Reading Dashboards Without Deceiving Yourself

Every modern workplace runs on dashboards, and most dashboard reading is quietly wrong — not because the numbers are fabricated, but because numbers don't lie; framings do. The skill worth building isn't making dashboards (tools handle that) but interrogating them: knowing which questions separate signal from decoration. This post is that question set.

Question 1: Compared to what?#

A number without a comparison is decoration. "4,200 visitors" means nothing alone; it becomes information against context:

• Last period (trend): up or down versus last week/month?
• The same period last year (seasonality): retail "surges" every December; gyms fill every January. Year-over-year kills most false alarms.
• A target or threshold: 4,200 against a goal of 5,000 tells a different story than against 3,000.
• Per unit: absolute totals flatter growth; per-user or per-session numbers reveal whether quality kept pace.

Dashboards showing bare current values without any baseline aren't informing you — they're performing. Good ones answer "compared to what?" before you ask.

Question 2: What's the denominator?#

Percentages hide their bases, and bases carry the story. "Signups doubled!" from 50 to 100 reads differently once you learn the ad budget also tripled. "97% satisfaction" from a survey of thirty voluntary respondents is a different artifact entirely than from all customers. Whenever a percentage moves, ask: did the numerator change, the denominator change — or both, in opposite directions? Some of history's most confident business decisions were made inside moving denominators.

Related trap: rates without exposure. "Our error rate is down 40%" — measured over how many attempts? A rate computed from three incidents is noise wearing percentages.

Question 3: Is this metric gameable — and does anyone benefit from gaming it?#

Every metric is a target, and targets get optimized — sometimes at the metric's expense (Goodhart's law in one line). Before acting on any KPI, ask who's evaluated by it:

• Support tickets closed rises when agents close tickets unresolved-but-closed.
• Response time improves when the hardest emails get deferred.
• "Active users" climbs when the definition quietly widens.

This isn't cynicism; it's systems literacy. The question isn't "are they lying?" — it's "what behavior does this number reward?" Dashboards measure the behaviors that survive measurement.

The visualization lies (mostly unintentional)#

Chart design steers conclusions before any math happens:

• Truncated y-axes: a chart starting at 90% turns a 2-point wiggle into a cliff. Check where zero (or a sensible floor) sits.
• Dual axes: two lines on independent scales always look correlated. Dual-axis charts have launched a thousand false causality stories; demand the actual scales.
• Cumulative charts: ever-upward cumulative curves mask flat-or-declining period performance — new additions may be shrinking while the total still climbs.
• Cherry-picked windows: a graph starting the week after a bad launch isn't lying about data; it's lying about range.

None of these require dishonest intent — defaults produce most of them. That's precisely why checking beats trusting.

Correlation, causation, and the third variable#

Two metrics rising together invites one true explanation and several impostors: A causes B; B causes A; C causes both; coincidence across a short window. Ice cream sales correlate with drownings — summer is the hidden driver, not dessert. The practical habit: for any compelling correlation, brainstorm one alternative explanation before believing the obvious one. If a plausible confounder exists, the honest conclusion is "interesting, needs investigation," not "proven."

The five-question checklist#

Run any important dashboard claim through:

1. Compared to what — trend, season, target?
2. What's the denominator, and did it move?
3. Who's evaluated by this number, and how would they optimize it?
4. What's the chart not showing — axis starts, window selection, missing segments?
5. Could a third variable explain this pattern as easily?

Ninety seconds of these questions will catch most dashboard-driven misjudgments before they cost anything — and make you the person meetings turn to when the pretty chart says something suspicious.

Related: the five analytics numbers that matter covers which metrics deserve dashboards at all, and spreadsheet mastery covers verifying them yourself.