Dataskera

Cohort Grid

One row per cohort, one column per period since that cohort started, colour carrying the value. Retention, decay and ramp-up read as a shape rather than as a table of numbers, and the average row underneath gives every cohort something to be compared against.

Data Requirements

Your Tableau worksheet should include:

  • 1 Dimension for the Cohort encoding. One row per value: a signup month, an acquisition channel, a release.
  • 1 Dimension for the Period encoding. The age of the cohort: 0, 1, 2 periods since it started. Numeric values sort numerically.
  • 1 Measure for the Value encoding. Fills and colours each cell.

Add Cohort Grid to your worksheet

First, download the manifest file. It tells Tableau where to load the extension from.

Download cohortgrid.trex

Direct link: https://dataskera.com/downloads/cohortgrid.trex

Then, in Tableau:

  1. On the Marks card, open the Mark type dropdown, and under Viz Extensions choose Add Extension.
  2. In the Add an Extension dialog, choose Access Local Viz Extensions, then open the cohortgrid.trex file you just downloaded.
  3. When Tableau asks, allow the extension to access the data in the workbook.
  4. The welcome screen appears. Sign in with your Dataskera account, or on Tableau Desktop choose Use on Desktop without an account. See Signing in & access below.
  5. Build the chart: drag the cohort dimension onto Cohort, the period dimension onto Period, and a measure onto Value.
Tableau Server or Cloud:
A site admin must allow the extension by adding its URL to the site's safe list, and the view needs the Download Summary Data permission. Extension URL: https://vizextension.dataskera.com/cohortgrid/cohortgrid.html

Signing in & access

Cohort Grid is free. Whether you need to sign in depends only on what you are doing. Viewing a dashboard never asks for an account, and authoring is free too.

Viewing a dashboard
No account

People who only view a dashboard are never asked to sign in and are never blocked.

Editing on Tableau Desktop
Optional

Sign in, or continue without an account in a single click.

Editing on Server or Cloud
Free account

Sign in with your Dataskera account to author on the web.

No account yet? Create one free in under a minute. No credit card required.

Format panel

Open the Format panel from the extension menu to customize the chart. It is for the dashboard author and is not shown to viewers. It has four tabs:

  • Cells Show: the raw measure, a share of the opening period, or a share of the cohort total; decimals for each.
  • Colour Scale: sequential or diverging, and where the midpoint sits.
  • Grid: cell gap and corner radius, the period header and its prefix, the cohort size next to each label, the size below which numbers are left out.
  • Average Row: an average row under the cohorts, treating every cohort alike or weighting each by the size it started at.
  • Color: a palette, a precise colour per category, and the overall opacity.
  • Legend: show or hide it, its position around the chart, an optional show/hide toggle for viewers, the colour swatch, and the content of each entry.
  • Tooltip: show or hide it, its colours, decimals, and a template you write yourself from the placeholders the chart provides.
A young cohort never drags the curve down:
Each period of the average row averages only the cohorts old enough to have reached that period, so the tail of the curve is not pulled down by cohorts that simply have not lived that long.
The opening period is a baseline, not a value:
In share-of-first mode every cohort opens at a hundred percent by construction. That column carries no information, so it is kept out of the colour scale and drawn neutral, where painting it at the top colour would make it indistinguishable from a genuinely extreme cell.
A late start is declared:
A cohort with no value in the opening period is measured from the first period it does have. That is the only defensible reading, and it is not the same baseline as its neighbours, so the chart says so.
Numbers or nothing:
Below the minimum cell size the numbers are left out rather than overprinted, and the chart says why. Hovering a cell still reads it.