Retention Analytics: Read Cohorts and Investigate Drop-Offs

What Retention Means

Retention is the percent of users who come back. Sign up 100 users in January, 40 are still active in February, your one-month retention is 40%.

Retention affects lifetime value and recurring revenue. A cohort curve shows how return rates change as users spend more time with the product.

Choose useful checkpoints

Day 1, Day 7, and Day 30 can help you examine early use, repeat use, and longer-term return rates. They do not establish whether users understood the product or built a habit. Choose intervals that suit the product's expected frequency of use.

Define what counts as returning. Exact-day retention measures activity on a specified day; “on or after” retention includes later returns. Those methods can produce different curves from the same events¹. Keep the definition consistent across comparisons.

Logo Retention vs Revenue Retention

Logo retention measures the percentage of customers retained over a period.

Gross Revenue Retention (GRR) measures the percentage of starting recurring revenue retained after churn and contraction, excluding expansion. It caps at 100%.

Net Revenue Retention (NRR) includes expansion from those same customers and can exceed 100%. Above 100%, expansion more than offsets churn and contraction during the measured period. That does not guarantee future growth.

Review GRR alongside NRR: expansion from some accounts can offset losses elsewhere without resolving those losses.

Reading a Retention Curve

Read the curve for your product rather than expecting a fixed three-stage shape.

• An early drop is a reason to review the first experience and acquisition audience. The curve alone does not identify the cause.
• A later drop may warrant research into repeat use and the reasons people return.
• A flatter curve shows more stable retention over the observed intervals. It does not prove that users will remain indefinitely.

Check the underlying events and talk with users before choosing a change.

Try It

How to Read the Output

Sample input: 1,000 January signups. Day 1 = 600 (60%), Day 7 = 400 (40%), Day 30 = 250 (25%), Day 90 = 200 (20%).

The count falls from 600 on Day 1 to 400 on Day 7, a decline of 200 users or 20 percentage points of the original cohort. The Day 7 count is one-third lower than the Day 1 count. For exact-day retention, however, those may not be the same users, so this is not proof that one-third of Day 1 returners left permanently.

The largest absolute decline from signup is before Day 1: 1,000 to 600. Both intervals deserve attention. Neither establishes whether onboarding worked or whether users need habit triggers.

Compare benchmarks carefully

Match the platform, product category, retention definition, and observation period before comparing rates. AppsFlyer's published app-retention table separates Android and iOS and shows substantial differences across categories². A mobile game and a B2B subscription service are not interchangeable reference groups.

Use your own earlier cohorts as another comparison. Record the source and period for any external benchmark instead of treating an undated table as a 2026 industry standard.

An example of testing a retention change

Duolingo describes an experiment that separated its daily goal from its streak requirement. The company reported improved Day 14 retention for that change³. This supports testing a specific mechanic in context; it does not show that adding streaks will improve every product's retention.

Five Common Mistakes

1. Reporting blended retention

A single average can hide differences between older and newer users. Group users by signup week or month and compare them at the same age.

2. Optimizing for Day 7 when Day 1 is broken

Review early drop-offs before assuming the problem begins later. An early decline can make later cohort counts small, but it does not rule out a separate issue affecting established users.

3. Mixing acquisition channels

Compare retention by acquisition channel. Differences can help you identify audiences or onboarding paths to investigate; they do not, by themselves, prove a channel attracts the wrong users.

4. Tracking logo retention but ignoring revenue

For example, losing one $10,000/year account while retaining ten $10/month accounts leaves 10 of 11 customers, or about 90.9% logo retention. Annual recurring revenue falls from $11,200 to $1,200, a decline of about 89.3%. The customer count alone would understate the loss.

5. Building Day 30 features for Day 1 problems

If research identifies broken onboarding, address that problem before testing re-engagement campaigns. Match the intervention to the cause you have investigated.

Investigate the drop point

These are hypotheses to investigate, not diagnoses that follow automatically from a percentage drop.

Cohort Analysis in 5 Steps

1. Group users by signup week or month. That is your cohort.
2. Track them at fixed intervals. Day 1, 7, 14, 30, 60, 90.
3. Build a cohort table. Rows = cohorts. Columns = day intervals. Cells = retention percent.
4. Read horizontally. Follow one cohort over time. Where does it drop?
5. Read vertically. Compare cohorts at the same interval. Are you improving?

Investigate meaningful differences while accounting for cohort size, acquisition mix, and whether enough time has elapsed to observe the interval.

Behavioral Cohorts

Group users by what they did, not just when they joined.

Pick one core action, such as uploading a photo, inviting a friend, or completing a purchase. Compare subsequent retention for users who did and did not take that action, using a consistent observation window.

A gap identifies an association worth investigating. Users who already value the product may be more likely both to take the action and to return. Test the proposed onboarding change before attributing retention to the action.

AI Prompts for Retention Work

These work in Claude, ChatGPT, or Gemini with your cohort CSV pasted in. Always include a grounding instruction so the model cites the rows it used.

Cohort Pattern Analysis

Analyze this cohort data:
[paste CSV with cohort, day, retention %]

Find:
- Best and worst cohorts at Day 7, 30, 90
- Anomalies (any cohort 20%+ off from average)
- Three hypotheses for the differences

Cite the specific rows you used. If you cannot find evidence, say so.

Drop-off Diagnosis

Daily retention for new users D0 to D30:
[paste numbers]

Find the largest absolute and relative declines. Use only benchmarks I supply, and check that their retention definition, category, platform, and period match. Suggest hypotheses to investigate and tests for those hypotheses. Do not infer causes from the curve alone.

Channel Segmentation

Retention by acquisition channel:
[paste signup date, activity dates, and acquisition channel for each user]

Calculate exact-day D7 and D30 retention per channel using only cohorts old enough to observe each interval. Identify the gap between best and worst, report cohort sizes, and suggest further analysis before making budget shifts.

A 30-Day Retention Sprint

Week 1. Calculate Day 1, Day 7, Day 30 for the last three months of cohorts. That is your baseline.

Week 2. Segment by acquisition channel. Find your best and worst channels. Note the gap.

Week 3. Pick the steepest drop. Design one experiment targeting it. Write the hypothesis, the change, and the success metric.

Week 4. Agree on the measurement plan, then ship the experiment behind a feature flag. Wait for cohort data to mature.

The month establishes a baseline and starts an experiment. Longer-term retention results need more observation time.

What This Connects To

Use retention alongside these related metrics:

• LTV Calculator for how retention turns into lifetime value
• CAC Calculator for the cost of acquiring new customers
• DAU/MAU Ratio for activity frequency among monthly active users
• MRR/ARR for how retention rolls up to revenue

Sources