Communication cohort analysis

Communication cohort analysis is a method of grouping people by a shared starting point and tracking how their communication behaviour changes over the periods that follow. A cohort might be everyone who joined a community in March, every customer onboarded in the same week, or every new hire in a given quarter. Each cohort is then followed period by period to see what share of them remain active in a defined channel.

The value of the method is that it separates the effect of time from the effect of mix. A flat overall engagement average can hide the fact that older cohorts are decaying while a flood of new members props the number up. By holding each group constant and reading across periods, you see the true shape of how engagement builds or fades after the start point.

The approach is the communication-focused cousin of cohort retention analysis. Where retention analysis asks who is still a customer, communication cohort analysis asks who is still talking, replying, posting, or opening. That distinction matters because communication is often a leading indicator. People go quiet before they leave, so a cohort whose email open rate and reply activity drop in period two is signalling risk well before it shows up in retention rate.

Communication cohort analysis is built rather than reduced to a single number. The output is a grid: each row is a cohort, each column is a period since the start point, and each cell is the engagement rate for that cohort in that period. The steps below define how to construct it and what each input means.

1. 1

   Define the cohort start event

   Choose the shared moment that groups people together, such as join month, first message sent, or onboarding completion date. Every person in a cohort must share this same start event so the periods line up correctly.

2. 2

   Choose the engagement signal

   Decide what counts as active communication for this analysis. It could be sending a message, replying to a thread, opening a digest, or attending a call. The signal must be consistent across all cohorts so the comparison holds.

3. 3

   Set the period length

   Pick the interval for each column, commonly a week or a month. Shorter periods reveal early drop-off in detail, longer periods smooth out noise and suit slower communication cycles.

4. 4

   Calculate the engagement rate per cell

   For each cohort and each period, divide the number of original members who showed the engagement signal in that period by the original cohort size. This produces the percentage that fills the grid.

5. 5

   Read across and down

   Reading across a row shows how a single cohort decays or strengthens over time. Reading down a column compares cohorts at the same age, which reveals whether changes to onboarding or content are improving the curve.

A worked example. The March cohort has 200 members. In period one, 150 of them send at least one message, giving an engagement rate of 75 percent. In period two, 90 are active, so the rate falls to 45 percent. In period three, 70 remain active at 35 percent. The April cohort, after an onboarding change, holds 60 percent in period two. Reading down the period-two column shows the change worked, even though the headline average across all members barely moved.

A cohort grid shows you that period-two engagement is sliding, but it does not tell you why or who should act. A metric tree decomposes cohort engagement into the drivers that shape it, so a dip in one cell becomes a route to a specific cause and a named owner. Cohort engagement sits below onboarding quality, content relevance, channel fit, and early value delivery, and each branch can be measured and owned.

This is where the gap between dashboards and decisions shows up. A dashboard can render a beautiful cohort heatmap and still leave a team staring at it, unsure whether the cause is a weaker onboarding wave, a content change, or a channel that the latest cohort simply does not check. The tree connects the cell to the cause.

Benchmarks for cohort engagement depend heavily on the channel and the audience, so the figures below are directional shapes rather than fixed targets. The pattern that matters is the slope of the curve. A gentle decline that flattens into a stable base is healthy. A steep drop that keeps falling points to a cohort that never found durable value.

Watch for the flattening point. A curve that settles into a stable base, even a modest one, means a loyal core exists and can be grown. A curve that keeps sliding toward zero means there is no floor, and acquisition is filling a leaky bucket. Compare the slope across cohorts rather than judging any single number.

Improving cohort engagement means improving the shape of the curve for new cohorts, not chasing the average up. Because each cohort is a controlled group, you can change one thing for a new wave and read the result cleanly against earlier cohorts. The levers below target the drivers that most often bend the curve.

1. 1

   Defining cohorts by outcome

   Grouping people by how engaged they became, rather than by a shared start event, produces a circular result. The cohort must be anchored to the beginning, never to the end.

2. 2

   Comparing cohorts at different ages

   A young cohort will always look more engaged than an old one. Comparisons only hold when you read down a column, lining cohorts up at the same number of periods since their start.

3. 3

   Changing the engagement signal mid-analysis

   Switching what counts as active between cohorts breaks the comparison. The signal must be defined once and applied identically across every group.

4. 4

   Letting the average hide the curves

   Reporting only a blended engagement average defeats the purpose. The whole point of cohort analysis is to see the shape that the average conceals.