cohort analysis for ecommerce: how to stop guessing about retention

TL;DR Cohort analysis groups customers by acquisition date and tracks their behaviour over time. It reveals which channels produce repeat buyers, when customers drop off, and whether your business is actually improving. We build cohort dashboards for ecommerce brands.

What Is Cohort Analysis (And Why Averages Lie)

A cohort is a group of customers who share a common characteristic — usually the month they made their first purchase.

Instead of looking at “average retention rate” across all customers (which means nothing), you track how each monthly cohort behaves over time. Did January’s customers buy again in month 2? Month 3? Month 6?

This is the difference between:

• “Our repeat purchase rate is 22%” (useless average)
• “January’s cohort has a 28% repeat rate at month 3, but March’s cohort only has 19%” (actionable insight)

The second statement tells you something changed between January and March. Maybe a product launch. Maybe an ad channel shift. Maybe a promotion attracted deal-seekers who never return.

What Cohort Analysis Reveals

1. Retention Curves by Acquisition Channel

Not all customers are equal. Cohort analysis by channel shows you:

In this example, email referrals produce customers with 2-3x the retention of Google Shopping. If you’re optimising purely on first-purchase CAC, you might be pouring money into channels that produce one-time buyers.

2. When Customers Churn

The retention curve shows you the “cliff” — the point where most customers drop off and never return.

For many DTC brands, it looks like:

• 50-70% never buy again after first purchase
• Of those who do return, most buy within 60 days
• After 90 days of inactivity, reactivation probability drops below 5%

This tells you exactly when to deploy retention tactics: email flows at day 14, 30, and 60 post-purchase. After day 90, those customers are essentially gone.

3. Seasonal Patterns

Cohorts acquired during peak seasons (Black Friday, Christmas) often have terrible retention. They came for the discount. They got their gift. They’re gone.

Meanwhile, cohorts from January-February (when only genuine interest drives purchases) often have the highest LTV.

This changes your perspective on “successful” holiday campaigns. High revenue in November means nothing if those customers never return.

4. Business Improvement Over Time

If your February 2026 cohort retains better at month 3 than your February 2025 cohort at month 3, your business is fundamentally improving. Better product, better experience, better retention flows.

This is the most important signal a cohort dashboard provides. It answers: “Are we getting better at keeping customers?”

How to Build Ecommerce Cohort Analysis

Step 1: Define Your Cohorts

The most common approach:

• Cohort definition: Month of first purchase
• Metric: Revenue (or orders, or % retained)
• Time periods: Monthly intervals since first purchase (Month 0, Month 1, Month 2, etc.)

Step 2: Get the Data

You need:

• Customer ID
• Order date
• Order revenue
• First order date (to assign cohort)
• Acquisition channel (for segmentation)

On Shopify, this lives in the orders and customers tables. Pull it into BigQuery via Fivetran for full flexibility.

Step 3: Build the Cohort Matrix

The core SQL logic:

1. Assign each customer a cohort (month of first purchase)
2. For each order, calculate months since first purchase
3. Aggregate: sum revenue (or count customers) by cohort x period

The output is a triangular matrix:

Step 4: Visualise

Three views that matter:

1. Heatmap — colour-coded cohort matrix (green = high retention, red = low)
2. Line chart — cumulative revenue per cohort over time (each line = one cohort)
3. Comparison chart — same cohort period, different years (e.g., all Month 3 values over time)

💡 This is what we do. We build automated cohort analysis dashboards in Looker Studio, connected to BigQuery, refreshing daily. Segmented by channel, product, and campaign. Delivered in ~3 weeks. Book a 20-minute discovery call — no pitch, just scoping.

Segmentation: Where the Real Insights Live

A single cohort view is good. Segmented cohorts are transformative.

By acquisition channel: Which channels bring customers who stick around?

By first product purchased: Do certain products lead to higher repeat rates? (This changes your ad strategy — lead with sticky products.)

By AOV bucket: Do high-AOV first purchasers retain better or worse?

By discount usage: Do customers who used a first-purchase discount have lower LTV? (Usually yes.)

By geography: Do certain markets have structurally different retention?

Each segment tells a different story and drives a different action.

Acting on Cohort Insights

Cohort data without action is just a pretty chart. Here’s what to do:

Common Mistakes

Mistake 1: Too-short observation windows. You can’t judge a cohort’s LTV at month 1. Wait for at least 3 months of data before drawing conclusions about a cohort’s quality.

Mistake 2: Ignoring sample size. A cohort of 50 customers is noise. You need 200+ per cohort for the retention curve to be statistically meaningful.

Mistake 3: Not controlling for seasonality. Compare same-season cohorts year over year, not adjacent months. January vs. January, not January vs. December.

Mistake 4: Only looking at revenue. Track both revenue retention (do they spend?) and customer retention (do they come back at all?). A small number of repeat buyers spending big can mask low overall retention.

The Technical Stack

For automated cohort analysis:

• Data extraction: Fivetran syncing Shopify orders daily into BigQuery
• Transformation: SQL models computing cohort assignments and matrices
• Visualisation: Looker Studio with filterable cohort heatmaps
• Refresh: Daily automated pipeline — always current

This runs in our cloud at Chartica. We manage the pipeline, the models, and the dashboards. You just open the report and make decisions.

Know someone drowning in spreadsheets? Share this guide with them.

If this sounds like more work than you want to take on, that’s what we do at Chartica. Book a 20-minute discovery call — we’ll scope it out, no pitch.