Use fuzzy metrics until you get good ones.

Early on, the things that matter most are the hardest to count. Is onboarding getting easier? Are we signing the right customers? Did the new pricing scare anyone off? Nobody has a dashboard for those. So the usual answer is to wait until the data's clean, the product is instrumented, and there are enough customers for the numbers to mean something.

The decisions don't wait. They get made anyway, on whatever's lying around: the last customer call, the loudest person in the room, the founder's gut on a good day.

I'd rather you measure it badly. On purpose, out loud, with a note on the number that says how it's off. I call those fuzzy metrics.

Three options, and two are the same

In any zero-to-one you have three choices. Run with no metrics. Pick metrics you believe are real, because you think they ought to measure the thing accurately. Or use fuzzy metrics.

The second and third are the same thing. A truly new product has never been measured before, so whatever number you pick, you're using what you believe is a proxy. The only difference is whether you say so.

Saying so is what makes the number useful. A team that knows its metric is a proxy checks it against other signals and keeps looking for a better one. A team that thinks its proxy is the truth starts optimizing the proxy, and wonders later why the business didn't follow.

Glasses off, mirror on

Anyone who wears glasses has gone without them for an afternoon. You still get around. You can tell the couch from the table and find the door. You just keep a thought running in the back of your head: I'm seeing this fuzzy, things aren't quite how they look, I'll account for that. You slow down where the shapes go soft.

Or think about the side mirror on a car. The warning is printed right on the glass: objects in mirror may be closer than they appear. Nobody stops using the mirror because of it. You check it, you change lanes, and you leave a bit more room than the picture says you need.

That's what a fuzzy metric is. You use it, and you know it isn't exactly the way you see it.

Blurry still beats closed. Most early teams think the choice is blurry or sharp. The real choice is blurry or eyes shut, and plenty of them pick eyes shut because it feels more honest than a number they don't trust.

Declare success and failure at the start

Here's what I tell teams in incubation, in zero-to-one, and in product generally. When you conceive the idea, say how you'll know it's working and how you'll know it's failed. Both. Write them down early, before anyone has built something they'd be sad to lose.

Your first answers will be rough. That's expected. As the idea develops, you learn how to actually tell. The way you measure success and failure needs to be developed just as much as the concept itself. Most teams pour months into the concept and give the measure an afternoon.

Failure criteria are the ones people skip. Once something's built, somebody will always find a reason to call it a success. A line you wrote before you started is much harder to argue with.

When I ran a concept of my own through G.I.V.E. (an AI intake kiosk for independent eye-care practices), every canvas carried failure criteria right beside the success criteria. One said the idea had failed if the practice's claim denial rate was unchanged ninety days after deployment. Another said it had failed if the doctor still had to ask the patient what brought them in. Neither was precise. Both were clear enough to argue about, which is the point. ("G.I.V.E. on one real decision" walks through the whole run.)

Benchmark, then optimize

On launch day you've never measured this thing before, so there's nothing to compare against. Make the first reading your benchmark. Write down what you think will happen and what it would mean if it did. Then act on the number, knowing you're not seeing it well, and sharpen it as you go.

A fuzzy number should carry three notes, written next to it where everyone can see them:

• How did we get it? "I tagged last month's support emails by hand." "We asked the ten customers who'd take a call."
• Which way is it probably wrong? The ten customers who'll take your call like you more than the ones who won't.
• By how much, roughly? A guess is fine. "Could be off by half" is a perfectly useful sentence.

Write that down and the number becomes something you can work with. Skip it and you have a bad number, which will turn up in a board deck six months from now looking like it came off a gauge.

Two habits keep the blur honest. Count the same way every time, even if the way is crude, because a consistent bad ruler still shows change and a new ruler every month shows noise. And cross-check with a second fuzzy number from a different angle, like what customers say against what the logs show. Where the two disagree is exactly where better data will pay off first.