Artificial intelligence

A dashboard is read or it dies: the ten-minute review

Most dashboards are not bad, they are abandoned. What keeps one alive is a habit rather than a feature.

Published on 1 July 2026 — Algeria Agency

There is a trajectory everybody will recognise. A dashboard is delivered, everybody looks at it for two weeks, then somebody opens it less and less, then nobody. Six months later a new dashboard is requested, and the cycle starts again.

The cause is almost never the tool. It is that no decision ever depended on what it displayed — and a screen whose reading changes nothing about what happens next is a screen people stop opening, which is rational behaviour.

This article describes the habit that prevents that: a short review, whose it is, the one gesture that makes it useful, and the signs a dashboard is dying. The companion article covers its design — a declared source of truth, one screen per decision, alerts on deviation — and none of it is repeated here.

It contains no chart and no figure, which is faintly ironic on this subject. Nothing it asserts concerns a quantity: it is a page about a habit, and a habit is not measured, it is held or not.

A dashboard nobody reads is not a dashboard

A dashboard is not an object, it is a practice. Delivered without the practice it is a web page displaying exact numbers with no effect on the business, which is indistinguishable from not having one.

The test is brutal and asked once: over the last three months, which decision was taken differently because of this screen? Not which was confirmed — confirming is pleasant and proves nothing — but which changed. If the answer is none, the problem is not that people are careless.

The consequence is that the question of what to display is secondary, and it is nonetheless the one that occupies every project. The first question is: who will read it, when, and what do they have the power to decide? An indicator read by somebody who can do nothing with it is inert, however accurate.

That leads to a rule that sounds severe and saves a great deal: display nothing nobody has the power to change. The interesting but out-of-reach figure belongs in an annual report rather than on a weekly screen, where it installs the habit of looking without acting.

The ten-minute review, and who has to be in it

The form that survives is short, fixed and standing. Ten minutes, the same day each week, dashboard on screen and nobody seated. Brevity is not a style: it is what stops the review becoming a meeting, and a meeting gets moved, then cancelled.

Three people are enough and two are the minimum. The one who can decide, the one who knows the ground — whoever answers customers, runs the workshop, packs the orders — and, at the start only, whoever built it. The third has to disappear after a few weeks, and if they cannot, the screen is too complicated.

The agenda is the same every time and fits in three questions: what moved, why, and what are we doing. Nothing else. The temptation to comment on the figures that did not move is strong and fills exactly the time available without ever producing a decision.

The day matters more than people think. A Monday review covers a finished week and allows decisions for the one starting; a Friday review covers a week nothing can be done about. It is a free detail and it changes how many decisions get taken.

The one gesture that matters: going down a level

All the value of a review sits in a single repeated gesture: see a number that moved, then go down a level to find out what it is made of. A review where nobody ever goes down is a reading aloud.

Going down means: does this total come from all products or one; from every week or one day; from all customers or two large ones. The answer is nearly always from one, and that sentence is what produces an action — never the total.

It is also the criterion that judges a dashboard better than any feature list: can you go down in one click, during the review, without leaving the screen? If somebody has to be asked for an export, the descent will not happen, because nobody interrupts a ten-minute review to wait for a file.

Two levels are nearly always enough. The total, then its breakdown. A third level helps one person in ten and lengthens the review for the other nine — and a screen allowing indefinite descent becomes an exploration, which is a different activity with no place at that moment.

What to look at when nothing moved

One week in three, nothing notable happened, and that is where the review is lost: with no event, the discussion drifts into impressions, or the review is cancelled since there is nothing to say.

Good practice is to hold a spare question, decided in advance and different each month. It is not about the week but about a hypothesis: do customers who buy twice come from a particular channel? do Saturday orders look like weekday ones?

That question is also what stops the dashboard being only for surveillance. A screen used only to check nothing is wrong ends up being opened only when something is wrong, which is precisely the moment it is too late for it to help.

A quiet week is finally the right moment for the one maintenance task that matters: looking at which screens nobody has opened in a month. That is section eight, and it gets done in the empty weeks rather than never.

The variation that is not one

The central difficulty of a weekly review is telling a movement from a flicker, and human instinct is poor at it: we see trends in three points.

Two guardrails are enough and neither is technical. The first is always looking at the same thing over a longer span than the period discussed: a week is read against twelve weeks. An isolated figure means nothing; the same one with a year of context comments on itself.

The second is the question of size: how many things is that percentage calculated on? A twenty per cent rise on five orders is one order. That question has to be asked out loud every time a percentage is spoken, and it settles half the false alarms.

It also has to be accepted, explicitly and as a group, that a variation may have no identifiable cause. We do not know why is a legitimate conclusion and it is nearly always better than the plausible cause that would have been found to close the discussion — because the plausible cause will then be remembered as a fact.

The figure nobody disputes

In every dashboard lives a figure everybody quotes and nobody has ever verified. It has been there from the start, it looks reasonable, and it has become furniture.

Those figures are dangerous precisely because they are consensual: they enter reasoning without being examined in it, and an error in one propagates to every decision resting on it, for months, with nothing stopping it.

The counter-measure is a modest annual habit: once a year, take the three most quoted figures and recalculate them by hand, over a short period, from the raw data. It is half a day. It finds something more often than people expect, and when it finds nothing it makes those three figures genuinely reliable rather than merely old.

The moment to do it is when you need them to be right: before an investment decision, before setting targets, before showing them to somebody outside. A figure verified the day before a decision beats one verified monthly on principle.

When the data contradicts what the team sees

This is the most useful moment in a review and the one handled worst. The dashboard says enquiries went up; the person on reception says the phone rings less. Both are sincere.

The wrong response is to settle it immediately in favour of the figure on the grounds that it is objective. It is, but it measures what it was asked to measure, and the discrepancy nearly always reveals a definition rather than an error: the enquiries counted may include a form nobody treats as an enquiry.

The right response is to treat the contradiction as the subject rather than as an obstacle. One question settles it: what exactly does that figure count, and what is the person counting when they say fewer? In most cases they are not talking about the same thing, and the dashboard comes out of the conversation with a sharper definition.

The other half of cases is more interesting: the figure is right and the impression wrong, or the reverse. Either way it is only discovered if somebody from the ground is in the review — which is section two, seen from this particular case.

The decision written down, and the one forgotten

A review produces decisions and they evaporate. A week later nobody remembers whether something was decided, nor what effect was expected — and the next review starts from nothing on the same subject.

The useful record is short: the date, what was decided, and above all what is expected to move. That third part is the forgotten one and the only one that allows learning: with no written expectation you cannot know whether the decision worked, you can only observe what happened and tell yourself it was intended.

Three lines in a file are enough. It is not a minute of a meeting and must on no account become one, because a long document will not be reread and its only function is to be reread at the start of the next review.

That opening reading — thirty seconds on last week decisions — is what turns a series of reviews into a cumulative practice. Without it you get fifty-two independent reviews a year, which is an expensive way of learning nothing.

The screens that die, and how to notice

A dashboard does not degrade all at once: it loses screens one by one, silently, and nobody notices because a screen nobody opens does not complain.

The sign is measurable and it is the only figure this page recommends looking at: which screens were opened in the last month, and by whom. Most tools know and nobody asks them.

A screen unopened for a month should be removed rather than improved. It is the rule its neighbour sets at design time and which is worth nothing unless somebody applies it afterwards; removing costs thirty seconds and makes the rest more readable, while improving a screen nobody looks at is the best way to spend a day with no effect.

There is an exception to write into the same rule: seasonal screens, useful only at one time of year. They get marked as such once, otherwise they are deleted every winter and requested again every summer.

What breaks a series, and why nobody mentions it

Broken series are the main source of false conclusions in a review, and they almost never come from the measuring tool. They come from the business: a renamed category, two merged, a new channel, an altered form.

The problem is not the break, it is that it is invisible in the weekly reading. The figure changes, everybody looks for a commercial cause, and the real cause is that you stopped counting the same thing. The discussion can run for weeks and end in a decision entirely founded on an artefact.

The defence is a habit of declaring rather than a mechanism: whoever changes something touching what is counted says so at the next review, in one sentence. That assumes people know what is counted, which is a further good reason to keep the dashboard small.

Note breaks where decisions are noted, in the same file and with their date. Six months later somebody will look at a curve and see a step; the only thing stopping them inventing an explanation is a line saying the definition changed on that date.

Taking over a dashboard you inherited

The case is common: you arrive, a dashboard exists, nobody knows who made it or why certain screens are there. The reflex is to redo everything, and it is nearly always premature.

Start with the usage inventory from section eight: which screens were opened last month. The answer is generally two or three out of fifteen, and it tells you what the business actually looks at, which is never what the structure of the tool suggests.

Then for each of those two or three, ask the section one question: which decision depends on it? You will get either a clear answer — in which case you have found the core to keep — or an awkward silence, which teaches you the tool is looked at out of habit.

Touch nothing else for a month. An inherited dashboard holds figures whose definitions you do not know, and altering them before seeing them live through a season is the best way to break a series somebody was using without telling you.

What we do, and what we refuse

Our share is making the review possible: going down a level in one click, the screen usage inventory, the file of decisions and breaks, and being present at the first reviews to demonstrate the gesture. Then we leave, and that is the objective — a review needing us every week has not been installed, it has been rented.

We refuse to run the review in your place. It only has value if the people present can decide something at the end, and we cannot. A supplier chairing your weekly review produces a commented report, which is exactly the inert object the first section describes.

We also refuse to add screens to a dashboard whose usage inventory shows two of fifteen are opened. The request comes regularly and it is sincere; meeting it would be selling work we know will not be looked at. Removal is the useful service at that moment, and it costs less.

Finally, the part that decides everything needs no tool: fix the day and time of ten weekly minutes, decide who is in it, and write three lines at the end. Do it once this week on the dashboard you already have, imperfect as it is. In a month you will know whether you are short of screens or you were short of a habit.

Frequently asked questions

Why is our dashboard no longer consulted?

Almost never because of the tool. Apply the test: over the last three months, which decision was taken differently because of that screen? Not confirmed — changed. If the answer is none, opening it serves no purpose and stopping is rational. The first question is not what to display but who reads it, when, and what they have the power to decide.

What does a good weekly review look like?

Ten minutes, the same day each week, dashboard on screen, nobody seated, and three questions: what moved, why, what are we doing. Two people minimum — the one who can decide and the one who knows the ground. Monday rather than Friday: an end-of-week review covers a period nothing can be done about.

How do you know a variation is real?

Two guardrails, neither technical. Look at the same thing over a longer span than the period discussed — a week against twelve. And ask out loud how many things the percentage is calculated on: a twenty per cent rise on five orders is one order. Accept too that we do not know why is a conclusion: the plausible cause invented to close the discussion will be remembered as a fact.

The dashboard contradicts what the team says. Who is right?

Usually both, and the discrepancy reveals a definition rather than an error. One question settles it: what exactly does that figure count, and what is the person counting when they say fewer? In most cases they are not talking about the same thing. It is also why somebody from the ground must be present: without them the contradiction never surfaces.

What should be done with screens nobody opens?

Remove them rather than improve them. Most tools know which screens were opened last month and nobody asks. Removing costs thirty seconds and makes the rest readable; improving a screen nobody looks at is the best way to spend a day with no effect. Mark seasonal screens separately, or they are deleted every winter and requested every summer.

We inherited a dashboard nobody understands. Where to start?

With the usage inventory: which screens were opened last month. The answer is generally two or three out of fifteen, and it tells you what the business actually looks at. For each, ask which decision depends on it. Then touch nothing else for a month — an inherited dashboard holds figures whose definitions you do not know, and altering them breaks series somebody was using without telling you.

Where we come in

Ten minutes in the same slot every week decides everything, and it needs no tool. What is missing is being able to go one level down inside those ten minutes.

We make every figure clickable through to the line that explains it.
We record which screens were opened last month, and by whom.
We remove the ones nobody opens, rather than adding more.

We will not hold the review for you: a meeting where nobody present can decide is a meeting that does not need to exist.

See the AI analytics & dashboards page Talk it over on WhatsApp

Let us talk about your project

A free audit, no commitment: we look at your online presence and tell you what is holding it back.

Request the free audit Write to us