Field notes from the founders

How to calculate LTV from cohort data (and predict it after three months)

November 15, 2026
At most D2C brands, lifetime value is a sentence. "About four orders, with a $90 AOV and a 30% contribution margin." Everyone quotes it, it multiplies out to a comfortable $108, and it exists nowhere: no system holds it, nobody remembers where it came from, and nobody has checked it since.A step up from that, finance or the founder keeps a spreadsheet called something like LTV_final_V3_copy_copy.xlsx. It gets updated quarterly at best, using a rough version of the textbook formula and an orders-per-customer figure that's either stale or estimated. Plenty of businesses sit here. It just about serves for board reporting and budget setting. Operationally it's as much use as a chocolate teapot.Calculating LTV is easy if you're prepared to wait 12 months, report the actuals and miss the boat. Using it to run the business continuously is hard. This post covers how to calculate LTV from cohort data properly, and then a simple way to predict a cohort's 12-month LTV when it's three months old, which is the step that turns LTV from a board-pack number into something you can act on.

Why LTV is hard to use

Your business keeps changing. The customers you acquired 12 months ago may look nothing like the ones you're acquiring now. Your channel mix has shifted, your offers have changed, you've launched products and your margins have moved. A 12-month LTV describes a business that no longer exists.Margin per customer isn't linear. Discounting front-loads the pain. A customer who joins on a heavily discounted first order earns you little or nothing on it, a bit more on the second, and full margin after that. So a cohort's cumulative margin climbs slowly towards a steady state, the undiscounted margin, at a pace set by your discount strategy. Subscription brands with an introductory offer show this most clearly. Any shortcut that assumes a flat margin per order gets the shape wrong, and with it the payback.Sample sizes force trade-offs. Small cohorts are noisy. You can have speed (judge cohorts early), granularity (by channel, product or offer) or accuracy, and however much data you have and however sophisticated your method, you'll be trading those three off against each other.

The lifetime revenue trap

Many "LTV" figures are lifetime revenue: everything a customer spends, before you pay for the product, the shipping and the discount that won them. Revenue numbers are popular for the same reason ROAS is popular. They're easy to produce and easy to understand, and they nearly always look great. Both are deeply flawed, because the money that pays back your acquisition cost is what's left after the costs of each order. LTV is contribution margin, and an LTV:CAC ratio built on revenue sets money you never keep against a CAC you paid in full.

How to calculate LTV from cohort data

Build contribution margin at order level, then add it up. Take every order and work out its contribution: revenue after discounts and refunds, minus product cost, fulfilment, packaging, shipping and payment fees. Then aggregate to customers and cohorts. Estimating an average margin in a spreadsheet loses exactly the detail that matters, especially discounts. If you don't have a database, an AI assistant such as Claude can do this for you in Python, from Shopify exports or by calling the Shopify API, and handle margins and discounts properly order by order.Some of it will always be estimated. We've never seen a business where contribution didn't need a few hard-coded figures or a lookup table: $3.90 for shipping, $1 for picking, $2.50 for packaging. These tend to be stable over time and across orders (or within a category), so they're fairly safe and easy to model. The rule is to pull as much as you can from a system and estimate as little as you can.Group customers into cohorts. A cohort is everyone whose first order fell in the same month. For each cohort, add up the contribution margin generated in each month since acquisition and divide the running total by the number of customers you started with, including the ones who never came back. That running figure is the cohort's cumulative contribution per customer, and its value at month 12 is the cohort's 12-month LTV.Pick a 12-month horizon. We recommend 12 months unless your business runs on an annual buying cycle (flowers, where customers come back for the same birthday or Mother's Day, are the classic example). Even then, if retention is that low, you can save yourself the pain and judge acquisition on first-order profit against what you spent. Choosing 24 months makes LTV look better, but it changes nothing about the business, and it makes the prediction harder to build.Divide by CAC for the same cohort to get LTV:CAC. Blended CAC (total marketing spend divided by Shopify's new customers) is the place to start.

A worked example

Take an illustrative subscription brand doing around $10m a year. The numbers are invented to show the mechanics.A full-price box is $60, and product, fulfilment, packaging, shipping and payment fees come to $36, leaving $24 of contribution. New subscribers get 40% off their first box and 20% off their second. So the first box earns nothing, the second earns $12, and every box after that earns $24.Of 100 subscribers acquired in January, 60 take a second box, 48 a third, then 42, 38, 35, 33, 31, 30, 29, 28 and 27 by month 12.Cumulative contribution per customer is $0 after month one, $7.20 after month two and $18.72 after month three. It reaches $46.32 by month six and $89.04 by month 12. Watch the margin climb as the discounts wash out: across everything the cohort has spent, contribution is 20% of revenue at month three, 28% at month six and 33% at month 12, edging towards the undiscounted 40%.With a $60 CAC, the cohort's 12-month LTV:CAC is 1.5x. On lifetime revenue, the same cohort spent $269.40 per customer, an LTV:CAC of 4.5x. Same customers, same spend, and one of those numbers gets you a very misleading board meeting.

Predicting 12-month LTV at three months

The worked example tells you how to calculate LTV. It doesn't tell you how to use it, because by the time January's cohort reaches month 12 the money has been spent and the decisions made.The simplest way to operationalise LTV is a ratio. Look back at your older cohorts and divide each one's 12-month cumulative contribution per customer by its 3-month figure. In our experience that ratio is fairly stable over time, and it holds up through significant changes to the operating model. It's worth validating on your own history before you rely on it, but it usually does hold.In the example, the ratio is $89.04 ÷ $18.72, about 4.75. Suppose last year's January to June cohorts all landed between 4.6 and 4.9. This year's August cohort has just completed its third month at $17.10 of contribution per customer. Multiply by 4.75 and its predicted 12-month LTV is about $81. CAC in August was $65, so the predicted LTV:CAC is 1.25x, down from 1.5x.That's a signal you get in November, nine months before the actuals would have told you. You now have a cohort-level LTV prediction that's three months old, and you can track it month after month to see whether the customers you're buying now are getting better or worse. We've seen this simple method get within about 5% of the eventual 12-month figure at cohort level, measured as mean absolute percentage error.

Projecting young cohorts is the point

A common warning is never to extrapolate a young cohort. We disagree, provided the method is validated on your own history, as the ratio above is. If you don't project, you can't act, and LTV ends up living in board packs and strategy papers that don't drive anything.

What LTV:CAC do you need to scale?

The 3:1 rule of thumb you'll see quoted comes from software. David Skok's widely cited SaaS Metrics 2.0 observes that the best SaaS businesses run above 3, on recurring revenue and software margins. It doesn't translate to a brand shipping physical product.From our experience of D2C, measured on contribution margin at 12 months: below 1x, you won't be able to scale. As you ramp up acquisition, CAC inevitably creeps up, and sometimes explodes, and there's no headroom to absorb it. Around 1.5x is safer. You probably need to sustain about 2x to give yourself enough headroom to push acquisition hard, especially if you need to stay cash-flow positive without venture backing.

Split it by channel

Once blended LTV:CAC is in place, split it by acquisition channel. That's the first real step towards operationalising LTV, because channels are where budget decisions get made.It means attribution, and attribution can't be solved. The platforms disagree on how to share out the credit, and before that they disagree on the data itself: browser pixels against backend orders, gross against net, refunds counted on the original order or on the day they're processed, and cookie consent deciding who each tool can see. Use a sensible default that everyone agrees on, such as Shopify's last non-direct click or GA4's data-driven attribution reconciled to Shopify's new customers, and stick with it. A consistent default that lets you act beats a perfect model you never get.

Where AI fits

Cohort LTV is the entry-level approach. Machine learning can go further: customer-level predictions that arrive faster, update continuously, can be cut any way you like, and let you personalise offers and messages one to one.If anyone tells you that's easy, or that AI makes it easy, be sceptical. LTV is an intrinsically hard problem with many pitfalls and trade-offs, and AI helps with them without making them go away. Fast, accurate customer-level predictions are the holy grail of LTV, and even companies with serious data firepower fail at it, or only ever partly succeed.That's the problem we work on for D2C brands without a data team. A company brain for D2C rebuilds LTV, CAC and contribution margin on one definition, from structured data across every system you run, reads the business LTV:CAC first, and has agents that act on it in the open.If you're below roughly $1m in revenue, your cohorts will usually be too small to separate signal from noise, and a careful version of the calculation above in a spreadsheet is the right place to start.If you'd like to see what your own cohorts say, book a call at signalovernoise.app/book-a-call or drop us an email at hello@signalovernoise.app.