Cohort Analysis for Marketers: GA4 Setup, LTV Calculation, and Retention Patterns

Cohort Analysis for Marketers: GA4 Setup, LTV Calculation, and Retention Patterns

Cohort analysis groups users by when they first acquired or converted, then tracks how a specific behavior (retention, revenue, engagement) changes for each group over time. It answers the question attribution reports can't: are customers from paid social in January retaining better than customers from paid search in February? And if retention dropped in March, is it a product problem or an acquisition quality problem?

Key takeaways

1. An acquisition cohort groups users by the month or week they first converted, it separates the question "how is this campaign performing now" from "how have customers from that campaign performed over 90 days."
2. D7 retention below 25% for SaaS or app products typically indicates an onboarding problem, not a product problem, the solution is first-week activation, not feature development.
3. LTV by acquisition cohort (ARPU × gross margin ÷ churn rate) is the only metric that tells you whether paying more per customer via one channel is actually justified by what those customers are worth.
4. 70% of organizations don't run control-group experiments to evaluate campaigns, cohort analysis without control periods is subject to the same confound: correlation without causation.
5. GA4's Cohort Exploration report builds acquisition cohorts by event in under 5 minutes, the most common setup error is leaving the cohort inclusion event as "First visit" instead of a meaningful conversion event.

The ICP problem cohort analysis solves: most marketing dashboards show performance in the current period only, this week's conversions, this month's CAC, yesterday's ROAS. These snapshots tell you what happened, not whether the customers you're acquiring are worth what you paid for them. A campaign that lowered CAC by 30% but acquired customers with 50% higher churn is a loss, but that verdict doesn't appear until 90 days later, well after the budget decision has compounded.

Acquisition cohort: A group of users defined by when they first completed a specific action (signed up, made a first purchase, started a trial). All subsequent behavior is tracked relative to that entry point.

Behavioral cohort: A group defined by a shared action at any point (users who watched a video, users who used a specific feature), not limited to acquisition events.

Retention rate: The percentage of a cohort that returns and completes a target action in a given period after acquisition. D7 = returns within 7 days, D30 = within 30 days.

Churn rate: The percentage of a cohort that does NOT return or cancels, the complement of retention rate.

01. What Cohort Analysis Reveals That Standard Reports Miss

Standard channel reports aggregate all users, new and returning, in the reporting period. A traffic spike from a viral campaign last month inflates this month's retention metrics because returning viral users are blended with new users. Cohort analysis separates these groups: the January cohort contains only users who first converted in January, regardless of when you're running the report.

This separation surfaces three insights hidden in aggregate data:

1. Acquisition quality by channel. If Google Search cohorts retain at 45% after 30 days while TikTok cohorts retain at 22%, you're paying a different effective price for the same conversion, the true cost per retained customer is roughly 2× higher from TikTok, even if the reported CPL is identical.

2. Product or acquisition problem? If every cohort's D30 retention drops starting from the March cohort onward, that's a product or onboarding change, the problem started at a specific point in time. If only the paid social cohorts from March retain poorly while organic cohorts hold steady, it's an acquisition quality problem, the channel changed who it was delivering.

3. LTV by channel before it's too late to act. The LTV/CAC ratio framework requires LTV data, but most teams calculate LTV from averages. Cohort analysis gives you LTV per acquisition channel, so you know whether the $200 CAC for a LinkedIn lead is actually returning 4× or 1.5×.

02. GA4 Cohort Exploration: Setup in 5 Steps

GA4's Cohort Exploration report is the fastest way to build acquisition cohorts for web and app properties.

Step 1: Open Explore. Cohort Exploration

In GA4, go to Explore in the left nav. Select Cohort Exploration from the template gallery. This opens a blank cohort report.

Step 2: Set Cohort Inclusion Event

Under Cohort criteria, click Cohort inclusion. This defines what triggers membership in a cohort. Change the default "First visit" to your actual conversion event, for SaaS: sign_up or start_trial; for e-commerce: purchase or first_purchase. Using "First visit" produces a cohort of all site visitors, which is too broad to be actionable.

Step 3: Set the Return Criteria

Under Return criteria, select the event you want to track post-acquisition. Options:

• Any event, measures general site/app re-engagement
• Specific event, measures a specific action (second purchase, feature activation, session start)

For SaaS: set return criteria to session_start to measure user return rate. For e-commerce: set to purchase to measure repeat purchase cohorts.

Step 4: Set Cohort Granularity and Size

Set Granularity to "Week" or "Month" depending on your business cycle. Monthly is standard for SaaS and longer-cycle products; weekly works for high-frequency apps. Set Cohort size to the number of periods to show per cohort (8 weeks = 2 months of cohort data visible).

Step 5: Add Segments for Channel Comparison

Add user segments (e.g., "Paid Traffic," "Organic Search," "Direct") to compare cohort retention across acquisition channels. This is where cohort analysis gets actionable: the difference in D30 retention between paid and organic cohorts tells you acquisition quality, not campaign volume.

Common failure: GA4's cohort report requires that conversion events have been active for at least as long as your cohort window. If you want 90-day cohorts, the conversion event must have been tracking for 90+ days. New events don't have historical data, set up events early and let them collect before you need the analysis.

Prooflytics

Turn scattered analytics into one clear picture

Every source in one brief. The whole picture. Your decision.

Start free trialSee pricing

14 days free · no credit card

03. Reading Cohort Retention Curves: D7, D30, D90

A retention curve plots what percentage of a cohort returns at each interval after acquisition. The shape matters more than any individual number:

Healthy retention curve: Steep early drop, then flattens into a stable baseline. Users who don't engage in the first week largely don't return, but those who do form a retained base that shows up consistently at D30 and D90.

Continuing decline: The curve keeps dropping with no flattening, no stable retained base. Indicates either a product failing to deliver on its promise or an acquisition audience that was never a good fit.

Benchmark anchors from the D7-D30 retention benchmarks guide:

• SaaS / productivity apps: D7 ~35-45%, D30 ~20-30%
• E-commerce: D30 repeat purchase rate ~25-35% (varies by category)
• Gaming / entertainment: D7 ~20-30%, D30 ~10-15%

D7 below 25% for SaaS is an onboarding signal, not a product signal, the problem is whether users complete the activation event (connect a data source, create their first project, invite a team member) in the first 7 days. Reducing first-30-day churn is largely a first-7-day activation problem.

04. LTV by Acquisition Cohort: The Formula and the Decision

The ICP problem LTV by cohort solves: standard LTV calculations use averages across all customers, which hides channel-level quality differences that are actually the most actionable data in acquisition analytics.

Simplified LTV formula per cohort:

LTV = ARPU × Gross Margin × (1 / Monthly Churn Rate)

Example for a SaaS cohort:

• ARPU: $150/month
• Gross Margin: 75%
• Monthly Churn Rate: 4%
• LTV = $150 × 0.75 × (1/0.04) = $112.50 × 25 = $2,813

If the paid social cohort from this acquisition month has a monthly churn of 6% instead of 4%, LTV drops to $1,875, a 33% reduction. If you paid the same $200 CAC for both, the CAC payback period for the paid social cohort is materially longer and may never recover margin.

Applied decision rule: when LTV by acquisition channel diverges by more than 25% between your highest and lowest channel, that's a budget reallocation signal, not an outlier. The channel with higher LTV justifies a proportionally higher CAC.

05. Cohort Analysis and Marketing Experiments: Avoiding Correlation Traps

The ICP problem cohort analysis without control groups creates: 70% of organizations don't run controlled experiments to evaluate marketing campaigns, and cohort analysis without control periods is subject to the same fundamental problem: you can observe that cohorts from a specific period retain differently, but you can't prove that a campaign caused it without a control group.

Marketing experiment structure as it applies to cohort analysis:

• Input: A hypothesis, "Changing the onboarding email sequence will improve D7 retention for paid social cohorts"
• Control group: The existing cohort (identical acquisition channel, no change)
• Test group: Same channel, new onboarding sequence
• Success metric defined upfront: D7 return rate (session event)
• Critical rule: Run both versions simultaneously, not sequentially, time is a confound. January cohorts and February cohorts are not the same experiment if seasonal behavior changes between them.

Harrah's Entertainment case illustrates this: when they compared their premium hotel+steak-dinner offer against a chips-only challenger offer, the challenge wasn't "which performed better", it was that most teams would have declared the premium offer superior based on cohort averages from different seasons. The controlled simultaneous split showed the opposite. The same bias applies to marketing cohorts: comparing December cohorts (holiday intent) to January cohorts (resolution intent) without controlling for season produces false acquisition quality signals.

Prooflytics HADI (Hypothesis, Action, Data, Insight) framework applies this structure: each hypothesis generates a test and a control cohort tracked over a defined window, with the delta between them constituting the measured campaign effect, not the absolute metrics from either cohort in isolation.

What to Watch: Cohort Analysis Warning Signals

1. Cohort D7 retention declining across consecutive acquisition months. If every new month's cohort retains worse than the previous, investigate product changes or onboarding flow changes that coincide with the decline.
2. Paid vs. organic D30 retention gap > 30%. Organic cohorts typically retain better because intent is higher. A >30% gap indicates the paid channel is acquiring low-fit users. Don't reduce paid spend, improve targeting or audience qualification.
3. Sudden LTV drop in a specific month's cohort. Cohort LTV declining in one specific month while adjacent months hold steady suggests an external event (competitor launch, pricing change, product bug), not a trend.
4. Cohort size declining for same-spend campaigns. Same budget, fewer conversions entering each cohort = CPL rising. This is a different signal from retention and requires CPC/conversion-rate diagnosis, not cohort analysis.
5. Cohort retention flat but net revenue retention declining. Retention (users returning) can be stable while expansion revenue from that cohort declines, indicating upsell failures or feature adoption problems. Check net revenue retention benchmarks alongside user retention.

Bottom line

• Acquisition cohort analysis isolates channel quality in a way attribution reports cannot, same CPL from two channels can produce radically different 90-day LTV.
• Build cohorts in GA4 Cohort Exploration using your actual conversion event, not "First visit", the report is only as useful as the entry event you define.
• D7 retention below 25% for SaaS is an onboarding problem, not a product problem, focus on first-7-day activation before feature investment.
• LTV by cohort = ARPU × gross margin ÷ churn rate, when cohort LTV diverges by channel by more than 25%, rebalance acquisition budget toward the higher-LTV channel.
• Cohort comparisons across different time periods are not controlled experiments, seasonal cohort bias is real; run simultaneous A/B splits to establish causation.

You can read independent reviews of Prooflytics on G2 and compare it to alternatives in the marketing analytics category.