Cohort Analysis for SaaS and E-commerce: The Retention Truth

Cohort analysis is the analytical technique that separates SaaS and e-commerce teams who understand their business from those who think they do. Aggregate metrics — average revenue, churn rate, LTV — hide compositional changes that cohort analysis surfaces immediately.

This guide covers cohort analysis end to end. What cohorts reveal, how to build retention curves, LTV calculation, tools (GA4, Mixpanel, Looker), and how cohorts surface insights that aggregate metrics miss.

The work is conceptual more than technical. Done right, cohort analysis becomes the lens through which the team understands customer behaviour, retention, and revenue.

What cohorts actually are

A cohort is a group of customers who share an attribute — most commonly, their acquisition month or first purchase month.

Examples:

• "January 2026 cohort": all customers who first purchased in January 2026
• "Q1 2026 organic cohort": all customers acquired via organic search in Q1 2026
• "December 2025 promotion cohort": all customers acquired during the December 2025 holiday promotion

Cohort analysis tracks how each cohort behaves over time — retention, revenue, expansion, churn — separately. The pattern across cohorts reveals trends that aggregate metrics hide.

We covered the broader KPI framework in our marketing KPIs selection guide. Cohort analysis is the lens that makes retention KPIs actionable.

Why aggregate metrics lie

Aggregate metrics combine cohorts of different ages, sources, and qualities. The result is misleading averages.

The "improving churn rate" illusion

A SaaS business reports churn dropped from 8 percent to 5 percent over 12 months. Looks like retention improved.

The reality: they grew rapidly. The denominator (total customers) ballooned with new customers who haven't had time to churn yet. The actual cohort churn rates didn't change.

Cohort analysis would have shown: every cohort still loses 8 percent per month after they've been customers for a year.

The "rising AOV" illusion

An e-commerce business reports average order value rose 15 percent. Looks like customers are spending more.

The reality: a promotion attracted low-spend customers who churned. The remaining customers are the high-spend ones who never left. The remaining customer's AOV didn't change — the mix changed.

Cohort analysis would have shown: every cohort's AOV stayed flat. The aggregate moved because of churn composition.

The "improving LTV" illusion

A business reports LTV is up 20 percent. Looks like customers are worth more.

The reality: they changed how they calculate LTV, or they extrapolated future revenue more aggressively. The actual revenue per cohort didn't change.

Cohort analysis forces you to look at real revenue accumulated per cohort. No extrapolation magic.

Section 1 — Building a basic cohort table

The simplest cohort analysis: monthly cohorts of new customers, tracking retention over subsequent months.

The data structure

For each customer:

• Acquisition month (cohort assignment)
• Monthly activity (revenue, sessions, key actions)
• Final status (active, churned, paused)

Aggregated:

• Rows: cohorts (by acquisition month)
• Columns: months since acquisition (Month 0, Month 1, Month 2, ...)
• Cells: retention rate or revenue for that cohort in that month

The result is a triangular table — recent cohorts have fewer columns filled in because they haven't had time to age yet.

Interpreting the cohort table

Reading down a column shows: across all cohorts, how does retention look at Month N?

Reading across a row shows: a single cohort's full retention curve over time.

The interesting patterns appear in both dimensions:

• Column trends: are recent cohorts retaining better or worse than earlier ones?
• Row patterns: at what month does retention stabilise? Is the curve steep or shallow?

The shape of the retention curve

Healthy retention curves are:

• Steep early drop, then flatten: typical of free trials or low-commitment products. Big early churn, then stable users.
• Gradual decline that flattens: typical of subscription products with steady churn baseline.
• Linear decline: usually problematic. No retention floor.
• Curving downward forever: very problematic. Business is losing customers faster than acquiring.

The plateau level matters more than the early slope. A business that plateaus at 80 percent retention is healthier than one that plateaus at 30 percent.

Section 2 — Retention metrics that matter

Multiple retention metrics, each measuring different things.

N-day retention

Of customers acquired in cohort X, what percentage are still active N days later?

Common N values:

• Day 1 (next-day return)
• Day 7 (one-week retention)
• Day 30 (one-month retention)
• Day 90 (long-term engagement)
• Day 365 (annual retention)

Rolling retention

Of customers acquired in cohort X, what percentage are active in any of the past 30 days as of day N?

Less strict than point-in-time retention. Captures customers who use the product irregularly.

Revenue retention

Of revenue from cohort X in month 1, what percentage of that revenue is still active in month N?

For subscription products, this is more meaningful than user retention. Some customers downgrade, some upgrade. Revenue tells the whole story.

Net revenue retention (NRR)

Of revenue from cohort X in month 1, what is the total revenue (including expansion) in month N divided by month 1 revenue?

NRR can exceed 100 percent if expansion revenue offsets churn. Healthy B2B SaaS targets 110 to 130 percent NRR.

Cohort LTV

Sum of revenue from cohort X over all time tracked, divided by initial cohort size.

Truer than calculated LTV (which extrapolates). Limited to the time period you have data for.

Section 3 — Cohort analysis for SaaS

SaaS-specific cohort patterns and metrics.

MRR cohort tracking

Track monthly recurring revenue per cohort:

• Month 0: initial MRR from the cohort
• Month N: remaining MRR (or expanded MRR)
• Calculate net revenue retention from this

Activation cohorts

Group customers by whether they completed key activation actions:

• Cohort A: completed onboarding + first key action within 7 days
• Cohort B: completed onboarding but not key action
• Cohort C: didn't complete onboarding

Retention curves for these cohorts differ dramatically. Often Cohort A retains at 80+ percent month-over-month while Cohort C churns within 30 days.

This insight drives onboarding investment.

Plan tier cohorts

Group customers by initial plan tier:

• Free trial converters
• Starter plan
• Pro plan
• Enterprise plan

Different tiers have different retention patterns. Enterprise typically retains better. Starter often churns fastest. Pricing strategy can be informed by cohort analysis.

Acquisition source cohorts

Group customers by acquisition channel:

• Organic search
• Paid search
• Paid social
• Referral
• Direct

Acquisition source predicts long-term value. Often organic search and referral cohorts retain 2 to 3x better than paid social cohorts of the same size.

This insight changes channel investment. Lower-cost paid channels with low retention are net negative. Higher-cost organic content with high retention is net positive.

Section 4 — Cohort analysis for e-commerce

E-commerce-specific cohort patterns.

Repeat purchase rate by cohort

Of cohort X (customers acquired in month X), what percentage made a second purchase within 90 days?

This is the leading indicator for LTV. Cohorts with high repeat purchase rate produce 3 to 5x more lifetime revenue than cohorts with low repeat purchase.

Months-between-purchases

For each cohort, what's the average time between consecutive purchases?

Useful for:

• Email cadence planning (send reminders before next likely purchase)
• Inventory planning (predict reorder volumes)
• LTV calculation (assume future purchases at observed cadence)

Discount cohorts

Compare cohorts acquired via discount vs full price:

• First-purchase AOV
• Repeat purchase rate
• Long-term cohort revenue

Often, discount-acquired customers have lower repeat rates and lower long-term revenue. Discount strategy needs to factor this in.

Product cohorts

Group customers by their first product purchase:

• "First-product: hero product" cohort
• "First-product: long-tail" cohort

Often customers who first buy your hero product retain better and expand more than those who start with niche items. Marketing focus should reflect this.

Seasonal cohorts

Customers acquired in different seasons behave differently:

• Holiday cohorts: large first purchase, low repeat (gift buyers)
• Summer cohorts: smaller first purchase, higher repeat (organic discovery)
• New year cohorts: medium first purchase, high repeat (resolution-driven)

Marketing strategy should account for these patterns.

Section 5 — Tools for cohort analysis

The tooling landscape supports various approaches.

GA4 cohort exploration

Free, integrated. Limitations:

• Limited to GA4 events
• Cannot easily combine with backend data
• Visualisation options are basic

Good for: marketing-cohort analysis on tracked behaviour.

Mixpanel and Amplitude

Built specifically for cohort analysis.

• Drag-and-drop cohort builder
• Native retention curves
• Behavioural cohorts (e.g., "users who did action X")
• Good for product-led SaaS

Looker, Power BI, Tableau

Custom cohort analysis via SQL on your warehouse:

• Most flexible
• Requires data engineering
• Best for full LTV calculations combining multiple data sources

Stripe and subscription billing tools

For SaaS, Stripe Sigma or ProfitWell give you basic cohort analytics out of the box from billing data.

Spreadsheet cohort analysis

For small-scale or one-off analysis:

• Export customer data to Sheets or Excel
• Build cohort columns manually
• Limited to small datasets but quick

For most SMEs, start with GA4 cohort analysis or your billing provider's built-in cohort reports. Graduate to Mixpanel or custom SQL when needs grow.

Section 6 — Cohort analysis pitfalls

These are the patterns we see most often.

Cohort comparison without controlling for season

Comparing November 2025 holiday-shopping cohort against May 2026 cohort. Different conditions. Compare like-for-like seasonal cohorts.

Confusing rolling retention with point-in-time retention

These measure different things. Pick one and be consistent.

Calculating LTV with implied future retention

True cohort LTV is "revenue accumulated so far". Anything beyond is extrapolation. Be explicit about which you're showing.

Insufficient cohort age

A cohort that's only 1 month old doesn't tell you long-term retention. Wait at least 6 to 12 months for meaningful patterns.

Tiny cohorts producing noise

Cohorts with under 50 customers produce noisy retention curves. Aggregate small cohorts (e.g., monthly into quarterly) for statistical stability.

Forgetting churn definition consistency

"Churned" can mean different things. Customer cancelled? Customer hasn't paid in 30 days? Customer hasn't logged in for 90 days? Pick a definition and stick to it.

Section 7 — Translating cohort insights to action

Cohort analysis is interesting. The actions it enables are what matter.

Investment in retention vs acquisition

If your cohort retention is poor, acquisition spend is wasted. New customers churn faster than you replace them.

Cohort analysis often surfaces this. The strategic response: investing in onboarding, product, and customer success before scaling acquisition.

Source-specific budget allocation

If organic search cohorts retain 3x better than paid social cohorts, organic deserves disproportionate investment. Paid social channels that look cheap on CAC are actually expensive on cohort LTV.

Onboarding intervention

If Cohort A (activated) retains at 75 percent and Cohort C (didn't activate) retains at 15 percent, onboarding is the leverage point. Every customer pushed from C to A is worth 5x more revenue.

Product investment priorities

If users of feature X have 40 percent better retention than users who don't use feature X, that feature is more important than its usage stats suggest. Product team should prioritise its expansion.

Pricing strategy

If high-tier cohorts retain better than low-tier cohorts, the pricing ladder is working. If high-tier cohorts churn at similar rates to low-tier, there's a value-delivery problem at the high tier.

A 14-day cohort analysis project

If you're running your first systematic cohort analysis, follow this sequence.

Days 1 to 3 — Define cohorts. Pick the cohort dimension (monthly acquisition is standard). Define the metrics to track.

Days 4 to 7 — Build the base cohort table. Use GA4, Mixpanel, or SQL on your warehouse. Cover the last 12 to 18 months.

Days 8 to 10 — Build segmented cohort views. By source, plan tier, first product, behaviour. Each segment is a separate analysis.

Days 11 to 12 — Synthesize insights. What patterns surfaced? What actions does each pattern suggest?

Days 13 to 14 — Prioritise interventions. Pick top 2 to 3 actions based on cohort insights. Plan implementation.

The output is a cohort framework that informs strategic decisions for the next year. Re-refresh quarterly.

A real example — Tunis estate agency cohort analysis

A Tunis estate agency we work with had stable monthly leads but declining revenue. Aggregate metrics looked fine. Cohort analysis revealed:

• Cohorts acquired via paid search converted at 8 percent (acceptable)
• Cohorts acquired via paid search closed deals 18 percent of the time (decent)
• Cohorts acquired via referral converted at 22 percent (excellent)
• Cohorts acquired via referral closed deals 41 percent of the time (excellent)

The aggregate "close rate of 25 percent" hid this dramatic source-level variation. Paid search was driving high-volume but lower-quality cohorts. Referrals were driving higher-quality cohorts but at smaller volume.

Action: shifted 30 percent of paid budget to referral incentives and organic content. Total qualified pipeline (close rate × lead count) improved 34 percent over 90 days at same total spend. The full story is in our Tunis estate agency case study.

Common cohort analysis mistakes

These are the patterns we see most often.

Comparing cohorts of different sizes. Small cohorts produce noisy retention curves. Use confidence intervals or aggregate.

Confusing aggregate trends with cohort behaviour. Aggregate metrics can move without cohort changes — composition effects.

Skipping segmented cohorts. Channel-segmented, behaviour-segmented cohorts surface insights that monthly cohorts miss.

Treating LTV as static. LTV varies by cohort, channel, product, season. Use cohort-specific LTV, not aggregate.

Cohort analysis without action. Insights are interesting. Shipped changes are what matter.

One-time analysis. Run cohort analysis quarterly. Patterns shift over time.

Frequently asked questions

How long after acquisition can I trust cohort metrics?

For most businesses, 3 to 6 months of cohort age gives meaningful patterns. Less than that and you're seeing noise.

Can I do cohort analysis without a developer?

For basic cohorts: yes via GA4, Mixpanel, or your billing tool. For custom segmented cohorts: usually needs SQL or developer help.

How big should a cohort be for reliable analysis?

50+ customers minimum per cohort. 200+ is preferable. Below 50, aggregate smaller periods into larger cohorts.

What's the difference between cohort retention and customer retention?

Customer retention is a single number (overall). Cohort retention is a curve over time per cohort. Cohort is more informative.

Should I use cohort LTV or model-extrapolated LTV?

Cohort LTV (revenue accumulated so far) is the truth as observed. Extrapolated LTV is a forecast. Use both, distinguish clearly which is which.

How does cohort analysis differ from journey analysis?

Cohort tracks groups of customers over time. Journey tracks individual customer paths. Both are useful, for different questions.

Get a cohort analysis audit

We run cohort analysis projects on SME accounts. Within 48 hours of discovery we deliver a sample cohort table, observed retention patterns, and recommended interventions.

Book a free 30-minute audit. We screen-share, walk through your customer data, and you leave with a clear plan.

Or explore our CRO service for the full system we run on accounts that need integrated CRO and analytics work.