Client case study · a premium DTC pet nutrition brand

What a month of advertising is worth, when it pays back, and what it is worth paying to keep the customer

Two questions from the same client, answered from the same order file. First: at a given cost per customer, how long does a month of paid acquisition take to pay for itself? Second: once someone has bought, how much should the Client give away to get them onto a subscription?

Every chart and table below is live. Change an input and the whole page recomputes — payback, the discount grid, the holdout sizing and the test thresholds all move together.

01 · The question

Can paid acquisition at a higher cost per customer pay back, and on what timeline?

The Client was buying new customers cheaply but not many of them. The proposal on the table was to bid much harder: a target of about $120 per new customer, against roughly $27 historically. The client needed an answer it could plan cash around, not a ROAS figure. How many months before a $120 customer has paid for themselves? How deep does the programme go into the red before it turns? Does any of it change if more people come back?

A second question followed from the first. If repeat buying is what makes acquisition pay, what is it worth spending to make more of it happen, and where should that money go?

Read every figure here as a floor, not a plan. The model repeats month one over and over: the same budget, the same cost per customer, the same repeat rate, for two years straight. Real programmes do not stand still. Creative improves, targeting sharpens, cost per customer comes down and budget scales in behind whatever is working. None of that improvement is built in.

02 · Data & method

Built from real orders, with one-time and repeat buyers modelled separately

The source is the Client's full order export from the ecommerce platform: 100K+ orders from ~15K customers across ~9 years of order history. Nothing in the model is a category benchmark or an assumed curve.

Email-matched cohorts

Customers are matched on email, not account ID. 18.2% of orders are guest checkout, and matching on account ID would merge every guest into one "customer" and inflate repeat rates.

Fully matured cohorts only

The repeat pattern comes from customers acquired in 2023 or later who have each had at least 18 months to come back, about a quarter of every customer the Client has ever won. 55.1% of them placed a second order.

Two buyer types, two first orders

One-time buyers spend $114.19 on their only order. Eventual repeat buyers spend $143.72 on their first order, 26% more. Moving the repeat rate therefore moves first-order revenue too.

Retention × basket, split apart

Repeat revenue in each month is the share of repeaters still buying multiplied by what those who stay spend. Each lever can be moved on its own, and the Retention input in section 04 scales only the first.

How long repeat customers keep buying

Customers do not stay forever, and the model does not pretend they do. Anyone who stops buying counts as zero from then on. The decline below is people leaving.

Share of repeat customers still buying, by month after their first order

Recent, fully matured cohorts (2023 onward, each with at least 18 months of history). Scales with the Retention input in section 04.

MonthStill buyingSpent by those still buyingAverage across all repeat customers

People leave; they do not spend less. The share still buying drops 65% between month 1 and month 24, while the ones who stay spend 23% more than they did at the start. Customers who stay get better. Most do not stay.

How long they last. Typically 13.6 months from first order to last (mean 16.7), with the best 10% running past 37 months. 72.7% of repeat customers have not ordered in six months or more.

Stopping at 24 months makes this cautious. The average repeat customer is worth $763 in revenue by month 24, $961 by month 36 and $1,277 by month 60. The model counts none of the $514 that arrives after month 24, so any payback month shown is the latest it could be.

03 · The do-nothing floor

Existing customers carry the business, and they are leaving

Before pricing any new customer, establish what happens if nothing changes. The existing file keeps buying with no advertising at all, but it decays. New customers are what replace it. This view stacks the two, using the acquisition inputs from section 04 and the recency mix from section 05.

Where the business stands (last 12 months)Value
New customers wonabout 80 a month
Average first order$149.57
Existing vs. new revenueexisting customers generate ~14× the monthly revenue of new first orders
Customers who bought in the last month7.1% of the customer file
Customers lapsed 25+ months68.1% of the customer file
—Existing-customer revenue for every $1 from new customers, month 1
—Change in total monthly revenue over 2 years
—Total revenue index in month 24 (month 1 = 100)
—Share of month-24 revenue from advertised customers

Existing customers and new ones, month by month

Indexed: total revenue in month 1 = 100. Volumes are indexed rather than shown in dollars.

Existing customers, no advertising New customers from advertising

The growth figure is not a target. At the default settings it is roughly flat, and that is the finding: every point of growth beyond it has to come from spending more, paying less per customer or getting more people to buy again.

04 · Payback at the default CPA

A $120 customer pays back. The programme takes longer than the customer.

Every month of advertising brings in a cohort of new customers. About 45% buy once and never return; the rest follow the measured repeat curve. Payback is an output here, never an input: it falls out of the fields below.

Out of every 100 new customers, how many ever place a second order. Measured: 55.1%.
Advertising cost per new customer. About $27 historically; $120 is the proposed target.
An illustrative budget. Changes how much cash is tied up, not how quickly one customer pays back.
What someone spends if they never come back.
First-order spend for customers who do come back: 26% more.
100 = as measured; 120 = 20% more repeaters still buying each month.

Profit margin is fixed at 50%, an illustrative margin assumption for what is left of each sales dollar after product cost and fees.

If more or fewer buy again
—Months for one customer to pay back what they cost
—Month the whole programme is in the black
—Profit from the first order
—Deepest point out of pocket
—Profit after advertising, 24 months

When the money comes back

This line bends upward only because repeat orders pile up, not because anything gets cheaper. Any efficiency found later pulls the crossing point left. Treat the month shown as the latest it would arrive.

Running total of profit after advertising

Profit minus spend, accumulated month by month at the illustrative budget. Crossing zero is the point you are square.

Monthly revenue against advertising spend

At the illustrative budget, in dollars per month on one scale. Spend is deliberately flat.

All revenue from advertised cohorts Advertising spend

How long one customer takes to pay for themselves

The grid builds itself around the current settings, so your combination is always in the middle and highlighted. This is one customer on their own; the programme takes longer because you keep paying for new customers while earlier ones are still paying back.

Repeat rate

Budget changes how much cash is tied up before payback arrives. It does not change the month itself.

Month by month

Every row of Spend and New customers is identical: this is the do-nothing case. Only repeat revenue grows, purely because earlier customers come back. The work of the next two years is to make the left-hand columns move.

MonthSpendNew customersRevenue from first orders Returning customersRevenue from repeat ordersAll revenue Revenue per $1 spentProfit after spendRunning total

It shows what revenue looks like at a given cost per customer. It does not prove you can buy customers at that cost. Bidding up to $120 might win the volume shown, or it might win the same customers already arriving at $27 and simply cost more. Only a live test answers that (section 09).

05 · Recency segmentation

How recently someone bought predicts almost everything

To measure the existing file, we took everyone on it two years ago, grouped them by how recently they had bought, and recorded what each group actually spent over the next 24 months. These are observed outcomes, not projections. Inputs are shares of the customer file; outputs are per 1,000 customers.

This view assumes no work to slow the decline: no win-back, no subscription push, no change in email cadence. The three buttons are the way to ask what that work would be worth.

If customers stay longer or leave sooner
—Revenue per 1,000 customers on file, 2 years, no advertising
—Profit on that revenue (50% illustrative margin)
—Monthly revenue index in month 24 (month 1 = 100)
—How far it falls over 2 years
Last orderedShare of fileBought again within 2 yearsRevenue per customer, 2 yearsShare of base revenue

Someone who bought last month goes on to spend $2,057 over the next two years. Someone who last bought two years ago spends $45: a 46-fold difference from one fact about them. And two-thirds of the file has not bought in over two years; those customers are worth about $8 each. It is why win-back aimed at the long-lapsed rarely pays, and why the recently active list is the one to protect.

Monthly revenue from existing customers alone

Indexed to month 1 = 100. Today's file, falling away at the rate its equivalents actually did.

06 · Retention offer design

Pricing a two-tier subscribe-and-save offer, year one only

Once someone has bought, what is it worth paying to get them onto a subscription? This prices a two-part offer: a bigger one-off enrolment discount on the second order, then a smaller standing discount on every order after it. Today the Client offers 5% on subscription and nothing else.

Everything here covers the first twelve months only. Each monthly cohort earns just the months left in the year: the cohort acquired in month 11 contributes one month, not twelve. Customers keep paying after that, but the harder question is whether the offer funds itself inside the year you pay for it.

25% Paid once. The average second order is $157.
5% Applies to everything they buy from then on. 5% today.
+3.0 pts 52.4% buy a second time today. Points, not percent: +3 means 55.4%.
50% 30% get there unprompted today.
Measured inputs and assumptions: change only with reason
Someone who returned only because of a discount is usually worth less than someone who would have returned anyway. Set low to stay cautious.
—Net profit after discounts, first 12 months
—Profit the offer creates
—Cost of the enrolment discount
—Cost of the standing discount
—Repeat-rate lift needed to break even

Where the money goes, first 12 months

LineCustomers a monthFirst 12 months

Enrolment vs. standing discount

Year-one net profit. Rows: the one-time enrolment discount. Columns: the standing discount that follows. Your current setting is highlighted.

Why this section says 52.4% and section 04 says 55.1%. Same customers, different base. 55.1% counts everyone. 52.4% excludes the roughly 8% of the cohort who started on subscription from their first order and come back 86% of the time. The offer can only reach people not yet subscribed, so it uses the lower number.

Year one is a deliberately hard test. A subscriber delivers only 47% of their lifetime value inside the first twelve months, so an offer that merely breaks even here is well ahead over a customer's life. What it should not do is lose money now on the hope that later years repay it.

07 · Holdout test design

Hold 20–25% back, and pick a metric the volume can actually read

Leave a fixed share of new customers out of the offer entirely. With volume this size, a before-and-after comparison cannot separate the offer from seasonality or a change in media mix. The holdout is also the only way to measure marginal convert quality, the assumption every gain figure in section 06 scales with.

25% Recommended 20–25%: enough control to read, without withholding the offer from most customers.
—Customers held out per month
—Months to read the enrolment shift
—Months to read the assumed repeat-rate lift
—Months to read the break-even lift

Sample sizes use a two-proportion test at 95% confidence and 80% power, split by the holdout share. Enrolment is read among second-order customers; repeat rate among all new customers, plus roughly three months for second orders to land (about three-quarters arrive within three months).

08 · Competitor benchmark

What the category gives away, and when

Across 11 competitors and retailer programmes the pattern is the same: one large discount to win the customer, then a small permanent one to keep them. Nobody gives 20% away forever. The Client was the only brand in the set with no welcome offer at all; the 5% subscription rate was the entire discount programme.

BrandCategoryFirst orderOngoing
OllieFresh cooked70%—
Spot & TangoFresh cooked60%—
Viva RawRaw60%—
The Farmer's DogFresh cooked50%—
Darwin'sRaw50%—
Chewy, select brandsRetailer50%10% on order 2, then 5–10%
BJ's RawRaw40%—
Chewy AutoshipRetailer35%5%
Stella & Chewy'sRaw35%5%
Open FarmRaw / freeze-dried25%5%
PrimalRaw20%—
The ClientPremium DTCnone5%

Two of these look bigger than they are. Open Farm's 25% is capped at $30 and Chewy's 35% at $20. The Farmer's Dog has at times given the first box away free; Ollie's 70% comes with a branded gift, Spot & Tango's 60% with $26 of chews. All require signing up to a subscription. Checked October 2026; these offers change.

Almost nobody runs a standing offer on the second order. Chewy's 50%-then-10% is the only one found, and even that looks like a temporary brand-funded promotion rather than policy. That gap is the opportunity, and the reason to test it rather than assume it works.

09 · What a live test must prove

Five things the model assumes and a test has to show

The model tells the client what the numbers look like if its inputs hold. A test is how the client finds out whether they do. In priority order:

  1. The higher CPA buys incremental customers. Run it as a budget-split or geo holdout and judge it on incremental cost per new customer, not platform-reported CPA.
  2. Paid cohorts repeat like the historical file. The payback month in section 04 rests on 55.1% coming back. Customers bought at a higher price may repeat less. Track each new cohort's second-order rate at months 1–3 against the measured curve; that is the earliest warning the payback month is slipping.
  3. The offer moves enrolment against the holdout. This is the primary read because it is the one the volume can support.
  4. Discount-driven converts are worth what the model assumes. Gains are discounted to 50% quality. Compare enrolees' order frequency once the enrolment discount is spent against organic subscribers. If they fall away faster, quality is below 50% and the year-one figures shrink with it.
  5. Year one breaks even at the tested discount levels. If the repeat-rate lift cannot be read in time, keep it as a guardrail (no significant drop), not as the success metric.