You pull the monthly report. Revenue per client is down 12% compared to last month. You dig in, build a slide, and walk into the room confident you have found the answer. Cohort analysis for analysts is exactly the discipline that would have caught you before that moment.
Then someone asks: "But are those the same clients we had last month?"
You pause. You realise you do not actually know. And in that moment, the entire analysis starts to unravel.
This scenario plays out in consultancies every week. Not because analysts are careless. Because the default way most of us are taught to compare data — this month versus last month, this quarter versus last quarter — contains a hidden assumption that almost never holds true. The assumption is that both periods contain comparable groups of people. They rarely do.
Why Aggregate Metrics Conceal More Than They Reveal
When you look at a blended average across your entire dataset, you are looking at a summary of multiple different groups compressed into a single number. That number is descriptive. It tells you what happened in total. But it actively conceals the behavioural mechanics underneath.
Here is a concrete example. Imagine your client acquired a large volume of lower-tier customers last month as part of a promotional campaign. This month, the blended Average Revenue Per User is down. On the surface, it looks like existing customers are spending less. But when you separate the data by when each customer was first acquired, you find something different: every historical customer is spending exactly the same as before. The apparent drop is entirely explained by the new, lower-value cohort diluting the average.
Without cohort analysis, you would have presented a problem that does not exist. And recommended a solution to fix it.
This is not an edge case. It is a structural flaw in how aggregate metrics work, and it affects every dataset where the composition of your population changes over time — which is most datasets.
What a Cohort Actually Is
A cohort is simply a group of people who share a common starting point. That is it. No advanced mathematics required.
In practice, a cohort might be clients onboarded in the same calendar month, survey respondents who participated in the same research wave, enterprise projects that kicked off in the same quarter, or users who signed up during the same promotional period. The defining characteristic is a shared entry point.
By grouping people this way, you create a basis for genuine comparison. You are no longer asking "how did this month compare to last month overall?" You are asking "how are clients who started their journey in January performing at the three-month mark, compared to clients who started in February at the same three-month mark?"
That is a like-for-like comparison. And it produces a fundamentally different, and more defensible, answer.
The Mental Shift at the Core of Cohort Thinking
The core of cohort analysis is not a formula. It is a question reframe.
Most analysts default to asking: "What changed between these two calendar periods?" Cohort thinking asks instead: "How is each distinct group performing relative to where it started?"
These feel similar. They are not. The first question compares snapshots in time, mixing together populations at different lifecycle stages, in different external conditions, with different compositions. The second question tracks each group on its own timeline, holding the starting point constant so that comparisons are genuinely meaningful.
Once you internalise this reframe, you start seeing the problem everywhere. Client retention analyses that mix tenured and new relationships. Campaign performance reports that blend audiences across different funnel stages. Quarterly business reviews that compare periods with fundamentally different market conditions.
The like-for-like comparison instinct is one of the things that separates analysts who get challenged in meetings from those who do not.
Three Questions to Ask Before Any Comparison
Before you finalise any comparative analysis, run through this checklist. If the answer to any of these is no, a straight period comparison is likely to mislead you.
1. Are these the same people or groups?
Has the composition of your population changed between the two periods? New clients, churned accounts, a promotional influx, a segment that dropped off — any of these will distort a blended comparison. If the groups are not the same, you are not comparing the same thing.
2. Are they at the same stage in their journey?
A client in month one of an engagement behaves very differently from a client in month twelve. A survey respondent completing a tracker study for the first time is not the same as one in their fourth wave. If your two periods contain people at different lifecycle stages, your comparison is conflating behaviour driven by stage, not by time.
3. Were external conditions comparable?
Seasonality, market events, client-side organisational changes, macroeconomic conditions — any external factor that differs between periods will show up in your numbers as if it were a meaningful trend. Before attributing a change to anything your client did or did not do, ask whether the external environment was sufficiently similar to make a comparison valid.
How to Apply Cohort Analysis in Excel and Power BI
Add two calculated columns: a cohort column (the month each client first entered) and a months-since-start column. Build a PivotTable with cohort on rows, months-since-start on columns, and your metric as values. In Power BI, create a calculated column and build a Matrix visual.
This Is Not an Advanced Technique. It Is a Better Question.
The mathematics involved are well within the range of any analyst comfortable with a PivotTable. What makes cohort thinking feel unfamiliar is the discipline of pausing before a comparison and asking whether it is actually valid.
To learn more about how you can automate your workflow check out our blog.
