More data is not a free good
Every metric added to a dashboard takes something from the ones already there. Not storage, which is cheap, but the scarce thing: the number of figures a person can hold in their head on a Monday morning and actually act on.
The failure mode is specific and it is not confusion. Nobody is confused by a forty-tile dashboard. They simply stop reading it, then stop opening it, and then keep it up on a screen because it looks like rigour. A page that gets glanced at rather than read has become decoration, and decoration is the most expensive kind of reporting because it produces the feeling of being data driven without any of the behaviour.
There is a second cost. With enough numbers on a page, at least one is always up and at least one is always down, which means any story can be told from the same dashboard by whoever is telling it. Fewer numbers is not just easier. It is harder to argue with.
The test: name the action before you keep the metric
For each number on the page, answer one question out loud. If this drops by a fifth next week, what do we do differently on Tuesday?
If the answer arrives immediately, keep it. Signups fell, so we look at the page and the traffic mix. Reply rate fell, so we look at the subject lines and the list. If the answer is a sentence about monitoring, watching, or keeping an eye on it, then no decision is attached, and the number is being reported because it exists rather than because anybody uses it.
Applied honestly, this deletes most of a typical dashboard, including several numbers that are perfectly true. True is not the bar. Actionable is the bar, and the two are not related.
Sort what is left by timescale
The surviving numbers split into two piles, and the biggest single improvement most teams can make is refusing to look at them on the same day.
Weekly, and they are inputs. Things published, ads launched, emails sent, tests started, calls made. Counts of actions. They are boring, entirely within your control, and honest at a week.
Quarterly, and they are outcomes. Revenue, retention, share, review volume, rank. They cannot say anything useful about seven days, and asking them to is how good work gets cancelled early.
That split is the whole of leading versus lagging, and mixing the timescales is the most common analytics error there is: a quarterly instrument pointed at a weekly question gives an answer that is confident, precise and meaningless.
The short list
For most companies selling something, six or seven numbers do the job.
Actions shipped this week. One count, whatever your inputs are. It is the only number that is fully yours.
Traffic, split by channel. The total is nearly useless. The mix is where the news is, including the share now arriving from AI assistants, which most analytics cannot attribute at all.
Conversion rate at the one step that matters. Usually the step immediately before money, or before the first real use of the product.
Average order value or its equivalent. Because it moves with a price test and a price test takes an afternoon.
Repeat or retention. The number that decides whether acquisition is an investment or a leak.
One quality signal from the customers. Review sentiment, refund reasons, the complaint that keeps recurring. It is the only one on the list that tells you why.
Everything else earns its place by surviving the Tuesday question. Follower counts, badge walls and press mentions rarely do.
A number you cannot open is a rumour
The last property, and the one we are strictest about in our own product: every figure should open into the thing it was counted from.
A conversion rate you cannot decompose is a claim. A median price that does not open into the list of products it was calculated from is a claim. The moment a number cannot be traced, it stops being evidence and starts being something people either believe or do not, depending on whether they liked the conclusion.
That is why the figures on the Benchmark are clickable, and why a measurement we could not verify is left out rather than estimated. A dashboard's credibility is set by its weakest number, not its best one.
Pick one competitor. See what we find.
Your first benchmark is free. One domain, measured against you, no card and no call.