The dashboard nobody opens is not a design problem

The dashboard went live in March. There was a session to walk everyone through it. For about three weeks the usage graph looked healthy, and then it went flat, and by June the weekly management pack was a spreadsheet again — the one someone rebuilds by hand every Monday morning.

Nobody decided to stop using it. It just stopped answering the question people actually had.

The number is worse than it feels

Gartner's tracking puts it plainly: 29% of employees use the BI tools their employer pays for. Only 16% of organisations reach full Power BI dashboard adoption, and 58% sit below a quarter.

Read that as a purchasing decision rather than an engagement metric and it gets uncomfortable. The licence count is set by headcount. The usage is set by whether people believe the thing. The difference between those two numbers is a line item that renews every year and returns nothing.

The usual fixes aim at the wrong thing

When adoption stalls, three responses come up, in this order, almost every time.

More training. If people were not using it because they could not, this would work. Usage would climb after each session and stay there. It climbs and decays instead, which is the signature of a motivation problem wearing a skills problem's clothes.

A redesign. Cleaner layout, fewer charts, better colours. This produces a nicer artefact that gets used exactly as much as the old one, because the reason it went unused was never legibility.

A mandate. Usage becomes a target. People open the dashboard, look at it for four seconds, and then go and check the number the way they were checking it before. You have bought yourself a metric, not a decision.

Each of these treats adoption as something you can do to the dashboard. It is not a property of the dashboard.

What is actually happening

75% of leaders say they do not trust their data for decision-making. A separate 2025 data integrity survey put the share of organisations that do not completely trust their data at 67%, up from 55% the year before — moving in the wrong direction, during the years everyone invested most heavily in analytics.

That is the whole explanation. People do not avoid a dashboard because it is ugly. They avoid it because they are not willing to be the person who repeated its number in a meeting and was wrong.

And trust is not something a dashboard can contain. It is a property of everything upstream of it.

Ask someone why they rebuild the spreadsheet and the answer is almost never about the tool. It is: last quarter it showed a different revenue figure to the one finance uses, and nobody could tell me which was right. One unexplained discrepancy is enough. After that, checking manually is not laziness — it is a rational response to having been burned once.

What repairs it

The work is unglamorous and it is not on the screen.

Give every number one owner, by name. Not a team. A person who can be asked "why did this move?" and will answer. An unowned metric is one nobody will defend, and a metric nobody will defend is one nobody will act on.

Agree the definition before you build the visual. Most conflicting-dashboard problems are two departments correctly calculating two different things and both calling it revenue. This is a half-day argument that people postpone for years because it is a conversation, not a task. Have the argument. Write down what won.

Show the number's provenance next to the number. Last refreshed at, sourced from, defined as. It sounds like clutter. It is the difference between a figure a manager will quote and one they will double-check — and double-checking is how you got the spreadsheet.

Instrument what gets used, then delete the rest. If a report has not been opened in ninety days, it is not neutral. It is a maintenance cost and one more thing that can disagree with something else. Fewer, trusted, owned reports beat a catalogue.

Fix one decision, not one department. Pick a recurring meeting, find the number it turns on, and make that number unimpeachable. Adoption spreads from a decision people had to make anyway — never from a launch.

The bit that is about to get harder

Gartner expects 80% of data and analytics governance initiatives to fail before 2027, largely because nothing forces the issue until something breaks publicly. They also predict that by 2028, half of organisations will adopt a zero-trust posture toward their own data, driven by the volume of unverified AI-generated content entering the pipeline.

Layering AI-generated analysis on a dataset nobody trusts does not resolve the doubt. It moves the doubt somewhere harder to inspect. If you cannot currently answer "who owns this number and how is it defined", a Copilot summary of that number inherits the problem and hides it behind fluent prose.

The order matters: earn the trust first, then automate on top of it.

Where to start

You do not need a governance programme to begin. You need one meeting, one metric and one name against it.

If the manual spreadsheet is still being rebuilt every Monday, that spreadsheet is the specification. Whoever maintains it has already decided which numbers are trustworthy and where they come from — that judgement is the asset, and it is usually undocumented and held by one person.

We build reporting people actually use as part of our data and analytics work, and we teach the modelling and governance side of it in our data science programme. If the dashboards are already built and simply not trusted, that is a shorter conversation than a rebuild — tell us what the Monday spreadsheet contains.