Why we deleted our own dashboards

We deleted our dashboards. Not the customer-facing ones, the internal ones we ran the company on. The funnel board, the usage grid, the revenue tiles, the pipeline view someone built in a weekend and we all bookmarked. Gone. We replaced the entire wall of charts with two things that come to us, and we got faster, not slower.

A dashboard is a question you have to remember to ask

A dashboard does nothing until you go look at it. It sits there, perfectly up to date, waiting for you to remember it exists, open it, scan it, decide which of the twelve numbers moved in a way that matters, and then work out why. It's a question you have to remember to ask, then go read, then interpret. Three jobs, and the dashboard does none of them.

So the failure mode is not that the chart is wrong. The chart is fine. The failure mode is Tuesday afternoon, when the account that mattered started slipping and nobody had the usage board open, because nobody has a board open at the moment an account starts to slip. You find out on Friday at the review, by which point it is not news, it is a postmortem.

We built tools that surface everything and notify you of nothing, then blamed people for not checking them. We ran those dashboards too. That is pull. You go to the data. The data never comes to you.

Push beats pull, but only if the push is trustworthy

The fix is to flip the direction. Instead of a system you check, a system that tells you what changed and what needs you. Not a notification firehose, which is just a dashboard that interrupts you, but a short, ranked account of what actually moved since you last looked, and what you specifically should do about it.

Two things make that possible, and we had to build both before we could delete anything. The first is the brain, a model of every account, deal, and project, built from the real activity that touches them, with a source on every fact. It tracks what normal looks like for each one. So it can tell that Northwind's usage dropping 30% for the third week running is a signal, while the same drop on a trial account in week one is nothing. That gap, between a useful sentence and an alert you learn to ignore, is the whole reason we could trust a push at all.

The second is Walle, the teammate that reads the brain and brings you the short list. Not 'here is a chart of activity.' More like: 'Northwind is down 30% three weeks running, renewal is in six weeks, the last thread with them was a pricing complaint, here is the source on each of those, and I have drafted a check-in for you to send.'

By Friday at the review it is not news, it is a postmortem. The point is to catch it on Tuesday.

The source is what makes push survivable

Here is the trap with any system that tells you what matters: the first time it tells you something wrong, you stop believing all of it, and now you are back to checking the dashboard, except you also have an inbox of useless pings. A push you cannot verify is worse than a pull you can.

So every line Walle brings you is clickable down to where it came from. The 30% is a query you can open. The renewal date is a record. The pricing complaint is the actual thread. We did not build a system that summarizes and asks you to trust the summary. We built one that summarizes and shows its work, because the only push worth reading is one you can audit in a click or two and then forget about.

Why this stays approval-first

Flipping from pull to push does not mean handing the wheel to the software. Walle surfacing 'Northwind needs a check-in' and drafting the email is not the same as Walle sending it. The drafting is automatic. The send waits for you. That is the line we hold everywhere: the AI does the noticing, the gathering, and the proposing, and the consequential call stays with a human. Push the information freely. Never push the irreversible action without a person saying go.

The Northwind situation now reaches whoever owns it the same afternoon it starts, not at Friday's review. The slip that used to become a postmortem becomes a check-in sent on Tuesday. Nobody opened a board and squinted at whether a line was good or bad news, because the things that actually matter, the account that goes quiet, the deal that stalls in silence, are exactly the things that are hard to chart and easy to say in a sentence. A dashboard is good at what is easy to plot. It is terrible at what is easy to miss.

We are not anti-data. The source sits under every claim. We're anti-homework. Make the system come to you, show its work, and wait for your call on anything you can't undo. Then delete the dashboard. We did, and we're not putting it back.