Fannel, cohorts, retention

The funnel shows where the script is lost, the cohort shows how comparable groups behave over time, retention shows whether users are returning to value. These tools complement each other, but the output depends on the exact events, the observation window and the meaningful segment.

Why do you need fannels, cohorts and retention?

How fannels (funnels) work

A fannel is a step-by-step process, a step-by-step process, like registering, filling out a profile, buying, and at every step, a part of the user falls off, and if you know where the metric is going, you know where to look for the problem.

**Example: there are 10,000 downloads on a mobile app, only 6,000 registered, 3,000 first order, and the feed shows the main gap between download and sign-up, so we need to improve onboarding.

Why do we need cohorts?

A cohort is a group of users who started using a product at about the same time or with a single action, usually looking at cohorts by registration date or on the first order.

Cohorts help to avoid confusing the effect of novelties with habit or shoals. If the retention of a new cohort falls below the previous ones, something may have broken or the test failed.

**Example: Developers are introducing new onboarding. Cohort analytics show that those who have undergone a new onboarding are more likely to reach payment than the previous cohort.

Why to monitor retention

Retention is the percentage of users who return to a product after a certain time. Weak retention = high churn, almost always a problem in the product or its value. Often count Day 1, Day 7, Day 30 retention.

** Case: Day 1 retention is 38%, Day 7 12%, and if push notifications are improved, the experiment is set.

How to build fannels and find bottlenecks

Stage-by-stage

List the key actions without which conversion will not happen. For e-commerce: visit → add to the cart → start registration → payment.

How to fix the numbers at each stage

In the reports, you see conversions between stages: how many users have gone through a step, how many have fallen off, if you think of it as sessions at one stage and you think of it as users at another, you’ll get errors.

Example: In a fintech service, the gap between filling out the questionnaire and linking the card increased by 9 percentage points after optimizing the UI.

Experiment: How to Measure the Impact of Change

When testing a hypothesis, for example, simplify the order form, compare the fannel before and after, see the difference in cohorts, count uplift not only at the conversion stage, but also on the retension.

** Case: In the FMCG A/B test, the new checkout test, Fannel showed that adding autocomplete increased conversions to payment, and a cohort of new users with a new checkout makes a second purchase more often in a week.

Amplitude: Funnel Analysis Overview

How to Collect Cohorts and Read User Behavior

Basic approaches to cohort construction

In product analytics, cohorts are most often built on:

• date of registration or first action
• channel of attraction (organics, advertising, affiliates)
• first product or function

If you look at all the data mixed up, the conclusions can be false: for example, users who left after the campaign spoil the general metrics.

How to Read Cohorts in Dynamics

Open a cohort report in BI or product analytics, compare the returns of different cohorts, and if a new cohort falls off faster, look for changes in product or marketing.

Example: An online education platform noticed a Day 7 retention drawdown in the cohort after launching a promo code: users activated trial periods and burned out without switching to a paid version.

Use heatmap, cohort table or line chart to visualize the image.Tools: Amplitude (Cohort Analysis Guide), Mixpanel.

Retention: How to calculate and what to use

Types of retention

There are several ways to count retention:

• Classic retention (return on a specific day)
• Rolling retention (returned at or later than the specified time)
• Bracketed (synonym rolling)

It is important not to be confused with the DAU/MAU: the latter shows the proportion of active, but does not take into account cohort.

Major traps and mistakes

A common mistake is to count cohorts by the wrong event or mix metrics of new and old users, and another anti-pattern is to try to improve the retention with mechanics that don’t affect the core value of the product.

** Case: Too many unnecessary fluffs temporarily level Day 1 retention, but by the end of the month lead to a cleanup of the app.

How to apply for product solutions

Retention is the main reference point for features, marketing campaigns or new UX. If the result of the experiment does not give a long-term effect on the cohort retension, change direction.

What to draw conclusions from analytics

How to track experimental effects

For any product test, measure metrics, not just conversions, but also cohort retention dynamics, otherwise you can imperceptibly worsen long-term performance.

** Example: Fintech added gamification. Conversion went up, but after a month the cohort of new ones went away: the mechanics weren’t about core value.

What decisions are most often made

• Change the form of onboarding if cohorts of new users lose retention.
• Remove the phases of the fannel, where most of the users are lost.
• Reverse feature or discount if new cohorts become worse on returns.

**Tip: Always look at analytics not just by general metrics, but by cohorts and retensions, which reduces the likelihood of making mistakes.

Anti-patterns and typical errors

False groupings and the mix of cohort/funnel

Don’t mix different channels of attraction or regions. Compare similar cohorts over distances, or you’ll get a porridge.

Focus only on the short retension

Day 1’s short-term growth doesn’t save if Day 7 and Day 30 are near zero.Don’t chase numbers without evaluating repeat cycles.

Chaotic gathering of events

If there is no unity in gathering events and interpreting steps, all results are not representative.

FAQ: A brief summary of the main

**How is the fannel different from the cohorts? A channel is the path of one user in a moment. A cohort is a group of users by time or event and their behavior in the dynamics.

**What if Day 1 retention is high and then drops sharply? Often, the cause is a strong trigger in the first stage, but there is no value to return.

**Can the fannel and cohorts be built on different events? Yes, the fannel is often built on key steps, cohorts on the entrance or the first action.

**How do you know which cohort is more important? Look at the cohorts by target action and by the channels that matter to the business, and you can see who pays off and how fast.

**What tools are needed for cohort analysis? Amplitude, Mixpanel, Looker, own BI-systems with cross-tables and heatmap.

**What if all the cohorts show a drawdown? Check for bugs, product changes, marketing, or external factors, traffic may have deteriorated or technical failures have occurred.