Market dashboards that lie less
Guide · Continuous Intelligence · 4 min read · last verified 2026-07-21
A dashboard that shows a number without its denominator, its method, and its observation date is reporting an impression rather than a measurement. Most market dashboards are built to be scanned quickly, and every design choice that makes a number faster to read also makes it easier to misread. The corrections below cost display space and buy back accuracy.
Show the denominator
"Mentioned in 12 responses" is uninterpretable. Twelve out of fifteen and twelve out of four hundred describe opposite situations, and the rendered number is identical.
Every count needs its base visible in the same view, not in a tooltip and not on a methodology page. Percentages need it more, not less: a percentage without a denominator hides how few observations produced it, and small denominators produce dramatic percentages that swing on a single item.
Where the denominator changes between periods, the change itself must be shown. A rise from 20% to 30% means something different when the base went from 10 to 200, and a dashboard that plots only the percentage is actively concealing the more important fact.
Put the method next to the number
The reader needs to know what produced the figure: which questions were asked, which surfaces were consulted, and over what span. That span is the measurement window, and its length and position change results independently of anything happening in the market.
A compact method label — question set version, source set, window length — sitting beside the number is enough. Its real function is to make method changes visible. When the label changes, the reader knows the series has a seam in it, which is the single most useful thing a dashboard can communicate about its own history.
Distinguish zero from not measured
Three states get collapsed into an empty cell, and they carry different meanings:
- Measured, found nothing — a real observation with real information
- Not measured this period — coverage gap, no information
- Measurement failed — source unreachable, query errored, capture incomplete
Rendering all three as 0 turns coverage gaps into apparent declines. A team then investigates a drop that never happened, or worse, celebrates a competitor's apparent disappearance that is actually a broken scraper.
Each state needs its own visual treatment, and the third needs to be loud. A silent failure that reads as a finding is the most damaging thing a dashboard can do.
Show what the number rests on
Sample size, coverage, and known gaps belong on the surface. Where a figure comes from a partial view — three of eight competitors checked, two of five regions covered — the partiality is part of the number.
Confidence displays should be legible rather than statistical theater. A count of underlying observations and an explicit list of what was not covered communicate more to a working reader than an error bar whose derivation nobody can reconstruct.
Date the observation, not the render
Dashboards commonly display the time they were loaded, which tells the reader nothing about the freshness of the data behind it. A panel showing "updated 2 minutes ago" over figures last collected six weeks earlier is a straightforward misrepresentation, and it is nearly universal.
Show the observation date per panel, since panels rarely refresh on the same schedule. Where data is older than its expected refresh interval, say so on the panel itself rather than leaving the reader to compute it.
Make the evidence one click away
Every figure should link to the material that produced it — the captured text, the response, the page, with its date. This is the evidence trail, and its presence changes behavior in a way no amount of annotation does.
When evidence is reachable, disputes resolve by looking. When it is not, they resolve by seniority. A dashboard whose numbers cannot be opened will eventually be trusted or distrusted as a whole, which is the wrong granularity for either.
Refuse to connect incomparable points
A line implies that adjacent points were produced the same way. When the method changed mid-series, drawing through the change asserts something untrue about the data.
Break the line at the seam, or render the segments distinctly, and label what changed. The chart becomes less elegant and stops making a claim it cannot support — the same constraint behind why trend lines need fixed methodology.
Anti-patterns worth removing
- A single large number with no context — impressive, uninterpretable, and the most-screenshotted element on any dashboard
- Rolling averages by default — they smooth away the step changes that are the most actionable events in a series
- Auto-scaling axes — noise rendered as a mountain range; fix the axis or show the movement floor
- Color thresholds with undocumented cutoffs — red and green are conclusions, and their boundaries need to be stated
- Period comparisons across method changes — the comparison that most often gets quoted, and the one most likely to be measuring the instrument
- Metrics with no owner — a panel nobody maintains keeps rendering long after it stopped meaning anything
An audit to run this week
Take each panel and answer four questions in writing: what is the denominator, what method produced this, when was it observed, and what does an empty value mean here. Any panel that cannot be answered on all four is either fixed or removed. Panels that survive the audit and then go unread for a quarter should also be removed — an unread panel still costs attention every time it is scrolled past, and its presence implies a coverage that nobody is actually maintaining.
What to watch
Watch for the first time someone screenshots a figure into a document without its context. That number now travels alone, and whatever the dashboard was doing to keep it honest no longer applies to it.