Customer Lifetime Value, Without the Myths: Cohorts, CAC Payback and What a Customer Is Really Worth

Two customers arrive at your store, each acquired by an ad, each costing you €54. The first buys a €60 accessory and is never seen again. The second buys the same €60 accessory, comes back in September, again in December, and brings €240 of profit within a year.

Your ads dashboard reports both as identical conversions. Same cost, same first order, same ROAS contribution. Every budget decision you make on that dashboard treats those two customers, and the campaigns that found them, as equally valuable. They are not, and the tool for telling them apart has been around for decades: cohort analysis, the honest version of "customer lifetime value."

This post explains how LTV and cohort tables actually work (in plain language, because the tables look scarier than they are), what real repeat-purchase data says instead of the recycled myths, how businesses have used cohorts to make real decisions, and how we built this into Enalitica so the math runs on profit rather than flattering revenue.

What LTV Actually Is (and the Revenue Trap)

Customer lifetime value is the answer to one question: how much is a customer worth to you over time, in total? Not their first order. Everything: the first purchase, the repeat in September, the gift order in December, minus the refund in January.

The trap is the unit. Most tools and most blog posts compute LTV in revenue, and revenue cannot repay an ad bill. If a customer generates €500 of revenue at a 40% margin, they produced €200 you can actually spend; a customer with €300 of revenue at 70% margin produced €210 and is the more valuable one, even though revenue-LTV ranks them backwards. Computed on revenue, an LTV:CAC ratio of 3.3 can hide a profit ratio of 1.3. Everything in this post, and everything in Enalitica, therefore uses profit-based LTV: revenue without VAT, minus product costs, shipping, payment fees, with refunds subtracted, the same engine as POAS.

Cohort Tables: Each Row Has Its Own Clock

A cohort is simply everyone who bought from you for the first time in the same month. The July cohort, the August cohort, and so on. The cohort table tracks each group's cumulative profit per customer as the months pass.

Here is the part that confuses almost everyone at first: the columns, labeled M0, M1, M2, are not calendar months. They count months since that row's customers first bought. Each row runs on its own clock:

For the 2026-07 row, M0 is July and M1 is August. For the 2026-06 row, M0 is June and M1 is July. When August arrives, the July row does not shift into M1; the July row stays where it is, and its M1 cell fills in with whatever those same customers bought during August. Empty cells with a dot simply have not happened yet.

Why build the table this sideways way? Because it lines every generation of customers up by age. Reading down the M0 column answers "are the customers I acquire getting better or worse in their first month?" without older cohorts winning just because they have had more time. And one elegant side effect: a single calendar month cuts across the table on a diagonal, so a July sale, price change or site outage shows up as a jump across every row at once.

The Three Numbers That Matter

A good cohort view compresses into three numbers per row.

1. nCPA: what a new customer cost you. Take a month's entire marketing spend (ad platforms plus your agency, tools and influencer costs) and divide by the number of first-time customers that month. Note this is a blended number that deliberately charges all marketing to new customers; repeat buyers ride free, which keeps it conservative.

2. Payback month: when the customer has earned their cost back. Follow the row's cumulative profit per customer until it crosses nCPA. Cross at M0 and your customers pay for themselves with the first order, the strongest position in e-commerce. Cross at M4 and you are financing four months of ad spend from cash flow, which is survivable with a healthy repeat rate and lethal without one.

3. LTV:CAC: the multiple. Cumulative profit per customer divided by nCPA. Below 1.0 the customer never repays their acquisition; above it, every increment is real gain. One honesty rule that most dashboards skip: a cohort that is only three months old can only give you a three-month ratio. Presenting an M3 ratio as "12-month LTV:CAC" flatters young cohorts and is one of the classic ways cohort analysis goes wrong.

What the Data Actually Says (Benchmarks Worth Trusting)

The internet is full of recycled benchmark numbers; the research for this post traced most of them to their sources, and the trustworthy residue is smaller than you would expect.

• Repeat rates are lower than the folklore. The most-quoted "average repeat purchase rate of 28%" traces back to one small study of retention-focused stores. A more recent dataset of 156,000 DTC customers found 18.8% placed a second order within a year, and among those who did, half did it within 30 days. Consumables (supplements, cosmetics, food) run structurally higher; furniture and durables structurally lower.
• Returning customers punch far above their weight. Adobe's classic transaction-data study found repeat purchasers were 8% of visitors but roughly 41% of revenue. The often-quoted escalation (a customer who bought twice is far more likely to buy a third time than a first-time buyer is to buy a second) holds directionally in every dataset we could find.
• Acquisition keeps getting more expensive. SimplicityDX estimated brands went from losing $9 per newly acquired customer in 2013 to $29 today. Rising acquisition costs are exactly why knowing your customers' downstream value stopped being optional.

And a benchmark warning from our own data: if your repeat rate is low (many stores see single digits), LTV analysis does not tell you to "invest in the long game." It tells you the opposite: the first order must be profitable on its own, because there mostly is no second one. That is a POAS conclusion, and it is just as valuable as discovering hidden repeat value, because it stops you from overspending on a promise your data says will not materialize.

The Myths Worth Retiring

LTV content is the most myth-dense corner of e-commerce writing. Four corrections, each traceable:

"Acquiring a customer costs 5 to 25 times more than retaining one." This exact phrasing comes from a 2014 Harvard Business Review article which cites no study for the range, and researchers who went looking for the original evidence never found one. Retention is often cheaper; how much cheaper depends entirely on your business, which is precisely what your own cohort table measures.

"Increasing retention 5% increases profits 25 to 95%." The real source is Reichheld and Sasser's 1990 work on banks and insurance brokers, single-company case results from 1980s service industries. Fine research, wrong universe to quote as an e-commerce law.

"A healthy LTV:CAC is 3x." The 3x rule was written by David Skok for SaaS companies: recurring multi-year contracts at 75-85% gross margins. E-commerce is transactional with far lower margins and a repeat curve that mostly plays out inside 12 months. The honest e-commerce floor is simpler: below 1.0 on profit-based LTV you are losing money on every customer; how far above 1.0 you need to be depends on your fixed costs, not on a SaaS benchmark.

"My analytics already shows retention." GA4's cohort report keys on the browser's client ID, not the person. A returning customer on a new phone, a cleared cookie, or a different browser enters the table as a brand-new user. Order-based cohorts keyed on the actual customer identity routinely show repeat rates dramatically higher than GA4 reports, which is the same pixels-versus-orders gap that distorts attribution.

What Cohort Analysis Changes in Practice

Three documented stories show the kinds of decisions this table drives.

Killing the discounts that killed the margin. When Drew Sanocki joined the bankrupt streetwear retailer Karmaloop as CMO, cohort segmentation showed a tiny "whale" segment (about 1.3% of visitors generating roughly 43% of revenue) while blanket 30-40% discounts were converting would-be full-price repeat buyers into deal hunters with negative lifetime value. The fix that followed from the cohort data: no discounts in the first 30 days after a purchase, when customers were likely to reorder at full margin anyway. The company went from bankruptcy to a strategic sale in under two years.

Raising spend because 60-day LTV said so. Common Thread Collective's brand Bambu Earth could not acquire customers profitably at its conversion rate and order value. The lever that worked was not cheaper ads but higher customer value: a skin quiz plus mini-kit entry products lifted 60-day LTV by 49%, which raised how much they could afford to pay per customer, which unlocked profitable scaling.

Discovering which first product creates the best customers. Flannel brand Dixxon found (in a vendor-published analysis) that customers whose first purchase was one product line were worth roughly double the customers acquired by another line a year later. The first product a customer buys is an acquisition channel in disguise.

That last pattern is why discount analysis belongs in the cohort table too:

One research-backed nuance before you kill your welcome offer: a field study by Anderson and Simester found discounts increased future purchases by brand-new customers while decreasing them among established customers, who learn to wait for deals. The danger is not the intro discount; it is training your loyal base. Your own two curves will tell you which effect dominates.

The Payoff: Knowing What You Can Afford to Pay

Direct-response marketers have repeated a line for decades, popularized by Dan Kennedy: "Ultimately, the business that can spend the most to acquire a customer wins." It sounds reckless until you see what makes it safe: you can only outspend competitors when you know what a customer is worth to you, and they are guessing.

The cohort table turns the slogan into arithmetic. If your matured cohorts show a customer is worth €72 of profit by month three, and you are acquiring at €54, then €72 is your allowable CAC at a 3-month payback target, and the €18 gap is your scaling headroom: the auctions you can win, the audiences you can afford to test, while competitors who only see first-order ROAS must stop bidding. If the numbers run the other way (nCPA above your M3 value), the same table is an early-warning system telling you payback is drifting out and cash flow will feel it.

How Enalitica Builds This (and Where We Refuse to Guess)

We built the cohort view into the profit dashboard rather than as a separate module, because LTV inherits every cost decision the profit engine already makes. A few choices worth explaining:

Profit, not revenue. Every cell is cumulative contribution profit per customer: revenue without VAT, minus product costs, shipping, payment fees, with refunds subtracted. Of the LTV tools we surveyed while researching this post, nearly all compute cohort LTV on revenue. A revenue toggle exists, mostly so you can see how much the flattering version overstates.

Honest maturity, no projections. Averages at month M only include customers old enough to have reached month M, young cohorts are labeled "maturing", and any ratio measured before twelve months carries a visible tag (like 2.9x @M3) instead of impersonating a 12-month figure. We also deliberately ship no predicted LTV: prediction models in this category publish no accuracy numbers, and practitioners report they routinely overestimate. We would rather show you a real number with a maturity label than a guess with confidence.

Four ways to slice the same customers. By acquisition month (is customer quality trending up or down), by the channel that brought the first order (which channel acquires customers who come back, using our own order-level attribution rather than platform claims), by first product (which entry products create your best customers), and by first-order discount versus full price (the Karmaloop question). Each view carries its own plain-language reading guide, because a cohort table you cannot read is furniture.

The allowable-CAC card. The table's conclusion, computed for you: what your matured cohorts say a customer is worth by M3 and M12, your ceiling at a 3-month payback target, what you currently pay, and the headroom between them.

The known limitation, stated. Customers are identified by email. A guest who buys twice with two different addresses counts as two people, so your true repeat rate is a bit higher than shown; the table reports its own coverage so you know how much weight to put on it.

Getting Started: A 30-Minute Checklist

1. Find your real repeat rate. Not GA4's; your order database's. Count customers with a second order within 12 months. This single number decides whether your growth story is repeat value or first-order profit.
2. Compute one cohort by hand. Take January's first-time customers, sum their profit through June, divide by their count. One row is enough to feel how the table works.
3. Compute your nCPA. A month's full marketing spend divided by that month's new customers. Compare it with the cohort profit from step 2: are you above or below water, and by which month?
4. Set your allowable CAC. Decide a payback target (3 months is a sane default for stores without deep cash reserves), read the matching cumulative profit per customer, and treat that as your bidding ceiling.
5. Check who your discounts acquire. Split cohorts by discounted versus full-price first order before your next sale season, not after.

In Enalitica the whole checklist is one screen on the profit dashboard: cohorts compute from your orders and costs automatically, in both profit and revenue, across all four dimensions, with the allowable-CAC card on top. If you want to see what your own customers are worth, book a demo or start on the free plan; the first cohort table usually settles a debate you have been having on gut feel for years.

Frequently Asked Questions

What is a good LTV:CAC ratio for an e-commerce store?

On profit-based LTV, 1.0 is the absolute floor: below it a customer never repays their acquisition cost. How far above 1.0 you need depends on your fixed costs and payback tolerance, not on the widely quoted 3x, which was written for SaaS businesses with 75-85% margins and recurring contracts. Many healthy stores run between 1.5x and 4x on a 12-month profit basis. More important than the ratio's absolute level is its trend across cohorts and the payback month attached to it.

How long before I can trust my cohort numbers?

A cohort is fully readable at the age you want to measure: trust an M3 number after three months, an M12 number after a year. The classic mistake is comparing a two-month-old cohort's total against a mature one's and concluding LTV is collapsing. Any tool (including ours) should label immature cohorts and never present a young ratio as a 12-month one.

Should LTV be based on revenue or profit?

Profit. Revenue-based LTV systematically overranks customers who buy expensive, low-margin, frequently returned products, and it can make an unprofitable acquisition strategy look healthy (a 3:1 revenue ratio at 40% margin is roughly break-even). If your current tool only offers revenue LTV, mentally multiply by your contribution margin before making decisions with it.

What about predicted LTV?

Useful in theory, unaudited in practice. The prediction models shipped by analytics tools in this category publish no accuracy or backtest numbers, and independent practitioners report systematic overestimation. If you use predictions, keep predicted and actual side by side until they converge; if you must choose, a labeled real number beats an unlabeled guess.

Do guest checkouts break cohort analysis?

They blur it. Identity is usually keyed on email, so one person using two addresses counts as two customers, which understates your repeat rate; shared family emails do the opposite on a smaller scale. Treat measured repeat rates as a floor. A tool should disclose what share of orders carried a usable identity so you know the base you are standing on.

My repeat rate is very low. Is LTV analysis pointless for me?

The opposite: it delivers a different, equally valuable verdict. A low repeat rate means nearly all customer value arrives with the first order, so first-order profitability (POAS above 1.0) is your entire game, and any "we will make it back on repeat purchases" reasoning is unsupported by your own data. The cohort table is how you prove that to yourself and, sometimes more importantly, to whoever keeps proposing deeper discounts.

Why do my columns say M0 and M1 instead of calendar months?

Because every row's customers started at a different time, the table aligns them by age instead: M0 is each cohort's own first month, M1 the next, and so on. Rows never shift as the calendar moves; their next cells simply fill in. This is what makes rows comparable, and it means one calendar month lives on a diagonal across the table, a handy way to spot the effect of a specific sale or outage.