Cohort and Funnel Analysis
Cohort analysis compares groups that share a starting condition across equal periods of time. Funnel analysis counts how many people or accounts progress through a defined sequence, so you can locate conversion loss and compare behavior between groups.
itData engineering and analytics | OpenSkills.info
Course pathWalk it in order
Look it upDip in anytime
Go furtherLeaves this page
Don't Panic
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.
Where this skill leads
Relevant careers
See how this topic contributes to broader role-level skill maps.
Sources
- https://support.google.com/analytics/answer/9670133?hl=en
Supports
- Cohorts as groups sharing an Analytics dimension
- Inclusion and return criteria, daily, weekly, and monthly granularity
- Elapsed-period cohort tables and incomplete recent observations
- Cohort claims in intro, slides, cheatsheet, quiz, links, video, and infographic
- https://support.google.com/analytics/answer/9327974?hl=en
Supports
- Open and closed funnels
- Ordered steps, direct and indirect following, conversion windows, breakdowns, and elapsed time
- Google Analytics reference rationale and landscape placement
- https://amplitude.com/docs/analytics/charts/funnel-analysis/funnel-analysis-how-amplitude-computes
Supports
- Any-order, sequential, and exact-order funnel semantics
- Property segmentation at funnel entry
- Amplitude reference rationale and landscape placement
- https://amplitude.com/docs/analytics/create-cohorts
Supports
- Reusable cohorts created from chart points, imported identifiers, and segmentation
- Behavioral and static cohort distinctions
- Cohort use with funnel and retention analysis
- https://docs.mixpanel.com/docs/reports/funnels
Supports
- Funnels as conversion measurement through a series of events
- Mixpanel landscape placement
- https://docs.mixpanel.com/docs/reports/retention
Supports
- Start and return behavior, retention rate, cohort buckets, and incomplete buckets
- Exact-period, unbounded, and consecutive retention distinctions
- Mixpanel reference rationale and landscape placement
- https://posthog.com/docs/product-analytics/funnels
Supports
- Sequential, strict, and any-order funnels
- Overall and previous-step conversion, cohort breakdowns, property attribution, and incomplete windows
- PostHog Awesome Link rationale and landscape placement
- https://posthog.com/docs/product-analytics/retention
Supports
- Start and return events, unique-user or group cohorts, elapsed-period retention tables
- First-time, first-ever, and recurring retention choices
- https://help.heap.io/hc/en-us/articles/37271972717073-Funnel-analysis-overview
Supports
- Person, account, and session funnel denominators
- Sequential actions, conversion windows, unique-user counts, and conversion arithmetic
- Heap reference rationale and landscape placement
- https://help.heap.io/hc/en-us/articles/37271980341009-Retention-analysis-overview
Supports
- Start and return event retention analysis
- Cohort analysis as engagement measured across time
- https://support.countly.com/hc/en-us/articles/4437429216409-Funnels
Supports
- Event-based ordered funnels, same-session and cross-session progression, and retroactive reporting
- Countly Awesome Link rationale and landscape placement
- https://support.countly.com/hc/en-us/articles/4414450842009-Retention
Supports
- Exact-period, unbounded, and consecutive retention calculations
- Cohort size denominators and elapsed-period rates
- https://support.countly.com/hc/en-us/articles/4405086657049-Cohorts
Supports
- Behavioral cohorts based on events and time windows
- Cohorts used to segment funnels and retention
- https://experienceleague.adobe.com/en/docs/analytics/analyze/analysis-workspace/visualizations/cohort-table/cohort-analysis
Supports
- Retention, churn, rolling, latency, and custom-dimension cohort tables
- Adobe Analytics reference rationale and landscape placement
- https://matomo.org/guide/reports/funnels/
Supports
- Funnels as expected action sequences and analysis of loss between steps
- Historical analysis, step trends, segments, and Matomo landscape placement
- https://github.com/sindresorhus/awesome
Supports
- Discovery route to the curated Analytics list
- https://github.com/oxnr/awesome-analytics
Supports
- Discovery of PostHog and Countly
- Relevance of selected ecosystem projects to event, funnel, and cohort analysis
- https://medium.com/airbnb-engineering/experiments-at-airbnb-e2db3abf39e7
Supports
- Visitor and user identity choices in multi-device journeys
- The need to wait for journeys with delayed completion before interpreting results
- Field Notes guidance on denominator and identity assumptions
