Your best ROAS month is probably your worst cohort

The governing number

Everything ladders up to 90-day customer value to CAC. It's a cohort version of CAC payback, and an earlier, observed version of LTV:CAC. LTV is a model built on churn assumptions. This is what customers actually spent.

• A cohort is every customer whose first-ever order falls in the period, deduplicated by customer ID or email.
• Customer value includes the first order and every rebill inside the window. Rebills are the point.
• CAC loads all paid spend onto new customers only. Returning customers carry no acquisition cost.
• 1.0x means the cohort paid back its media by day 90. Under 1.0x, you're lending money to every new customer.

Everything below runs on contribution unless it's labelled gross. The real numbers from that brand are on gross sales, because that's what was being reported at the time.

Trap 01

Banked and projected are different numbers

The month after that 5x read, an all-hands slide showed 90-day CV:CAC at 1.10x and 1.23x. Both were projections, pulled up by two low-CAC months that hadn't reached a rebill yet. The fully matured cohorts behind them sat at 0.51x and 0.61x on gross sales. The blended 12-month payback ratio across all cohorts was 0.77x.

Nothing on the slide was false. It just led with numbers that hadn't happened. The fix is a labelling rule that never bends.

The honest headline when projections look good is "leading indicator on track, banked payback still maturing." It is never "we hit 1.23x."

One more trap sits inside monthly cohorts. Reading month index 3 gives a customer acquired on the 1st about 120 days of value and a customer acquired on the 30th about 90. Index 3 is generous by up to a month. Move to weekly cohorts read at week 13, or footnote every slide with "index 3 reads 90 to 120 days."

Trap 02

Contribution is the real payback

1.0x on gross sales pays back the media and nothing else. Discounts, landed COGS, pick and pack, shipping, gift COGS and returns all come out before a dollar repays CAC.

The illustrative cohort further down reads 1.39x gross and 0.87x on contribution. Run both in parallel for eight weeks, then move the headline to contribution and keep gross as the trend line. Never put a gross number next to a net number without a label, and footnote any series that crosses an analytics tool change. Two tools on the same brand reported the same month's discount rate at 15.4% and 17.5%.

Trap 03

Average CAC hides your last dollar

Cohort CAC is an average. The decision you make every week is about the next dollar, and that's marginal CAC.

An illustrative month. Spend goes from $200,000 to $260,000 and new customers go from 2,105 to 2,210. Average CAC moves from $95 to $118, which looks like a manageable 24% rise. The extra $60,000 bought 105 customers at about $571 each.

That's the shape of the month at the top of this piece. Average CAC rose 31% and the blended number made it look like a pricing problem. It was a ceiling. Meta spend had nearly doubled without a proportional rise in new customers, and blended CAC hid it. Non-brand search CAC in the same period spiked to five to eight times the blended low.

Track marginal CAC by channel weekly. When it crosses the 90-day contribution a new customer delivers, about $87 in the example below, the next dollar is losing money no matter what the average says.

Trap 04

Cheap entry products buy cheap customers

In the same month, new-customer order value fell 27%. Nothing changed in pricing. The mix shifted toward the lowest-priced products in the range, and away from the flagship and the bundles.

CAC on its own celebrated it, because cheaper products convert cheaper. CV:CAC caught it, because value fell faster than cost.

• Tag every cohort by the hero product in the first order and read 90-day CV per entry product.
• Point paid creative at the products whose cohorts pay back, even when the first order costs more to win.
• Watch new-customer AOV weekly beside CAC. A falling AOV with a flat CAC is the early warning.

The science

The levers, in order of what they're worth

An illustrative cohort, run on contribution. $100 CAC. An $85 first order and a $72 rebill after 15% subscribe-and-save. 60% of buyers take subscription, 70% survive to rebill one and 55% to rebill two. Each unit costs $17 landed and each order $9 to pick, pack and ship, plus 3% payment fees and 2% returns.

On gross, this cohort reads 1.39x and looks healthy. On contribution it's 0.87x, so every customer is still about $13 underwater at day 90. The rebill also carries a thinner margin than the first order, 59% against 64%, because subscribe-and-save comes off the price and none of it comes off the cost.

On contribution, price is the biggest lever. A price rise drops straight through while unit cost stays where it is, which is also why deep subscribe-and-save discounts cost more than they look. Retention looks small here because only two rebills fit inside 90 days. It compounds after that, so a retention-heavy brand should read a 180-day ratio beside the 90-day one. A subscriber who makes it through rebill two is far more likely to still be there at month twelve.

The art

The subscription mechanics that actually move it

• First-rebill survival as its own cohort view. Subscribers who complete rebill one, divided by subscribers acquired. Split it by entry product, entry offer and channel. It's the largest revenue leak in most subscription businesses and it's rarely on the dashboard.
• Read offer tests through survival. A 20% first-order discount usually wins on signups. Test it against a milestone gift at rebill two and read the result through first-rebill survival and 90-day CV:CAC. Signup volume is the wrong scoreboard for an offer test.
• Watch discount stacking. A first-order discount on top of subscribe-and-save on top of free shipping can put the first order underwater on contribution before CAC is even counted.
• Make skip and pause easier than cancel. A paused subscriber can be reactivated. A cancelled one has to be reacquired at full CAC.
• Ship on the real usage rhythm. Customers who fall behind on product cancel to catch up. Match cadence to how the product is actually used, and let people change it in one text.
• Brief box two like a launch. The second delivery happens thousands of times a month and it decides survival. Treat it as a creative problem.

The brand

Brand has to show up in the cohort table

Brand earns its place in this model when it shows up in the numbers. The clearest example from that brand was a three-week demand ramp into a major sale period that cut CAC by more than half. Demand moments move the denominator hard.

Always-on brand spend is harder to see. Run it as its own budget line with a geo holdout or a matched-market test, and read the result through CAC and first-rebill survival in the test markets. Full-price customers who join for the product survive rebill one at a higher rate than customers who join for the discount. That's where brand pays back inside 90 days.

The window

Ninety days is a choice

Ninety days fits a brand that funds growth from its own cash. It's one quarter of working capital and roughly one inventory turn. A well-funded brand with strong retention can deliberately run a 6 to 12 month payback and win share while others wait for cash. That's a legitimate strategy, as long as the board is choosing it with banked numbers in front of them.

Set the window to two or three replenishment cycles of the core product, and keep a 90-day read as a leading indicator even when the governing window is longer.

The whole company

One spend decision every week

Every week ends with one spend decision tied to the ratio, logged with the number that justified it.

For channel decisions we keep a separate attributed CAC on 7-day click, no view-through, new customers only, rebills excluded. Platform ROAS includes view-through and returning customers, so it never decides where the next dollar goes.

This only works if the whole company is accountable to it. Paid media owns the denominator. Lifecycle and retention own the rebills. Product owns entry mix and AOV. Operations owns cadence and box two. Finance owns landed cost, which is what makes the contribution number possible. When each team reports its own metric instead, the business can lose money on every customer it buys while every team has a good month.

Building it

Thirty days to stand it up

Any DTC brand with six months of order history and daily spend by platform can run this. Week one connects the order system and every paid platform and loads costs. Week two reconciles one full month of net sales to finance and backfills six to twelve months of cohorts. Week three reads the matured cohorts and sets bands. Week four runs the first weekly read and logs the first decision.

The logic, tool-agnostic:

WITH first_orders AS (
  SELECT customer_id, MIN(order_date) AS first_date
  FROM orders
  WHERE is_test = FALSE AND channel = 'dtc'
  GROUP BY customer_id
),
cohorts AS (
  SELECT customer_id, first_date,
         DATE_TRUNC('week', first_date) AS cohort_week
  FROM first_orders
),
cv AS (
  SELECT c.cohort_week, c.customer_id,
         SUM(o.gross_sales) AS cv_90
  FROM cohorts c
  JOIN orders o
    ON o.customer_id = c.customer_id
   AND o.order_date <= c.first_date + INTERVAL '90 days'
  GROUP BY 1, 2
),
spend AS (
  SELECT DATE_TRUNC('week', spend_date) AS cohort_week,
         SUM(spend) AS paid_spend
  FROM ad_spend
  GROUP BY 1
)
SELECT cv.cohort_week,
       COUNT(*)                                  AS new_customers,
       s.paid_spend / COUNT(*)                   AS cohort_cac,
       AVG(cv.cv_90)                             AS cv_90,
       AVG(cv.cv_90) / (s.paid_spend / COUNT(*)) AS cv_cac_90,
       CASE WHEN CURRENT_DATE >= cv.cohort_week + INTERVAL '97 days'
            THEN 'Banked' ELSE 'Projected' END   AS status
FROM cv
JOIN spend s USING (cohort_week)
GROUP BY cv.cohort_week, s.paid_spend
ORDER BY cv.cohort_week;

Multiply cv_90 by your contribution rate for the contribution view, and add entry product and first-order channel to the cohort key for the splits.

FAQ

Questions founders ask us

What is a good CAC payback period for a subscription brand?

For a brand funding growth from its own cash, pay back acquisition cost on contribution margin within 90 days. That's one quarter of working capital and two to three replenishment cycles on a 30-day product. Well-funded brands with strong retention can run 6 to 12 month payback on purpose, as long as the board decides with banked cohorts in front of them.

What's the difference between 90-day CV:CAC and LTV:CAC?

LTV:CAC is usually a model built on churn and margin assumptions. 90-day CV:CAC is observed. It's what a cohort of new customers actually spent in their first 90 days, divided by what it cost to win them, and it arrives nine months before a 12-month LTV read.

Should CAC payback be measured on gross revenue or contribution margin?

Contribution margin. 1.0x on gross sales pays back the media and nothing else. At a 60% contribution rate, a 1.0x gross cohort is 0.6x on contribution.

What is marginal CAC and how do you calculate it?

Change in paid spend divided by change in new customers. It tells you what the next dollar buys, which the average never will. Track it weekly by channel.

Why is ROAS misleading for subscription brands?

Platform ROAS counts the first order, includes view-through and returning customers, and is graded by the platform selling you the ads. In subscription, the first order is usually the one that loses money. The value is in the rebills, and ROAS never sees them.

More from Magnetic: How we name brands · Services