Why your LTV number is wrong (and how to fix it)
Most LTV numbers reported to the board are point estimates on shaky cohorts. A field-tested framework for measuring lifetime value the way an operator can actually defend.
Ask five people in a DTC brand what LTV is, and you will get five answers. The CFO quotes a number from a board deck. The head of growth quotes a different one from the paid dashboard. The CRM lead quotes a third from the ESP. All three are wrong, and all three are being used to justify budget decisions worth millions of euros a year.
This is not a modelling problem. It is a definitional problem. Lifetime value is a claim about the future, made from data about the past, discounted for margin, refunds, and cost to serve. Every one of those inputs is a design decision, and most brands never make the decision explicitly.
The formula on the whiteboard is not LTV
Average order value times purchase frequency times gross margin. This is the formula every marketing textbook uses, and it is the formula on the whiteboard of every DTC brand under €50M in revenue. It is not lifetime value. It is a snapshot of last month's behaviour, projected as if it were permanent.
The formula assumes the customer base is stationary. It is not. New cohorts behave differently from mature ones. Discounted cohorts behave differently from full-price ones. Paid acquisition cohorts behave differently from organic. Averaging across all of them produces a number that is directionally wrong for every segment inside it.
Five failure modes we see in every audit
1. Cohorts too small to be stable
A cohort of 180 customers with a 4% repeat rate has a 90% confidence interval wide enough to drive a truck through. Brands routinely act on quarterly cohort deltas that are inside statistical noise. The rule we use internally: below 500 customers per cohort, do not act on movement of less than 8 percentage points.
2. Discounted orders counted at full margin
If your LTV multiplies AOV by a static gross margin, and 34% of your revenue is discounted, your LTV is overstated by roughly the difference between blended and net margin. On a brand at 60% gross and 34% discount contribution at an average 22% off, that is 4-6 points of margin quietly missing from the projection.
3. Refunds and returns in a separate system
Most brands calculate LTV from Shopify order data and never join to the returns system. Categories with return rates above 12%, apparel, footwear, some beauty, systematically overstate LTV by the return rate. If you cannot show me a customer-level view where returns are already netted out, your number is wrong.
4. Cost to serve treated as fixed
Support tickets, shipping subsidies, packaging, replacement units. These are variable per customer and heavily skewed. The top decile of customers by revenue is often the top decile by cost to serve as well, and net margin on that segment is frequently lower than the median.
5. Attribution collapsed into LTV
The most expensive mistake. Reporting LTV by 'channel' when the channel is only credited for the first order flattens the compounding effect of retention into an acquisition metric. LTV is a customer property, not a channel property. Channels influence which customers you get, not the value they produce.
The framework we use
We measure LTV at three horizons, per cohort, per acquisition source, net of refunds, at contribution margin, with cost to serve subtracted at the customer level. It sounds heavy. It is not. It is the minimum needed to make a decision you can defend.
| Input | Common practice | What we do |
|---|---|---|
| Time horizon | Blended lifetime | 90, 180, 365 days |
| Segmentation | Blended cohort | Cohort × acquisition source |
| Revenue | Gross | Net of refunds and cancellations |
| Margin | Sticker gross margin | Realised contribution margin per order |
| Cost to serve | Ignored or averaged | Attributed per customer |
| Extrapolation | Linear | Curve fitted, plateau modelled |
What changes when you fix it
Two things, always. First, the ranking of acquisition channels changes. In every audit we have run, at least one channel that looked profitable on a first-order basis was actually the worst on 365-day LTV, and one channel that looked expensive was the best. Second, the retention budget becomes defensible. When the finance team can see 90-day, 180-day, and 365-day LTV per cohort, the argument for lifecycle investment stops being a matter of taste.
"You cannot manage what you extrapolate. You can only manage what you observe."— Every operator eventually
How we approach this at Loiale
We rebuild LTV in the warehouse, not in the ESP or the ads platform. The output is a single view, per customer, per cohort, refreshed daily, that finance, growth, and CRM all read from. It takes two to three weeks. It is the single highest-leverage piece of analytical work we do inside a brand, and it is almost always the first.