Bucketing & Grouping
Continuous things do not group themselves. Hand a column a time unit, an equal-width step or a list of breakpoints, and the query hands back labelled bands with a sort key that holds their order.
Assigning a row to a band
4 of 8 rows read- 32<50
- 840250 – 1K
- 11<50
- 2,4001K+
- 18650 – 250
- null
- 95
- 1,120
CASEWHEN employee_count IS NULL THEN '(none)'WHEN col < 50 THEN '<50'WHEN col < 250 THEN '50 – 250'WHEN col < 1000 THEN '250 – 1K'ELSE '1K+'END
- <50sort 02
- 50 – 250sort 10
- 250 – 1Ksort 21
- 1K+sort 31
- (none)sort null0
Product figures from the platform’s own defaults - not customer averages
How it works.
Three ways to band
Truncate a date column to a unit. Cut a numeric column into equal-width bins from a start you name. Or list breakpoints and let the query build the ranges between them, which is how uneven bands like a small, medium and enterprise split get made.
Every band emits two expressions
A text label for the axis and a separate numeric or chronological key for the ordering. Without the second one a band called ten thousand plus sorts before a band called two hundred, because the axis would be reading letters.
Three sources, resolved in order
The banding written on this widget wins. Failing that, the banding an operator set as the field's default applies everywhere that field is grouped on. Failing both, a date column is banded automatically from how long the window is, moving through hours, days, weeks, months, quarters and years as the window lengthens.
Mismatches are dropped rather than run
A time unit aimed at a numeric column, or a numeric band aimed at text, would produce SQL that fails at execution. The compiler checks the declared column type first and falls through to plain grouping, so the chart still draws while the operator fixes it.
Two moments in every number.
Every number passes through the same seven. Bucketing & Grouping is the lit ones, and everything either side of it is a different page in this category.
- 01Query
what is asked, and of what
- 02Shape
the measure and the dimension
- 03Draw
the visual chosen for that shape
- 04Arrange
where it sits on a dashboard
- 05Filter
narrowing it after the fact
- 06Drill
going from the number to the rows
- 07Share
who else gets to see it
The specifics.
8 facts- Time units
- Day, week, fortnight, month, quarter, year. A fortnight pairs ISO weeks and labels itself with both
- Numeric
- Equal-width bins from a start, or a list of breakpoints that must be distinct
- Cap
- Fifty bins or fifty breakpoints, above which the request is refused
- Overflow
- On a numeric banding, anything under the first boundary and anything at or over the last collapse into their own labelled bands, and empty values get a band named none. A time banding leaves an empty value unlabelled and sorts it last
- Empty bands
- A band matching no records still draws, at zero, so a hole in the middle of a range reads as a hole rather than closing up
- Row budget
- The widget's cap, padded by the band count so an authored band is never cut off, ceiling of a thousand
- Not offered
- On text columns, which group on their raw value
- Labels
- Abbreviated to thousands, millions, billions and trillions, with the currency symbol added by the renderer when the field is money
What this is assembled from.
A number on a dashboard was built somewhere before it was drawn. These are the parts it is made of.
More in Reporting & Analytics
11 capabilitiesSeven data shapes, twenty-two visuals, shareable without the portal.
Hand a board to a team or a CRM role, at read or at edit, by the person who built it. Roles outlive people, which is why a share points at one, not a mailbox.
Columns, funnels, treemaps, radar, scatter, gauges and pivots - offered only where the data shape supports them, so no chart can misread its own data.
Build as many as the floor needs, arrange them on a grid, and give each team the one that answers their question rather than everybody's question.
Click a bar and the rest of the dashboard follows it, so the next question is asked by pointing at the answer instead of rebuilding the filter.
Wire a chart's clicks to a second dashboard. The value travels on the URL and is read by the board that was named before it paints for the first time.
Colour a number by the rule that matters - under target, over threshold, inside a band - so a dashboard reads correctly from across the room.
Draw the target on the chart itself, so performance is read against the number that was promised rather than against the shape of the bars.
Six palettes including a print set, pinned per widget, each carrying eight series colours plus the axis, grid, card and tooltip colours drawn around them.
Thirty-two bytes of entropy, an expiry you choose, revocation on the spot, and only the hash kept at rest, on a board that draws its figures for a reader.
Name a slot at the call site and one authored widget answers it everywhere inside the product, computed against the records of whoever is signed in.
Total calls, answered rate, average duration and missed calls, split by direction and outcome, over the call records rather than a sampled export.
The rest of the platform.
Five more categories, all on the same record and the same bill. Each card names three of its capabilities, so you can tell from here whether it is worth opening.
Reporting & Analytics
See bucketing & grouping on your own floor.
Thirty minutes, your numbers and your data. We will set bucketing & grouping up live and you can decide from the thing itself rather than from this page.
14-day trial · no card · migration included