A heatmap is a visualization that uses color intensity to represent the magnitude of values across a two-dimensional grid, so patterns and outliers stand out at a glance. Each cell sits at the intersection of two dimensions — say, day of week and hour of day — and its color encodes the measure: darker (or hotter) means more, lighter means less.
When to use a heatmap
Heatmaps shine when you’re scanning for patterns across two dimensions at once, rather than reading precise values:
- Time-by-time grids. Orders by day of week and hour of day reveal the Tuesday-morning spike and the weekend lull in one glance.
- Cohort retention. Sign-up month down the rows, months-since-sign-up across the columns, retention rate as color — the classic retention triangle is a heatmap, and a fading color band shows churn faster than any table of percentages.
- Category-by-category comparisons. Support tickets by product area and severity, or feature usage by plan tier.
- Geographic density. A map shaded by how much activity falls in each area is a heatmap where the grid is the world.
When not to use one
Color is a blunt instrument. Humans compare lengths far more precisely than shades, so if readers need exact values or fine distinctions between similar numbers, a bar chart or a plain summary table serves them better. A heatmap says “look here,” not “this is 4.7% higher than that.”
Heatmaps also need enough data density to be worth the format. A grid that’s mostly empty cells, or a 3×2 matrix with six values, has no pattern to reveal — the color scale just adds decoding overhead to numbers a table would show plainly. And choose the color scale honestly: a diverging scale (two hues meeting at a midpoint) implies a meaningful center such as zero or a target, while a sequential scale implies a simple low-to-high range. Using the wrong one suggests structure the data doesn’t have.
Heatmaps and Metabase
Metabase doesn’t have a general-purpose heatmap chart type, but it covers the two most common heatmap jobs with other tools:
- Matrix-style heatmaps are built with conditional formatting on tables and pivot tables. Build a pivot table with one dimension on rows and another on columns, then add a color-range rule to the cells — the result reads exactly like a heatmap, with the actual numbers still visible in each cell. This is the standard way to build a cohort retention grid in Metabase.
- Geographic heatmaps are handled by maps: a grid map shades latitude/longitude buckets by density, and a region map shades countries or states by a metric — a choropleth, which is a heatmap over regions.
If your data is one-dimensional — a single numeric column whose distribution you want to see — a histogram is usually the better instrument than forcing it into a grid.
Related terms
Further reading
Put it to work
- Churn rate — Metric
- Marketing analytics — Overview
