Your store acquired 1,200 new customers last month. Good news? Maybe. It depends entirely on what those customers do next.
Standard ecommerce metrics -- monthly revenue, total customers, average order value -- treat your customer base as a single undifferentiated mass. They tell you what happened across all customers in a given period. What they don't tell you is whether the customers you acquired in March behave differently from those you acquired in June, or whether your retention is improving or deteriorating over time.
Cohort analysis fixes this blind spot. It groups customers by when they first purchased, then tracks each group's behavior over subsequent periods. The result is one of the clearest views you can get of customer quality and business health.
What a cohort is
A cohort is simply a group of customers who share a common starting event in the same time period. In ecommerce, the most common cohort definition is month of first purchase.
Every customer who made their first purchase in January belongs to the January cohort. Everyone who first purchased in February belongs to the February cohort. And so on.
Once you've defined cohorts, you track each one over time: what percentage made a second purchase within 30 days? Within 60 days? What's their cumulative revenue at month 3, month 6, month 12?
This time-based view reveals patterns that aggregate metrics hide completely.
How to read a cohort retention table
A cohort retention table is the standard format for presenting cohort data. Here's what one looks like:
| Cohort | Customers | Month 0 | Month 1 | Month 2 | Month 3 | Month 4 | Month 5 | Month 6 |
|---|---|---|---|---|---|---|---|---|
| Jan | 840 | 100% | 28% | 19% | 15% | 13% | 12% | 11% |
| Feb | 920 | 100% | 26% | 18% | 14% | 12% | 11% | -- |
| Mar | 1,100 | 100% | 22% | 14% | 11% | 10% | -- | -- |
| Apr | 780 | 100% | 30% | 22% | 17% | -- | -- | -- |
| May | 850 | 100% | 31% | 23% | -- | -- | -- | -- |
| Jun | 910 | 100% | 32% | -- | -- | -- | -- | -- |
Here's how to read it:
- Month 0 is always 100% -- that's the month the cohort made their first purchase.
- Month 1 shows what percentage of the cohort made another purchase in the following month. For the January cohort, 28% of the original 840 customers (about 235 people) purchased again in February.
- Each subsequent column shows the percentage still active in that month.
The diagonal reading (top-right to bottom-left) tells you how different cohorts perform at the same stage of their lifecycle. The January cohort had 28% retention at Month 1. The March cohort had 22%. The June cohort had 32%.
What the numbers tell you
Declining cohorts: something went wrong
In the table above, the March cohort stands out. It's the largest cohort (1,100 customers) but has the worst Month 1 retention (22%) and the steepest drop-off. That's a red flag.
Large cohort size + low retention often means you ran a broad acquisition campaign that attracted low-intent buyers. Maybe you ran a deep discount promotion, or a viral social media post brought traffic that was curious but not committed. The raw customer acquisition numbers looked great. The retention data tells the real story.
When you spot a declining cohort, ask: what was different about that month? Common causes include:
- Heavy discounting that attracted one-time bargain hunters
- A marketing channel shift (new platform, new audience) that brought lower-quality traffic
- Product issues -- a new product launched that month had quality problems or didn't meet expectations
- Fulfillment problems -- shipping delays or packaging issues during that period damaged the first impression
Improving cohorts: something is working
The April through June cohorts show improving retention -- 30%, 31%, 32% at Month 1. Something changed that made new customers more likely to come back.
When you spot this pattern, work backward to identify the cause. Possibilities include:
- Better post-purchase experience -- maybe you implemented a follow-up email sequence or improved packaging
- Higher-quality acquisition -- a shift in ad targeting or channel mix brought more intentional buyers
- Product improvements -- new products or formulation changes increased satisfaction
- Seasonal effects -- some months naturally produce higher-intent buyers (holiday shoppers vs. January browsers)
The goal is to find the cause and replicate it. If your improved cohort retention coincides with launching a post-purchase email sequence, that sequence is demonstrably valuable -- and you can quantify exactly how valuable by comparing the pre- and post-change cohorts.
The retention curve shape
Beyond comparing cohorts to each other, look at the shape of each cohort's retention curve.
Healthy pattern: A steep initial drop (Month 0 to Month 1) that flattens out over time. The January cohort goes from 100% to 28% (steep) but then flattens: 19%, 15%, 13%, 12%, 11%. The customers who survive the first drop-off tend to stick around. This is normal.
Concerning pattern: A retention curve that keeps dropping at a steep rate and doesn't flatten. If a cohort goes 100% > 28% > 16% > 8% > 4% > 2%, the curve never stabilizes. You're not building a base of repeat customers -- you're churning through one-timers.
Excellent pattern: A curve that flattens early and stays flat, or even ticks up. If a cohort stabilizes at 15% by Month 3 and stays at 14-16% through Month 12, those customers have formed a purchasing habit. The "ticking up" scenario happens when seasonal effects or loyalty programs re-engage customers who had lapsed briefly.
Revenue cohorts vs. retention cohorts
The table above shows retention -- what percentage of customers are still purchasing. Revenue cohorts show a different, complementary view: the cumulative revenue generated by each cohort over time.
| Cohort | Customers | Month 0 | Month 3 | Month 6 | Month 12 |
|---|---|---|---|---|---|
| Jan | 840 | $54,600 | $88,200 | $109,600 | $142,800 |
| Feb | 920 | $57,040 | $89,240 | $108,560 | -- |
| Mar | 1,100 | $60,500 | $82,500 | -- | -- |
| Apr | 780 | $52,260 | $91,260 | -- | -- |
Revenue cohorts reveal something retention tables don't: the quality of repeat purchases. The April cohort has fewer customers than March but generated more revenue by Month 3. This tells you April's customers are not only returning at higher rates -- they're spending more per order when they do.
Revenue per customer by cohort age is the clearest measure of customer quality. Calculate it by dividing cumulative cohort revenue by the original cohort size:
- January cohort, Month 12: $142,800 / 840 = $170 per customer
- April cohort, Month 3: $91,260 / 780 = $117 per customer
If April is already at $117 by Month 3 while January was at $88,200 / 840 = $105 at the same point, the April cohort is tracking significantly higher. Whatever drove the April cohort is producing better customers.
Connecting cohorts to business changes
Cohort analysis becomes powerful when you overlay it with a timeline of business changes. Keep a simple log of major changes by month:
- January: Launched new product line
- February: Increased ad spend on Instagram by 40%
- March: Ran site-wide 25% off sale
- April: Hired customer service rep, launched post-purchase email sequence
- May: Switched to faster shipping carrier
- June: Launched referral program
Now the cohort data has context. March's poor retention coincides with the 25% off sale -- confirming that deep discounts attract low-retention customers. April's improvement aligns with better customer service and post-purchase emails. May and June continue the improvement, possibly compounded by faster shipping and the referral program (which brings in customers referred by existing loyal buyers -- a higher-quality source).
Without the cohort view, you'd see total revenue going up and assume everything is fine. With it, you can see that March's acquisition strategy was actively harmful to long-term value, even though it produced the largest cohort.
Common cohort analysis mistakes
Using too short a window. A 3-month cohort analysis shows initial retention but misses the long game. For most ecommerce businesses, 12 months is the minimum useful window. Some product categories (furniture, electronics) need 24+ months.
Ignoring cohort size differences. A cohort of 200 customers with 35% Month 1 retention is not necessarily better than a cohort of 1,200 with 28% retention. The larger cohort still produced more absolute repeat customers (336 vs. 70). Always consider both percentage and absolute numbers.
Not segmenting within cohorts. A monthly cohort includes customers from every acquisition channel, product category, and price point. If your Facebook ads and Google search bring fundamentally different customer types, combining them in one cohort obscures both. Segment by channel when possible.
Comparing cohorts at different lifecycle stages. Your most recent cohort will always look worse in absolute terms because it's had less time to mature. Compare cohorts at the same lifecycle point -- Month 3 to Month 3, Month 6 to Month 6 -- not the latest snapshot.
How Spark handles cohort analysis
Building cohort tables manually from raw order data is possible but tedious. You need to identify each customer's first purchase date, assign them to a cohort, then calculate retention and revenue for each subsequent period across all cohorts. It's a multi-join SQL query at best, a multi-tab spreadsheet at worst.
Spark by MishiPay builds cohort analysis automatically from your connected store data. Ask "Show me my customer cohort retention for the last 12 months" and you get a retention table with trends highlighted. Ask "Which acquisition month produced the highest-value customers?" and Spark identifies the best-performing cohort with revenue-per-customer comparisons.
You can also drill into specific cohorts: "What's different about the March cohort?" prompts an analysis of that cohort's characteristics -- acquisition channels, first-purchase products, average first order value, and discount usage -- compared to better-performing cohorts.
The bottom line
Cohort analysis answers the question that aggregate metrics can't: is your business getting better at acquiring and retaining customers over time?
A growing business with declining cohort quality is building on a weakening foundation. Revenue goes up because you're acquiring more customers, but each batch is less likely to come back. Eventually, acquisition costs outrun the declining lifetime value, and growth stalls.
A business with improving cohort retention and revenue is doing the opposite -- compounding. Each new batch of customers is more valuable than the last, which means your acquisition spend becomes more efficient over time.
Track your cohorts monthly. Overlay them with your business changes. The patterns will tell you more about the health of your business than any top-line revenue number ever will.
Track your cohort retention automatically
Spark builds cohort retention and revenue tables from your Shopify, WooCommerce, or Square data. See which customer groups are thriving and which are fading.