Dashboard taxidermy: your dashboard died, you stuffed it

You know the Monday where someone asks “do we have a dashboard for that?” and you say “yes, actually” with a little too much enthusiasm. You send the link. They say thanks. Then you open it yourself for the first time in two months and notice the third chart has been showing an error since August.

Nobody told you. Because nobody saw it. Because nobody opens it.

Congratulations. You’re a taxidermist.

What dashboard taxidermy is

Dashboard taxidermy: keeping a dead dashboard preserved and posed so it looks alive. It refreshes on schedule. It’s linked from the wiki. It’s in the onboarding doc. Nobody has made a decision with it in a quarter.

It isn’t laziness. Deleting a dashboard feels like admitting the work was wasted, and keeping it costs nothing you can see. So it stays. Glass eyes, perfect fur, completely dead.

The real danger isn’t the wasted work. It’s that a stuffed dashboard still gets quoted. Someone finds it six months from now, assumes it’s maintained because it looks maintained, and puts the number in a doc.

Every stuffed dashboard goes through the same four stages.

Stage 1: The Launch Glow

Week one is great. The dashboard answers a question someone actually asked. There’s a kickoff, a walkthrough recording, a “this is super helpful.” Half the team opens it.

This is where you make the mistake that kills it, and it doesn’t look like a mistake. You built it as a place to go. A destination. Something people have to remember to visit.

People don’t remember to visit things. They respond to things that come to them.

Stage 2: The Top-Line Glance

By week three, traffic settles into its real shape: one person, once a week, for about four seconds.

They read the big number at the top. Up or down. Green or red. Then they close the tab.

And you feel it, that quiet little resentment. You built twelve charts. There’s a cohort breakdown in there you’re honestly proud of. There’s a funnel whose filters took a whole day to get right. All of it, unread, below the fold.

Here’s the uncomfortable bit: the stakeholder is right. They had one question, “is the thing OK?”, and the top line answered it. The other eleven charts were answers to questions they never asked. You weren’t building for them. You were building for a version of them with a free afternoon, and that person doesn’t exist.

Stage 3: The Silent Rot

Now that nobody’s really looking, things start to break.

An upstream column gets renamed and one tile starts erroring. A filter default points at a value that no longer exists, so a chart quietly shows nothing. A table stops loading, and the top-line number freezes at last Tuesday’s value. Still green. Still beautifully formatted.

Nobody notices, because the four-second glance only checks whether the number looks reasonable. And a frozen number looks extremely reasonable. It’s the most stable metric in the company.

This is where taxidermy turns dangerous. A dead dashboard nobody reads is wasted work. A dead dashboard one person still glances at every week is misinformation with a refresh schedule.

Stage 4: The Mount

Last stage. The dashboard is furniture now. It’s referenced in three docs and one OKR. Its owner on paper moved teams in the spring. Deleting it would need a meeting, and nobody wants to be the person who killed “the revenue dashboard.”

So it stays on the wall. Well lit. Seen by visitors. Never fed.

Be honest: you probably have a few of these right now. Maybe one you built.

The way out: make it talk, or let it die

A dashboard has two honest options. It earns attention by speaking up when something matters, or it gets archived. Everything in between is taxidermy.

Three moves.

Move 1: The Last-Decision Test

Go through your dashboards and ask one question of each: when did someone last decide something because of this? Not open it. Decide something.

If nobody can name a decision, archive it. Archive, not delete, so nobody has to win an argument first. You’ll miss fewer of them than you expect, and the ones you do miss, you’ll rebuild smaller.

Move 2: Shrink to the top line, on purpose

For the survivors, accept what stage 2 already told you. People read the top line. So make the top line excellent and treat everything else as optional depth.

One question per dashboard. One number that answers it. The rest is there for the day that number moves and someone needs to know why.

Move 3: Make it come to them

This is the big one. A dashboard that waits to be visited dies. A dashboard that taps you on the shoulder when its number goes wrong stays alive, because it earns attention instead of demanding it.

That means checks on the numbers that matter. Not “email me this dashboard every Monday,” which is just a screenshot people learn to ignore. Alerts that fire only when something is actually off: the table went stale, the row count fell, the metric left its normal range.

Then silence means something. If the dashboard didn’t speak up, the number is fine. Which is the only thing the four-second reader wanted to know in the first place.

Where arcodash fits

I’m building arcodash for exactly this, and it’s in early beta, so weigh what follows accordingly.

In arcodash a dashboard isn’t only charts. Its numbers can have checks attached, and the checks do the talking:

• Checks that alert when a number goes wrong. Freshness, row counts, nulls, or a baseline monitor that learns normal from the metric’s own history, so you don’t have to pick thresholds. That’s stage 3 handled: the frozen number gets flagged instead of glanced at.
• “Why did this fire?” in Slack. When a check fails, AI takes a first pass at what moved and puts it in the alert. The stakeholder gets the top line and the first clue in one message, in the place they already are. Treat it as a lead, not a verdict.
• AI summary on the dashboard. A few plain-language lines on what the charts are showing, each one linked to the chart it came from. For the four-second reader, that’s the format they wanted all along.
• Ask with AI. When the summary raises a question, anyone can ask it in plain English and get SQL they can read and edit, instead of filing a ticket for chart thirteen.

The AI runs on your own Gemini, OpenAI or Anthropic key. There are 19 connectors, and it’s from $59/month for the org with unlimited seats, so nobody has to ration who gets to see the thing.

Your homework

Pick your most-mounted dashboard, the one you already suspect is stuffed. Run the Last-Decision Test on it, honestly.

If it fails, archive it and feel nothing. If it passes, put a check on its top-line number so it can speak up the next time it’s wrong.

You can do that second part free for 14 days, no card, at arcodash.com. Start with one number and see if it has anything to say.