Most operators analyse churn. They study who left, try to figure out why, and hope that stops more of it. The data is right there in Stripe so it feels productive. But it is backward-looking. You are learning from customers you already lost, after the fact.
The more useful question runs the other way. Who stays? What do they share? And can you spot the pattern early enough to double down on it before twelve months of LTV data proves it for you?
I have watched operators run this analysis on their own top customers and it is the same story every time. The signal is already there. It just lives in four tools, and nobody is reading the four tools together.
What a valuable cohort analysis actually is
A valuable cohort is not “high-LTV customers.” It is the specific combination of things those customers share: the acquisition channel, the first product they bought, the email they opened in week one, the one support ticket they filed and how it went. That combination is what predicts long-term retention. It spans four tools at minimum, which is why no single dashboard shows it to you.
Most retention reports start with aggregates. Percentage retained, by cohort, by month. A valuable cohort analysis starts with individuals. You pick twenty of your best customers and you trace what each of them actually did. After fifteen traces the shared pattern gets loud. After twenty it is obvious.
That pattern will not show up in any aggregate view. The signal only exists when you read four tools at once.
Three customers who stayed, and what each one tells you
Here are three patterns, rewritten as customers. All in your top 20% by LTV. All still subscribed. Completely different paths, and each one points to a different lever.
James, the organic search power user. James signed up through a Google search for something specific in your category last September. He picked the Pro plan on day one instead of starting on the free tier. And on day three, he used your migration wizard to import his data from a competitor’s tool. He has never filed a support ticket. Six months in, his usage has only climbed, and he is on track to become a multi-year subscriber.
No single tool tells you that “Google Search plus Pro tier plus used the migration wizard in the first 72 hours” is a retention pattern. Your analytics tool has the acquisition source. Stripe has the plan tier. Your product analytics tool (Mixpanel, PostHog, Amplitude, whatever you use) has the feature usage event. The pattern only emerges when you read all three at once. Trace ten more customers like James and it holds. That is a lever you can pull. Not “do more SEO,” but “push the migration wizard into the first-session onboarding checklist so more organic Pro signups trip the same conversion pattern James did.” That is a different team’s backlog, and the lever only exists because you joined three tools to find it.
Maria, the referral high-spender. Maria came from a friend’s referral link in July. Her first order was three products at once. She contacted support in month two with a sizing question. The agent sorted it fast, she left a 5 out of 5, and she has been on the premium tier for nine months.
Referral customers arrive warmer because someone they already trust vouched for you. What is interesting about Maria is the single support interaction. Omniconvert’s research shows that customers who contact support once and have a positive experience retain better than customers who never contact support at all. The support moment is not friction here. It is a trust-building event. Referral plus high first-order AOV plus one positive support ticket is a cohort worth naming. Your referral program is not just acquiring customers. It is acquiring the ones who already trust you by the time they arrive.
Priya, the Instagram slow-burner. Priya came from an Instagram ad in August. Her first order was tiny, a single low-price item she picked up on a promo. On first-order data alone she looked like a throwaway one-timer. But she opened every Klaviyo email for six weeks, clicked through to product pages, came back for a second one-off order in week four, and then converted to a monthly subscription in week eight. Nine months later she is in your top 10% by LTV.
Express Analytics found that email engagement in the first 30 days is one of the strongest predictors of long-term retention. Priya’s first purchase said almost nothing about her value. Her Klaviyo engagement said everything. But the outcome only showed up in Stripe two months later, and by then a less patient operator would have already written her off. If you had looked at her first 30 days of email behaviour instead of her first-order AOV, you would have flagged her as high-potential before she upgraded.
Three customers, three levers. James tells you to push the migration wizard into the onboarding flow so more organic Pro signups activate the same way he did. Maria tells you to invest harder in referral and treat post-support follow-up as a retention moment, not a ticket close. Priya tells you to stop judging Instagram customers by their first order and start watching their first 30 days instead. Treating all three as “retained customers” and doing nothing different is how growth decisions get made on averages.
Why aggregates hide your best customers
Blended LTV. Blended CAC. Blended retention rate. These are the numbers most operators track, and they describe nobody in particular.
Your best cohort might retain at 95% while your worst retains at 30%. The blended number, 62%, represents zero actual customers and anchors every growth decision you make. Recurly’s 2026 benchmark data puts subscription churn at 52% within the first year. That is an average. Some cohorts in that dataset churn at 70%. Others at 20%. The difference between those two groups is the most important signal in your business, and the average erases it.
Same problem with CAC. If your Google Search customers retain at twice the rate of paid social, their effective cost per dollar of eventual LTV is much lower even at a higher headline CAC. But blended CAC treats both channels as equal contributors. You end up over-investing in the cheap customers who leave in three months and under-investing in the expensive ones who stay two years.
None of this is a data quality problem. It is a connection problem. Retention lives in Stripe. Acquisition lives in Shopify. Engagement lives in Klaviyo. Support lives in Gorgias. Each tool is accurate in its own lane. The cohort pattern that joins all four lanes is where the growth lever lives, and the lever only exists when you read the tools together.
How to do this yourself in 30 minutes
Here is the version that works with four browser tabs and a text document. It is manual, it is rough, and the first time you do it you will see at least one pattern you did not know was there.
The walkthrough below uses a DTC stack as the example (Stripe, Shopify, Gorgias, Klaviyo). If you are B2C SaaS, swap Shopify for your analytics tool, Gorgias for Intercom or Zendesk, and Klaviyo for your product analytics tool (Mixpanel, PostHog, Amplitude). The process is identical. You are still joining billing to acquisition to product behaviour to support, just through different doors.
Step 1, 10 minutes: pull your top 20% by LTV from Stripe. Filter for active subscriptions, sort by total revenue, export the top quintile. You want customer email, subscription start date, total revenue, and current plan. Sample 15 to 20 rows if your list is huge.
Step 2, 5 minutes: trace their first order in Shopify. Take those emails into Shopify. What was the acquisition channel? What product did they buy first? What was their first-order AOV? You are looking for clustering. Do your best customers come from one or two channels? Do they tend to start on a specific product?
Step 3, 5 minutes: check their support history in Gorgias. Search for the same emails. How many filed tickets? What happened on each one? What were the CSAT scores? You might find your best customers either never contact support, or contact once and have a great experience. Either pattern is useful.
Step 4, 5 minutes: check their email engagement in Klaviyo. Pull up their profiles. Look at open rates and clicks across the first 60 days. Did they engage with onboarding flows? Did they click through campaigns? You are looking for whether early email behaviour correlates with retention in your specific business.
Step 5, 5 minutes: find the shared pattern and name it. Group what you found. What is over-represented? What clusters? Write the cohort down. “Google Search / mid-tier first purchase / high onboarding engagement” is a cohort. “Referral / high first-order AOV / one positive support ticket” is a different cohort. Each has its own growth lever.
Run it on fifteen customers. Then run it again next week on fifteen different ones. After two or three rounds the same patterns keep surfacing, and you will know which ones to double down on and which ones to quietly pivot away from.
Where this falls apart at scale
The manual version works for 20 customers. It breaks when you try to keep the pattern fresh across your whole customer base, or when sampling stops being reliable. At 5,000 customers and three active acquisition channels, twenty-sample patterns start drifting with each run and you lose trust in what you are seeing.
That is the work Riven does in the background. Stripe, Shopify, Gorgias and Klaviyo get linked by customer identity, the same person rather than just the same email string, and cohort patterns surface automatically. What takes you 30 minutes for 20 customers runs across every customer you have, every day, without a CSV export in sight. When a new cohort starts forming, you see it in hours. When one quietly degrades, you see that too, in time to pivot rather than write the post-mortem.
That is the whole point. The slow version of this analysis tells you what happened. The fast version tells you what to do about it, while it still matters.
A note on what to do with the answer
Finding your most valuable cohort is half the job. The other half is treating it like a decision input instead of a curiosity.
When you name a cohort, write the lever next to it. “Double down on mid-tier organic.” “Pivot Instagram budget into the flows that caught Priya.” Then put that lever in front of whoever owns the budget. Cohort analyses that sit in a Notion page nobody reads do not change anything. Cohort analyses that show up as a decision in next week’s budget meeting do.
A churn report tells you who you lost. A cohort retrospective tells you who you should be finding more of. Start with your best, trace what they did differently, then double down where it is working and pivot where it is not. It takes 30 minutes the first time. It will change how you think about growth after that.
Frequently asked questions
How do I define “valuable”?
Start with total lifetime revenue. Total money paid across the customer’s entire relationship with you. It is the simplest and most reliable measure. Later you can factor in referral value, support cost per customer, and expansion revenue. For your first pass, Stripe total revenue is enough.
How many customers per pattern before I can act on it?
Around 15 to 20 showing the same pattern. Below that you might be seeing noise. Above 30 you usually have enough to make a budget decision. If your sample is small, run it again next month and see whether the cohort holds.
What if my best cohort is too small to scale?
That is common and it is still useful. If your strongest cohort is “referral premium buyers” and they are 5% of your base, that tells you two things. Protect those customers fiercely, because they are disproportionately valuable. And try to engineer the same conditions in other channels. Can your referral program produce more of them? Can your onboarding nudge other customers toward premium behaviour?
Should I do the same analysis for churned customers?
Yes, but second. Start with your best because the signal is clearer and the action is more motivating. Once you know what your top cohort looks like, run the same process on your highest-churn group. The contrast between the two usually reveals the specific intervention point where a customer either becomes a long-term subscriber or drifts toward cancellation.