Retention by channel and campaign

What is retention by channel?

Retention by channel is the same cohort retention you already track, split by the install source credited to each user, so every channel, campaign or creative gets its own row of Day 1, Day 7 and later values.

Install source: the channel and campaign an attribution tool credits for an install, or organic when no campaign is credited.

Blended retention: retention for all installs together, with every channel pooled into one number.

Why blended retention hides what is happening

A blended number averages channels that behave very differently. If organic users retain better than paid ones, the blended figure depends as much on how many installs came from each as on how well any channel retains.

That leads to two common misreadings:

• A blended drop is blamed on the product when it was caused by more paid spend.
• A blended rise is credited to a product change when organic installs simply grew.

The channel mix trap

In the sample app on the home page, a made-up app, a paid campaign scales up partway through the period. Classic Day 1 retention looks like this:

Blended Day 1 falls by about 7 points, yet neither channel changed. The mix moved, not retention. This is a version of Simpson's paradox, and the mix section of the reading guide shows it in the full table. You can reproduce it in the cohort explorer by switching the channel filter.

How to cut the cohort table

Work from broad to narrow, and only go one level deeper when the level above shows something worth explaining.

1. Channel first. Organic against each paid network. This tells you whether the mix explains a blended move.
2. Campaign next. Within a channel, one campaign often drags the average down.
3. Creative last. Creatives with the same targeting can bring different users, especially if one promises something the app does not do.

At each level, keep the same retention definition, the same cohort dates and the same day boundary. The channels section of the reading guide lays out the table, and there are more ways to slice cohorts in this piece on six ways to cut cohorts.

How small is too small?

Small cohorts swing a lot, because one or two users returning can move the rate by several points. Read small cohorts as directional, not as results.

• Pool several days or a week of cohorts before comparing campaigns.
• Look for a pattern that repeats across cohorts rather than a single high or low cell.
• As a rule of thumb, if adding or removing one returning user would change your decision, the cohort is too small to act on.

The sample size notes explain why this matters more at creative level, where cohorts are smallest.

Keep campaign names consistent

Retention by campaign only works if campaigns roll up cleanly. If the same campaign appears as three slightly different names, its users are split across three small rows that all look unreliable.

• Agree one naming pattern for channel, campaign and creative, and stick to it.
• Use the same case and separators everywhere.
• Fix names at the source rather than cleaning them in a spreadsheet each week.

This guide to inconsistent UTM naming covers the usual problems and fixes.

Read paid and organic on the same definition

Before comparing channels, check that both use the same retention definition. Rolling is always at least as high as classic and keeps rising as late returns arrive, so a rolling figure for one channel against a classic figure for another is not a comparison. Classic is the safer basis. The guide to why paid D1 is lower than organic goes deeper on that gap.

Where do you get retention by channel?

Attribution tools (MMPs) and product analytics tools can split retention by install source; examples to check include Linkrunner, AppsFlyer, Adjust, Singular and Branch. Linkrunner reports classic, rolling and cumulative retention as separate columns. Whichever tool you use, check its documentation for which definition its retention column uses before you compare numbers across tools or against a benchmark.