When I tell my homepod mini to turn off the living room lights, I’d guess it works about 95% of the time. The other five percent of the time, the homepod didn’t activate for some reason, or someone interrupted me when I was talking. Or the wifi was down for a second. Or it thought I was saying something else. Or it asks me which room again. Or, for some reason, the lightbulb lost its connection to the base unit. There’s a long chain of events where any breakage is a problem.

I like voice control enough that I’ve decided to live with this success rate. I could probably improve some of this if I tried, but it’s not really worth it to me. And I always have a fall-back solution of just turning off the lights the old-fashioned way with the switch.

One of my frustrations with digitization is that we’ve switched primarily from mechanical devices that operate nearly 100% of the time to digital devices that now work around 98% of the time. And that 2% is kind of a big deal.

When I bring this up, I’m often told I’m overreacting. It’s only one percent worse, right? Is that worth making a big deal out of?

Absolutely, it is! And the premise that it’s ‘only one percent worse’ is a deeply flawed statement. If my aforementioned lights went from a 1% connection rate to a 2% connection rate, it would be fair to say it’s twice as good. But if it went from 80% to 90% success rate, that would also be twice as good. When something is expected to work, the failure rate determines the percentages. In both, the failure rates are cut in half.

We often incorrectly state that something works 100% of the time, which is never true. It will break or have issues at some point. But these types of things, for example, a mechanical light switch, are more like 99.999%. So, if replaced with something that works 99% of the time, it should rightly be considered three orders of magnitude worse. Significantly worse. Do that to enough devices, and you’ll start to feel like everything is broken all the time. Those extra digits of reliability really matter.