options in short windows
Relevant titles only, and a clear "nothing fits" screen instead of filler.
People watch less overall
Guardrails on evening viewing and retention. "Keep watching" is one click away.
Launch problems across teams
Employees first, then 1%, 10% and the 50/50 test, with instant rollback.
Maya's evening · 21:00 to 22:55
Play
optional
too little time
21:00Maya opens Netflix
Evening banner, 18:00 to 02:00Where people decide what to watch
One action: StartNo settings, no extra choices
Always in the menuFor eligible profiles, day or night
Nothing else on Home changes, so it stays as fast as today.
21:002 hours is already selected
One questionNothing else to decide
Six presetsThe end time updates as you choose
Or a custom time20 min to 4 hours. The last choice is remembered.
Returning members just confirm. The second of three clicks to playing.
21:00Six options, all end by 23:00
What you're watching comes firstEach run starts where you stopped
Then new titles that fitFilter by genre in one click
When each one endsRounded up: 22:54 shows as 22:55
Six options: two to continue, four new.
Optional
Start from any episodePick an earlier one to watch it again
One more episode?See the overrun first, then decide
One clear rule
Up to 5 min over counts as a fit
6 to 20 min over: offered, minutes shown
More than 20 min over: never offered
Filling the free 30 minutes with a short film comes in Phase 3.
22:55Maya's plan is complete
Autoplay stops at the plan's endNothing starts on its own
Continuing is a choiceOne click away, never automatic
The end time stays in the playerWhenever the controls are open
This is where we measure completion.
22:20Omer has 40 minutes
Says plainly that nothing fitsThen shows the closest options
The overrun, in minutesOnly up to 20, the same rule as screen 4
Three different reasonsContinue, fits your time, popular now
Running 15 minutes over by choice is fine. Finding out at 23:15 is not.
Leaving out the ads plan, even though it brings most new sign-ups. Without predictable ad breaks, we can't promise an end time.
Measuring the whole test group, not only people who use the feature. Otherwise we'd measure who chooses it, not what it does.
How often people stop mid-title, the 60% completion baseline, and whether 15% try it. The plan rests on these three assumptions.
Questions?
Maayan Nakav & Tal Shami
| Decision | What we chose | Why | What we rejected |
|---|---|---|---|
| How it works | The member sets the limit, we offer options | Play Something failed because it chose for the member | A shuffle or auto-play button |
| Platform | TV app only | Evening viewing happens on TV, and one platform gets the test running fastest | Mobile and web at the same time |
| Market | US and Canada | Highest revenue per member, one main language | A global launch |
| Plans | Ad-free plans only | Ad breaks make the end time unpredictable | The ads plan with an estimated ad load |
| Price | No separate charge | Netflix packages value into plans, and paying for discovery adds friction | A paid add-on |
| Number of options | Up to six | Enough for a real choice, few enough to decide in seconds | A full personalized row |
| Content | Films and episode runs | Series are about 3 of every 4 hours watched | Films only |
| End time | On every option, rounded up to 5 minutes | It turns a filter into a plan, and rounding up protects trust | Showing the exact minute |
| Who it's for | The active profile only | Choosing for two people is a different problem | Joint picks at launch |
| When nothing fits | Three labeled options, overrun up to 20 minutes | An empty screen breaks trust for the people who need the feature most | An empty state, or quietly showing longer titles |
| Kids profiles | Not included | Bedtime planning for children is a separate product problem | A child-friendly time window |
| Prioritization | RICE across ten ideas | Comparable scores, with the assumptions out in the open | Judgment alone |
| Primary metric | Whole test group against whole control group | Comparing only feature users with control rewards self-selection | Feature users against control as the headline |
| Rewatching | The member can change the starting episode | Members sometimes want to watch the last episode again, and forcing the resume point blocks it | Always starting where the member stopped |
| An episode that almost fits | Offer one more episode when it runs 6 to 20 minutes over, with the minutes shown | The member decides with full information, and up to 5 minutes is within our accuracy anyway | Silently dropping or adding it |
Reach is the number of eligible members we expect to try the feature in a quarter. Every step after the first is our assumption, kept on the low side.
Bars are to scale. Membership has grown since 2024, so the real number is likely higher.
Impact scale. 3 massive · 2 high · 1 medium · 0.5 low · 0.25 minimal.
The whole launch scores about 0.59: 4.7M × 2 × 50% ÷ 8 months. Confidence is only 50% because Play Something failed on adoption.
Effort. 6.5 months of work for the four launch items, 8 once tracking, the A/B setup and testing across TV devices are added.
| Metric | Exact definition | Today | Goal |
|---|---|---|---|
| Time to start watching PRIMARY | Median time from choosing a profile to playback, in evening sessions (18:00 to 24:00 local). Every member in the test group against every member in control, whether or not they opened the feature. | About 12 min, the US industry average (assumption) | At least 2% lower than control |
| Finish the plan | Share of sessions started from the feature where the member watched at least 90% of the planned runtime, in the same session | About 60% for evening sessions started from Home (assumption) | 75% |
| Use it again | Share of members who used the feature once and used it again within 14 days | New feature | 30% (assumption) |
The test. Ad-free members on the TV app in the US and Canada, split 50/50, for four weeks.
Success. The primary goal is met with statistical significance, plus at least one secondary goal.
Guardrails. Total evening viewing hours and 28-day retention in the test group must not fall below control.
Why 2% and not 25%. Inside sessions that use the feature we expect starts about 25% faster. Most sessions won't use it, so across the whole group the effect is about 2%.
Events we track. Feature opened, time selected, options shown, option selected, playback started, planned run completed, left without playing.
←→ or the buttons below