The idea
A customer is not worth a single number. They are worth what they have spent so far, and that number keeps growing every month they stay. Every customer metric in this chapter is a count taken at a moment in time: how many are new, how much have they spent, what did they cost to find. Forget the clock and you will compare people who have been around for three years with people who arrived last spring.
Start with the simplest piece: who is new? A customer is new in the period of their first booked order, and in no other period. Everything else follows from that one date.
Figure 1
One customer, four orders, one 'new' year
| customer_id | order_date | status |
|---|---|---|
| 125 | 2023-01-01 | completed |
| 125 | 2023-01-08 | completed |
| 125 | 2023-01-15 | completed |
| 125 | 2023-01-26 | completed |
| customer_id | first_order | new_in |
|---|---|---|
| 125 | 2023-01-01 | 2023 |
- Customer 125 has four booked orders in January 2023. We need to decide what is 'new' about them, and in which period.
- A customer's first order is their earliest booked order. Take the minimum
order_datepercustomer_id: here 2023-01-01. - Customer 125 is new in 2023, once, and never new again. The other three orders belong to a returning customer.
Why it exists
A store has to decide how much it can spend to win a customer. That decision needs three numbers: how many customers the spend won, what each one is worth, and how long it takes to earn back the cost. None of them is stored in a table. You build them from orders, and each build has a choice of rule that changes the answer.
Harbor & Pine is a good case because its customers come back: 81% of buyers have ordered more than once. Repeat buying makes lifetime value large, and it makes every mistake large too. A customer's third order is worth as much as their first, but it only exists if you have waited long enough to see it.
How it works under the hood
New and returning customers
Take the minimum order_date per customer among booked orders (completed or refunded). That
is the first order. An order is a new-customer order when it is that first order, and a
returning-customer order otherwise. Counting customers per year then means counting distinct
first-order dates in that year: 574 in 2023, 525 in 2024 and 641 in 2025.
Repeat rate and frequency
The repeat rate is the share of customers with at least two booked orders. Frequency is orders per customer. Both need a denominator and a time limit. "Ever" is easy to compute and easy to misread: 97.9% of the 2023 cohort have a second order, against 58.8% of the 2025 cohort. That does not mean the newer customers are worse. Most of them simply have not had time yet.
Lifetime value
Historic LTV is net revenue per customer to date: the discounted line value minus refunds, divided by the customers in the cohort. It is a fact about the past, with no forecasting in it.
Figure 2
Net revenue per customer so far, by year of first order
- The 2023 cohort has produced $6,209.53 of net revenue per customer, almost six times the 2025 cohort's $1,126.30.
- It looks like 2023 customers were better. But they have had up to three years to order, and an average of 11.6 orders each. The 2025 cohort has had at most one year and has averaged 2.5.
- The gap is mostly age. A fair comparison asks what each cohort had earned after its first 12 months, not what it has earned by today.
The trap is reading it across cohorts. The 2023 cohort's number contains three years of orders and the 2025 cohort's contains one at most, so the first number will always be larger. To compare cohorts, cut every one at the same age, for example the first 12 months after the first order, or compare cohorts only once they have the same number of months behind them.
A predictive LTV answers a different question: what will a new customer be worth? The simple version multiplies average order value, orders per year and expected lifespan, then applies a margin. It rests on assumptions you must write down, such as how long a customer stays.
CAC and payback
CAC (customer acquisition cost) is marketing spend divided by new customers over the same period. The blended version uses all spend and all new customers. Channel-level CAC needs to know which channel found each customer, and the data does not always say: of 641 new customers in 2025, only 47 named Google Ads and 209 left the field blank. Divide the $340,778.33 of paid search spend by those 47 and paid CAC comes out at $7,250.60, which measures who fills in forms, not what customers cost.
Figure 3
From marketing spend to payback
- Marketing spend $510,641.32 in 2025
- New customers 641 first orders in 2025
- CAC spend ÷ new customers
- Margin per customer what a customer earns
- Payback months until margin ≥ CAC
- Add up every dollar of marketing spend in the year: $510,641.32 in 2025.
- Count the customers whose first booked order falls in the same year. Spend and customers are aligned on the same dates: 641 customers.
- Divide once. $510,641.32 / 641 = $796.63 per new customer. This is blended CAC, because it spreads all channels over all new customers.
- The 2024 cohort cost $841.70 per customer and earned $1,020.11 of margin each in its first year.
- Margin passes CAC after 9.9 months: that is the payback. Both sides use margin, never revenue, because the ad has to be paid out of what is left after the goods.
The payback period is the number of months a cohort needs before its cumulative margin reaches its CAC. For the 2024 cohort that is 9.9 months. After that month every extra order is profit toward the next campaign, which is why a short payback is often more useful to a founder than a high ratio: it says how fast the money comes back.
LTV:CAC compares what a customer is worth with what they cost. Payback says when the cost is earned back, which matters because cash goes out first and comes in over months.
Putting the clock into every definition
When you write any of these metrics down, add the clock to it. "Repeat rate" should say within how many days or as of which date. "LTV" should say to date or first 12 months. "New customers" should say by first booked order in the calendar year. Three extra words stop the next person from comparing a three-year-old number with a three-month-old one.
Common misconceptions
"LTV is the average revenue of our active customers"
Active customers are the survivors, the ones who did not leave. Averaging only them leaves out everyone who bought once and never returned, so the number is too high for the people a campaign will actually bring in. LTV belongs to a cohort, counted from its first order, including those who stopped.
"Anyone who ordered this year is a new customer"
A customer ordering in 2025 might have been ordering since 2023. New means first booked order in the period. If you count by order date alone, you count the same person as new again every year, and CAC looks cheaper than it is.
"CAC is what we spent on ads divided by sales"
CAC divides by customers, not by orders or revenue. A customer who orders eight times is still one acquisition. Dividing spend by orders makes every number look cheaper in a store like Harbor & Pine, where customers repeat, and hides how much it really costs to bring someone through the door the first time.
Check yourself
One question. Answering it ticks this lesson off on the map; it never locks a level.
Not quite. The numbers are real but they are measured at different ages. Compare the cohorts at the same age before drawing any conclusion about quality.Try another answer.
Not quite. CAC and LTV are separate numbers. A cheaper cohort does not make its revenue higher, and the 2025 CAC ($796.63) is not far from 2024's ($841.70).Try another answer.
Right. Age is the difference. Compare each cohort's first 12 months and the gap shrinks.
Not quite. A customer is new only in the year of their first booked order. Counting them again would break the cohort, which is why the first order is fixed once.Try another answer.
Lesson done: it is ticked on the map.