A few years ago, I read Matthew Walker’s Why We Sleep. It contains a lot of scary assertions:

  1. 99% of the population needs 7-8 hours of sleep. Only 1% of the population carries a gene mutation that makes them resilient to sleep deprivation. You and I aren’t in that 1%.
  2. Poor sleep is correlated with increased rates of cancer and heart disease. The WHO has even classified shift work as a probable human carcinogen.
  3. The less we sleep, the sooner we die. If the above wasn’t enough, poor sleep leads to diabetes, Alzheimer’s, and DNA damage.
  4. You can’t reverse the damage. We can’t build up sleep debt during the week and pay it back on the weekend. The damage caused by sleep deprivation is irreparable.

People have also criticized Why We Sleep. For example, this withering rebuke by Alexey Gusey argues that Walker misrepresented, cherry-picked, and invented facts to suit his narrative.

Who do we believe? I don’t know. I’m not a doctor or scientist, and I don’t play one on the internet. I do know that I feel and perform better when I sleep well. So, sleep became a project for 2020. I improved my sleep in the same way that I would improve a product as part of my “day job”: Set a goal, define a way to measure success, and test.

Like most things in life, I learned that most of what we do doesn’t matter, but some things work really well. These are the 10 rules that emerged after a year of experimentation.

“You Can’t Manage What You Can’t Measure”

I doubt Peter Drucker was thinking of sleep optimization when he said this, but it’s certainly applicable. I bought an Oura ring to set a baseline and measure whether my sleep was getting better or worse. Data is less fallible than perception, and I wanted a way of quantifying my experiment.

The Oura ring validated what I suspected: I wasn’t sleeping particularly well. I have high “restlessness,” and I wake up multiple times each night. Starting, this was an average night:

An average night’s sleep in early 2020.

8 Hours in Bed Doesn’t Equal 8 Hours of Sleep

One of the first things you notice when you start tracking sleep is that we’re inefficient. My sleep “efficiency” ranges from 75-90%: I’ll get ~6.5 hours of sleep for every 8 hours I spend in bed. If you want 8 hours of proper sleep, expect to spend 9-10 hours in bed.

Cool, Dark, and Quiet

You’ve probably heard of the Pareto principle: 80% of our results come from 20% of our effort. Our sleep environment—temperature, light, sound—is the Pareto principle of sleep. If you control these, you’ll almost certainly improve your sleep. No “hack” can compensate for a poor sleep environment.

Case in point: We recently moved to a new apartment. Unlike most San Francisco apartments, this one has air conditioning. And unlike our old apartment, it’s also quiet. I set the temperature to 67°F every night (the ideal sleep temperature being 60-67°F). This cool and quiet combination made a profound difference, with Oura showing a ~15% gain in my sleep quality almost immediately.

A few other tools for augmenting our sleep environment:

Supplements Don’t Do Much (Unless They’re Targeted)

I tried 11 different supplements. Some of them feel like they’re working, but most don’t make a measurable difference. I suspect this is because they aren’t solving the right problem. Taking magnesium might help if you’re magnesium-deficient, but what if you’re not?

Supplements can work when applied to a specific problem. As I mentioned, I struggle with waking up during the night. I stumbled upon research that linked this to increases in cortisol. When our cortisol rises, we wake up. I did some more digging and found an over-the-counter supplement called Phosphatidylserine that decreases cortisol. I take it selectively—usually when I’m stressed— and it makes a measurable difference (increasing my sleep efficiency by ~7%).

Small Meals, Never Late

Without fail, a large meal before bed results in poor sleep, even worse when it contains red meat or sugar. The optimal time to stop eating is around 3 hours before bed. My sleep scores are consistently better—and I feel better—when I stop eating early.

No More than One Drink

I made a trip last year with friends who all have Oura rings. We drank more alcohol than usual, and we all had a ~50% reduction in sleep quality: high heart rates, reduced REM and deep sleep, and low efficiency. Experimentation suggests that two drinks are the limit before impacting sleep quality.

No Caffeine After Midday

Caffeine is an easy one: don’t drink caffeine late. The half-life of caffeine is 5 hours, so an afternoon coffee is almost certainly still in your system at 10pm. I only drink decaf after midday. I experimented with cutting out caffeine entirely for a month, which didn’t make a meaningful difference.

Exercise to Increase Your Need for Sleep

We all sleep well after a long hike. There’s a direct relationship between exercise and how well we sleep. The scientific concept is “sleep pressure,” meaning our body’s need for sleep. The more we exercise, the more “sleep pressure” we accrue. This helps us get to sleep and stay asleep.

I sleep best when I’m exercising regularly, ideally for 30-60 minutes a day.

Mental Health Is as Important as Physical Health

We’ve all had nights where anxiety has made it difficult to sleep. One study found that 3/4 of depressed patients displayed symptoms of insomnia.

I don’t suffer from depression, but I certainly some anxiety and stress. For me, anxiety usually indicates that I’m neglecting something important. It surfaces when there’s a disconnect between how I’m spending my time and what my mind thinks I should be doing.

Paradoxically, trying to sleep better can become a cause of stress, resulting in poorer sleep. Be mindful of this.

Found a Routine and Stick with It

Having a consistent sleep and wake schedule helps your body find a groove. You’ll feel tired at the same time and begin to wake up naturally before your alarm. My routine: start winding down at ~11pm, shower, tea, and then read on a Kindle until I’m sleepy.

My final recommendation, and what I do after $2,500 and 2,500 hours of experimentation: