Problem Statement

Most people who set a bedtime on their iPhone still blow right past it, and the reminder meant to stop them is the easiest thing on the screen to ignore. It shows up at a fixed time, in the same shape every night, competing against apps built by thousands of engineers to keep you watching, and it loses. The market's answer has been a thousand sleep apps, but every one of them runs into the same wall: it's just another app asking you to stop using apps. Apple is the only player that owns the whole phone, the watch on your wrist, and the data on whether you actually slept. So the question I'm solving is: how do I use that position to make Apple's bedtime experience the one people actually follow, not by reminding harder, but by shaping the hour before bed, without it tipping into the kind of nagging or guilt that gets the feature switched off?

Core Tension

On the surface this looks like a volume problem. The reminder is too quiet, so turn it up. If that were the whole story, it wouldn't be worth a case study. The reason it's hard is that there are three failures pulling against each other, and fixing any one of them in isolation makes another one worse.

The first is the ignored failure, and it's where Apple sits today. The reminder fires at the same time, in the same shape, every single night, and the brain does what brains do with anything repetitive: it stops seeing it. By week two it's wallpaper. The feature technically works, it fires on schedule, it just doesn't change anything. And it's losing to an opponent that isn't playing fair. On the other side of that notification is an app built by thousands of engineers whose entire job is to make the next video start before you've decided whether you want it.

The second is the switched-off failure, and it's what happens the moment you try to fix the first one with force. Make the alert bigger, add friction, lean on guilt, and people don't suddenly comply. They go into settings and turn it off. That's worse than being ignored, because an ignored feature still has a shot at the night someone's actually ready to listen. A disabled one doesn't. Nobody wanders back into settings to re-enable the thing that nagged them.

The third is the exception-night failure, and it's the one that makes this genuinely tricky. Real life isn't consistent. There's the deadline night, the movie with your partner, the late flight home. A system strict enough to work on a normal Tuesday ends up punishing the nights that are legitimately different, and every unfair nudge teaches the user one thing: this feature doesn't get me. But bend the other way, make it easy to wave off for a good reason, and suddenly every night has a good reason. Flexibility for the exception becomes a loophole for the rule.

So the three are stuck together. Turn the reminder up and you trade being ignored for being switched off. Turn it down and you're back to wallpaper. Add flexibility and the whole thing becomes infinitely snoozable. The job isn't picking a volume. It's building something that knows what kind of night it is, and shapes the hour before bed around that, instead of firing the same message at the same time and hoping tonight's different.

Who I'm Solving For

Start with the scale, because it's the reason this is worth getting right. Apple's active device count passed 2.5 billion at the end of 2025, and in the US, iPhones make up roughly 60% of new smartphones sold. Whatever Apple ships here doesn't nudge a niche, it nudges a meaningful chunk of the country's bedtimes.

But "every iPhone user" isn't a user, it's a market, and designing for a market means designing for nobody. The 16-year-old on TikTok at 1am, the nurse coming off a night shift, and the person who's never once thought about their bedtime all count, and they need completely different things. So I'm cutting it tighter.

Who I'm solving for is the person who's already set a bedtime and keeps blowing past it. That's the most interesting group by far, because they've already told Apple what they want. Nobody has to convince them sleep matters, they agree. The problem is the gap between what they decided at 2pm and what they actually do at 11:45pm, and closing the gap between intent and action is exactly what a product is good at. It's still an enormous group, so I'm not giving up scale to get focus.

Who I'm not solving for, yet, is the person who's never set a bedtime at all. That's an adoption problem, getting someone to care in the first place, and it's a different case study. Mixing it in would blur both.

And here's the part that shapes everything downstream. It's tempting to build this around types of people, the night owl, the early riser, the shift worker. But a person isn't the same every night. The same night owl has a deadline Tuesday, a lazy Saturday, and a 6am flight Thursday. So the model doesn't key off who someone is. It keys off what kind of night it is: what they're doing right now, what's waiting for them tomorrow, how they slept last night, and whether tonight is a normal one or a real exception.

The Model: Four Kinds of Night

If the model keys off the night, the first job is sorting nights into kinds. I landed on four, and the test for each was simple: does this kind of night need a genuinely different hour before bed? If not, it doesn't get its own type.

A normal night is the routine. Nothing special tomorrow, nothing off about today, so the wind-down runs as planned. A high-stakes night has something waiting on the other side, an early alarm, a big meeting, a flight, so the wind-down starts earlier and leans a little harder. A recovery night follows a bad one. The Watch already knows you slept five hours, so tonight the system is protective, not just routine. And an exception night is the legit late one, the deadline, the movie, the late landing, where the right move is for the system to step back instead of nagging.

The obvious objection is that the system will guess wrong, and it will. Calendars hold work meetings, not evenings. Nobody puts "dinner with friends" in there. The tempting fix is to just ask every evening which kind of night it is. But that's a prompt at the same time, in the same shape, every single night, which is the exact thing that turned the original reminder into wallpaper. Within two weeks people would tap the first option without reading. So instead of asking, the system guesses, tells, and lets you override. It picks a night type and says so in one quiet line, "early start tomorrow, wind-down starts at 9:45," and doing nothing means yes. Changing it is one tap. It only actually asks when it's genuinely unsure, and every override teaches it something, so the Thursday it keeps getting wrong eventually stops being wrong.

Underneath all four sits one principle: you can't just take scrolling away, you have to replace it with something. Telling someone to stop is a fight. Handing them something better to do is a trade.

The Hour Before Bed

The original reminder makes one move: it interrupts you, once, at a fixed time, and hopes. That's a fight it loses every night, because one interruption against an endless feed is no contest. So the core idea here flips it. The phone shouldn't interrupt you once, it should get gradually more boring. Fade, don't fire.

This is where Apple's position actually matters. A sleep app can send you a notification and that's about it. Apple owns the whole phone, so it can change the room instead of shouting in it. Notifications go quiet. The screen warms and dims step by step. Near the end, color drains to grayscale, and it turns out a feed is a lot less gripping in gray. None of these moves ask you anything. They just make the thing you're doing slightly less pulling, slightly more often, until putting the phone down stops feeling like a sacrifice.

And at the right moment, it offers the trade: a sleep playlist, a podcast, an audiobook. All audio, on purpose, so your eyes get a break while your brain still gets something. That's the off-ramp, and it's what makes this a trade instead of a fight.

How far the fade goes depends on the night. On a normal night it runs gently and stops short of anything forceful. On a high-stakes night it starts earlier and goes further, and if the user has opted in, it ends with a screen lock. Opt-in is the whole point. A lock someone chose is a commitment. A lock someone didn't choose is the fastest way to get the feature switched off forever. On a recovery night it starts earliest and speaks softest, showing last night's sleep once as the reason, framed as protection rather than rules. And on an exception night it steps back almost entirely. One gentle check much later, "still up? Alarm's at 7," and no guilt in the morning. A system that punishes the nights you had a real reason to stay up teaches you it doesn't get you.

One thing I'm deliberately not pinning down is exact timing, how many minutes before bed each fade starts. I could make up a number, but it'd be confidence cosplaying as rigor. The shape is the decision. The minutes are what testing is for.

How I'd Measure It

The trap with any reminder feature is measuring the reminder. Did it fire, did people tap it, did they dismiss it. Those numbers are easy to get and almost useless, because a perfectly tapped notification that doesn't change when anyone sleeps is still a failure. So I'm measuring the thing we actually care about.

The North Star is nights on target: how often someone actually falls asleep close to the bedtime they set. Apple is in an unusual position here, because the Watch measures sleep directly. Most products have to guess whether they worked. This one can check. For people without a Watch, the last phone unlock of the night is a decent stand-in. Not perfect, but it answers the right question.

Under that sit three inputs that tell me why the number is moving. Off-ramp pick rate tells me whether the trade is working, whether people actually reach for the playlist or podcast instead of the feed. Override rate tells me whether the night-type guess is any good. If people keep correcting it, the model's wrong, not the user. And fade depth, how far into the wind-down people get before the phone goes down, tells me whether the gradual approach is landing or whether everyone's holding out until the very end.

Then two guardrails, one for each way this breaks. Feature disable rate watches the switched-off failure. If it climbs, we pushed too hard, and no improvement in nights on target is worth losing those people for good. And exception-night creep watches the loophole. If the share of nights marked "exception" keeps growing, flexibility is eating the rule.

What success looks like is nights on target going up while disable rate stays flat. That's the shape that matters, because it means we didn't trade one failure for another. We beat the tradeoff instead of just picking a side of it.

Tradeoffs