A dashboard is a decision surface, not a place to put the data you have

Most dashboards are an inventory. Someone lists what the system records, puts each item in a card, and arranges the cards by how they look together. The result is dense, impressive in a demo, and useless on a Tuesday — because nothing on the screen tells the user what to do differently.

Design from the decision backwards

1. Name the decision in a sentence a customer would say: "which of these should I fix first?"
2. Name the one number that changes the answer to that sentence.
3. Define that number precisely: what is counted, over what population, with what denominator, in what window.
4. Decide what the number looks like when it is bad, and what the screen offers at that moment.
5. Only now choose the visual — and if a sentence would be clearer than a chart, write the sentence.

The denominator is the product

A percentage with an undefined denominator is not a metric, it is a rumour. "Visibility: 42%" means nothing until the screen can say forty-two percent of what — of which questions, on which engines, over which time window, excluding which failures. Two products can compute that number differently and both be internally consistent while telling a customer opposite things.

Write the definition down before the query. Every customer-facing number should be able to state its concept, its grain, its population, its denominator and its scope — and if one of those is unknown, the product does not get to show a confident number.

Build the refusal state first

Every analytics product has states where an honest answer is not available: not enough data, a provider failed, the comparison is not valid. Most products render a zero. A zero is a claim — it says the thing was measured and found to be nothing — and it is usually a lie. Design the not-measured state as carefully as the success state, with words that say which reason it is.

Every control must do something

A filter that does not change the result, a date range that only has one meaningful value, a segment dropdown with one segment — each of these teaches the user that the interface is decorative, and that lesson transfers to the controls that do work. Ship fewer controls and make all of them real.