Don't Panic: Cohort and Funnel Analysis

Cohort and funnel analysis are two ways of making an event stream admit what it knows. A cohort puts people or accounts who share a starting condition into the same row, then watches them across equal elapsed periods. A funnel puts required actions into sequence and counts who reaches each one. The events have not become wiser. They have merely been asked a question with fewer escape routes.

The first important idea is the analysis unit, which is the thing being counted. A person, an account, and a session are not interchangeable. A person can make several visits. Several people can act for one account. An anonymous visit and a later authenticated visit can be one journey or two, depending on the identity rule. The denominator is therefore not a small technical detail hiding under the chart. It is the floor under the chart.

The second idea is that every picture is assembled from rules. A cohort needs an inclusion condition and a return condition. A funnel needs step conditions, an order rule, an entry rule, and a conversion window. Sequential order permits unrelated events between the required steps. Strict order does not. An open funnel can admit someone at a later step. A closed funnel asks everyone to begin at step one. The numbers change because the population changes, which is inconvenient only if the metric was expected to be a decorative houseplant.

The surprise is that recent cohort cells are not bad news by default. They are often unfinished observations. A weekly cohort has not had time to populate later weeks, so compare mature elapsed periods rather than treating the newest diagonal as zero. Funnels have their own version of this problem: a short window can reject a valid slow journey, while a long one can join separate attempts into a fictional success.

When a result moves, begin with counts as well as rates. Lower entry volume, a weaker transition, and weaker later retention are different shapes with different next checks. The Cheatsheet holds the formulas and diagnostic signals. The Reference tab explains how analytics systems implement the rules. The Practice Reference turns the rules into a repeatable analysis, and the Exercise gives you a small event stream to interrogate without alarming a real dashboard.

A cohort or funnel can show where behavior differs and which group carries the difference. It cannot prove why. Use paths and session evidence to investigate what happened. Use interviews to understand intent. Use a controlled experiment when the decision needs a causal answer. The chart has done useful work when it narrows the next question. Asking it to settle the whole case is how a helpful table gets promoted to oracle.

Don't Panic: Cohort and Funnel Analysis

Relevant careers