The dashboard nobody reads

We built a metrics wall for one of the portfolio companies in the spring. Thirty-one tiles. Real-time. It looked like the bridge of a ship.

Eleven days later I checked the analytics on the dashboard itself — which is a thing you should always do, and which almost nobody does — and found that after the first week, it had three regular viewers. All three were on the team that built it.

The dashboard wasn't wrong. Every number on it was accurate, live, and correctly computed. It was simply unreadable, in the specific sense that no one could look at it and know what to do next. And a number that doesn't change a decision is a number that costs money to maintain and returns nothing.

Why it failed

Three reasons, and I think they're the standard three.

It answered a question nobody was asking. The tiles were organized by data source — everything we could get from Stripe, everything from the product, everything from support. That's a convenient way to build a dashboard and a useless way to read one. Nobody arrives at a screen wondering "what does Stripe know." They arrive wondering "are we okay, and if not, where."

It had no thresholds. Thirty-one numbers, no indication of which ones were fine. A reader had to hold the healthy range for every metric in their head to extract any signal at all. That's not a dashboard, that's a quiz. And when everything requires interpretation, people stop interpreting.

Nobody owned any of it. It was "the dashboard." Group ownership of a number is the same as no ownership of a number, and it shows up as a specific failure: a metric goes bad, everyone sees it, everyone assumes someone else is on it, and it stays bad for a month in full view of the entire company.

What replaced it

A weekly post in a shared channel. Eleven numbers. Plain text. Generated by a cron job at 8am Monday.

Each line looks like this:

onboarding · median time to first import
  4.1 days (target < 3)  ▲ 0.6 from last week   owner: @dana

That's it. The number, the target, the delta, and a human being's name.

It is dramatically uglier than the wall of tiles. It is also read by everybody, every week, and it has produced more course corrections in two months than the dashboard did in its entire lifespan.

What made the difference

The target. A number without a target is trivia. Adding "target < 3" converts every line into a yes-or-no question that anyone can answer in half a second without knowing anything about the business. This one change did more than the other three combined.

The delta. Direction beats level almost every time. A metric sitting outside its target but improving for six straight weeks needs no intervention. A metric comfortably inside its target that has moved the wrong way three weeks running is the most valuable thing on the page — it's a problem you get to fix while it's still small.

The name. Not a team, not a function. A person. It ends the diffusion of responsibility instantly, and — this surprised me — the owners like it. Being publicly accountable for a number turns out to be much less stressful than being vaguely accountable for an area, because a number tells you when you're done.

The push. The dashboard waited to be visited. The post arrives. That distinction is the entire ballgame: the gauge that requires a decision to look at it will eventually stop being looked at, and it will stop precisely during the busy weeks when the readings matter most.

The uncomfortable finding

Going from thirty-one numbers to eleven meant deleting twenty numbers that were real, accurate, and genuinely interesting.

That was harder than it should have been, and the argument for keeping each one was always the same: "but what if we need it?" The answer is that it still exists. It's queryable. Nothing was deleted from the warehouse — it was removed from the attention surface, which is a completely different and far scarcer resource.

The wall of tiles wasn't a measurement system. It was a hedge against the discomfort of choosing. Picking eleven numbers means saying out loud which parts of the business you're watching, which means being wrong in public when the failure comes from somewhere you weren't looking. Thirty-one tiles let you avoid that by watching everything, which is the same as watching nothing while feeling responsible.

The rule we kept

A gauge that doesn't change a decision isn't a gauge. It's a fact, and facts are cheap now — the marginal cost of computing another number rounds to zero, which is exactly why discipline about which ones you display matters more than it used to.

Before a number goes on the weekly post, it has to survive one question: what would we do differently if this moved? If nobody can answer, it doesn't go on. It can live in the warehouse with the rest of the facts, available to anyone who has a reason to ask.

The pipe is supposed to be clear. Clear is not the same as crowded.