The Most Dangerous Dashboard Is the One Everyone Trusts But Shouldn't.
In my first year building dashboards, I thought the goal was accuracy. Get the numbers right, get the visuals clean, get it out the door.
I was wrong. The goal is trustworthy accuracy — which is completely different.
The audit that changed how I build
At MQ Centre of Applied AI, I inherited a set of Power BI reports that had been in use for over a year. The team used them weekly. Leadership made resource decisions based on them.
I ran a lineage audit. Of the 14 key metrics on the main dashboard, 3 were calculating correctly, 8 had logic errors of varying severity, and 3 were pulling from deprecated data sources that no longer refreshed.
The team had been confidently wrong for months. And because the dashboards looked polished — nice colours, clean layout, consistent branding — nobody had questioned them.
What makes a dashboard feel trustworthy (regardless of accuracy)
Visual polish. Consistent number formats. Fast load times. Leadership saying "I use this every week." These all signal trustworthiness. None of them have anything to do with whether the underlying numbers are correct.
This is the trap. The better your design skills, the more dangerous your errors become.
What I do differently now
1. Every metric has a data lineage note. Not buried in documentation — right there in the report, accessible via a tooltip or info button. Where does this number come from? What's the filter logic? When was it last validated?
2. Every report has an accuracy date. Not just a data refresh timestamp. A "last validated" date — the last time a human checked that the numbers matched a trusted source.
3. I validate against source systems before UAT. Not "does it look right." "Does it match this SQL query against the raw data."
The dashboard that improved from 82% to 98% accuracy at MQ didn't just fix 12 metrics. It changed how the team relates to the numbers — from assumed-correct to verified-correct. That shift is worth more than any design improvement.