Measuring a thing precisely and controlling it are different skills, and the gap between them is where most startups actually die. That gap has been widening for a decade, because instrumentation keeps improving and the willingness to be corrected by it does not come bundled with the software. You can chart a storm in perfect resolution and still get soaked. The chart never learned how to stop rain.
Stewart Butterfield’s team spent years and real money building a game called Glitch, and it flopped. What survived the wreckage was the internal chat tool they had cobbled together to coordinate while building it, and that leftover became Slack. The company did not set out to build workplace software. It set out to build a game, noticed what people actually kept using, and rebuilt itself around the accident. That is the real shape of finding the thing customers want, and it is a much less flattering story than the one most pitch decks tell.
What Slack Learned from the Failure of Glitch
Notice what did not happen. The plan was not validated by research. It was discarded by evidence, which is a different and considerably more painful process. Nobody at Tiny Speck sat in a room and reasoned their way to a chat product, and if they had tried, the reasoning would have produced a worse product than the one that emerged from watching what people did when nobody was designing for them.
They shipped something, watched the data contradict the thesis, and had the discipline to believe the data over the thesis. Every part of that sentence is harder than it reads. Believing the data means conceding that the years and the money went into building the wrong thing, in public, with a team that joined for the original idea. The instrument that produced the signal was cheap. The willingness to act on a signal that indicted the plan is the expensive part, and it is the part no tool supplies.
Why More Dashboards Do Not Mean More Control
That discipline is rarer now than it sounds, partly because founders today have more instrumentation than any generation before them and mistake the abundance for progress. Dashboards refresh in real time. You can watch churn happen at the click level, segment it by cohort, and trace it back to a specific release, all before lunch. It looks like control, and it feels like control, which is worse.
It is not control. What the dashboard actually delivers is better information about a system you still do not run. The market moves on its own logic rather than on your roadmap’s, and no amount of additional resolution changes whose logic is operating. High-resolution visibility into something you cannot steer produces a specific failure: you spend your attention on explaining the movement rather than on deciding what to do about it, because explanation is available and action is not.
The same trap shows up in investing, where it is even better funded. Every retail platform now hands you real-time price feeds, sentiment scores, options flow, and alerts down to the second. All of it is genuinely useful information, and all of it is a trap for the same structural reason: precision about a system is not leverage over it. The market does not submit to your view just because your view is well-instrumented. The traders who do fine over long stretches tend to look almost boring by comparison, holding a short list of convictions and updating them slowly, and they are not seeing less than everyone else. They have stopped mistaking visibility for control, which frees up the attention that everyone else is spending on the feed.
Fewer Bets Make the Signal Legible
So what separates the founders who catch the signal from the ones who drown in their own metrics is not more data. It is fewer bets. The startups that survive the early years almost always describe the same move in hindsight: they stopped spreading time and money across everything that was defensible and narrowed hard, usually down to two or three things they were willing to be publicly wrong about.
The reason is not that focus is a virtue in the abstract, though people will tell you that afterward. It is that a scattered company cannot tell the difference between a real signal and noise, and the inability is arithmetic rather than temperamental. When you are doing twelve things at once, every metric is trending somewhere, every trend has three plausible causes, and any story you want to tell about the week has supporting evidence. When you are doing three things, you know what moved and why, and you know it fast enough to act while the information is still worth something. Narrowing does not just conserve resources. It raises the signal-to-noise ratio of everything the instrumentation reports back to you, which is the actual reason it works.
Hostile Customers Are Data, Not Threats
Where all of this gets tested in real time is customer hostility, and it is the part founders handle worst. Someone posts publicly that your product wasted their week, or that your support team did, or your billing did, or that the thing you shipped broke their workflow the day before a deadline. The instinct is defensive: explain the context, minimize the scope, wait for the thread to fall off the timeline.
The better move, and the harder one, is to treat the hostile customer as a live data point rather than a threat to be managed. Reply constructively. Fix the actual thing. Do it in public if the complaint was public, because the audience for that exchange is not the person who complained. You cannot control whether someone likes you, and trying to is its own instrumentation trap, the customer-relations version of believing that a chart controls the weather. What you can control is whether the next hundred people who read the exchange see a company that responds like an adult or one that disappears into a support queue and hopes.
None of this is an argument against tools, which would be a foolish argument to make. It is an argument about what tools are for. Slack’s founders did not need less data about Glitch in order to find the exit; they needed the willingness to let the data overrule the plan, and the data was useless without it. The founders who survive customer anger are not the ones with the better crisis playbook, they are the ones who stopped needing the customer to be wrong. The investors who last are not the ones staring hardest at the feed, they are the ones who decided in advance how few things they would actually act on.
The instrumentation always improves, reliably, every year, whether or not anyone gets better at using it. The humility to be corrected by it does not come standard with the dashboard, and it is the only part of the stack that was ever actually load-bearing.
Leave a Reply