The Real Cost of a Customer on Amazon

Most sellers can tell you their customer acquisition cost to the dollar. Almost none of them can tell you what their next customer will cost. Those are different numbers, and the gap between them is where profit quietly disappears.

The standard calculation is total ad spend divided by total orders. Run that on a healthy account and you get something reassuring, say $18 a customer against a $42 order. ACOS looks acceptable, the account is growing, nobody panics. The problem is that this number is an average across every keyword, placement, and audience you bought, and it therefore tells you nothing about the decision you actually face, which is whether to spend the next $500.

Why blended CAC hides your worst spend

Ad spend does not buy orders at a constant rate. You buy the cheapest conversions first: your brand terms, your exact-match head keywords, the placements where intent is highest. As you increase budget you are forced down the intent curve into broader match types, competitor terms, and display audiences that convert at a fraction of the rate. Cost per incremental order rises the whole way down.

Here is an illustrative campaign, sorted from cheapest to most expensive spend. These numbers are hypothetical, not from a real account.

A “tier” here isn’t a category Amazon assigns — it’s just a $500 slice of your own spend once every search term is sorted from cheapest to most expensive. Here’s what that sorting looks like at the individual query level, continuing until the running total crosses $500:

The first five queries above are Tier 1 in the table below: $500 of spend, 55 orders. The sixth query tips the cumulative total past $500 and becomes the first row of Tier 2 — notice it isn’t dramatically pricier than the query before it, it just happens to be the one that crosses the line. Roll up every query in this hypothetical campaign the same way and you get:

The $18 blended CAC is real. It is also useless. The first $500 of daily spend acquires orders at $9, while the last $500 acquires them at $50, and the average sits comfortably in between, concealing both facts. Every dollar in tier 5 is subsidised by the brand-term traffic in tier 1 that you would have won anyway. Cut tiers 4 and 5 entirely and you lose 22 orders a day and $1,000 a day of spend, which is a trade most sellers should look at hard before dismissing.

Calculating marginal CAC from reports you already have

This does not require a data warehouse. It requires one report and a sort.

1.   Pull the Sponsored Products search term report (or the placement report, if you are diagnosing placement bids) for a stable 30 to 60 day window.

2.   Sort ascending by cost per order, which is spend divided by orders at the row level.

3.   Walk down the sorted rows accumulating spend, and cut a new tier every $500 or $1,000 of cumulative spend. Under $50/day, use $100–$200 tiers instead — at low volume, $500 buckets collapse into one or two tiers and the curve stops telling you anything.

4.   Record cumulative orders at each tier boundary.

5.   Compute the marginal cost of each tier.

The formula is just a first difference:

Marginal CAC = (Spend at tier B − Spend at tier A) ÷ (Orders at tier B − Orders at tier A)

Plot cumulative spend on the x axis and cumulative orders on the y axis and you get a curve that flattens as it climbs. The slope at any point is your marginal CAC at that budget level. That slope, not the average, is the number to compare against value.

Two adjustments matter before you trust the output. First, orders are not customers. Use new-to-brand orders as your acquisition denominator, because a repeat buyer purchasing again is not an acquisition, however the campaign reports it. If new-to-brand is 65% of ad orders, your $50 marginal cost per order in tier 5 is really $77 per new customer. Second, ad-attributed orders inside the attribution window are not the same as incremental orders. Some of these people would have bought anyway. Every CAC figure produced this way is therefore a ceiling on your true incremental CAC, which is why the honest version of this exercise eventually involves geo holdouts or brand-term pauses rather than reports alone.

One more filter before you act on any of this: treat a tier built from fewer than roughly 20 orders as noise. A five-order bucket can move 40% depending on which day you happened to pull the report, and a “cut” call on a noisy tier is a coin flip dressed up as math.

Where Subscribe & Save changes the equation

Now the other side of the ledger. You cannot decide whether $77 is too much to pay without an estimate of what a customer is worth, and this is where most models fall apart, because the repeat rate gets guessed.

Subscribe & Save is the best fix available on Amazon, and it is undervalued precisely because sellers see the discount and stop thinking. Yes, an S&S customer is worth less on the first order. At a 10% discount on a $42 item with 32% contribution margin, you give up $4.20 of a $13.44 contribution, leaving $9.24 per shipment, which is a 31% cut to first-order profit.

What you buy with that is a cadence you can observe instead of a repeat rate you invent. Subscription reporting gives you active subscribers, skips, and cancellations by period, which is enough to compute a monthly cancellation rate directly from your own account rather than borrowing a category benchmark.

From a cancellation rate you get an expected number of shipments:

Expected shipments ≈ 1 ÷ monthly cancellation rate

At 10% monthly churn that is 10 shipments, worth $92.40 in contribution. Treat that as an optimistic upper bound: it assumes a constant hazard rate, and real S&S cohorts churn hardest at the second and third shipment before stabilising, so a young subscriber base will underperform the geometric estimate. Survival analysis on your own cohorts is the rigorous version of this calculation, and it is worth doing once you have twelve months of data.

Compare that to a non-subscriber. At 1.4 lifetime orders and $13.44 of contribution each, they are worth $18.82. The subscriber is worth roughly five times more, which means acquisition cost and retention mechanism cannot be evaluated separately. A customer acquired at a terrible marginal CAC can still be a good trade if they subscribe.

Folding LTV into the same table

The full model needs one more line:

LTV = Average order value × Contribution margin % × Expected number of orders

Run it for both customer types and blend by your actual S&S conversion rate. At 40% of new customers subscribing:

Now put value next to marginal cost, per new customer, tier by tier. Illustrative throughout.

Tiers 4 and 5 lose money on every customer they bring in, and the blended 1.74x ratio is what hides that. Cut them and the account-level ratio moves to 2.45x on $1,500 of daily spend. Cut tier 3 as well and you reach 2.92x on $1,000, at the cost of 46 orders a day. Which of those you choose depends on whether you are optimising for profit or for rank velocity, but you should at least know you are making the choice.

One nuance worth acting on: if you can identify which tiers convert to S&S at above-average rates, and search term level subscription data lets you get close, then those tiers deserve a separate LTV. A tier that converts 70% of buyers to subscriptions carries an LTV near $70, which makes a $64 marginal CAC roughly breakeven rather than clearly negative.

The decision rule

The last dollar you spend should acquire a customer worth at least as much as that dollar. Stop comparing ACOS to a target and start comparing marginal CAC to marginal LTV.

In practice, three thresholds:

•    Marginal LTV:CAC above 3.0x: scale the tier. The commonly cited 3:1 rule of thumb from DTC and SaaS financial modelling is a heuristic about portfolio health, not a law, and it exists mainly to buy a margin of safety against LTV being overestimated.

•    Between 1.0x and 3.0x: hold, and use payback period to break the tie. A 3.5 month payback is fine if you are not financing inventory; it is dangerous if you are.

•    Below 1.0x: cut, unless you can defend the spend as a rank or launch investment with an explicit end date.

Give subscribers a longer leash. Their LTV estimate has lower variance than a non-subscriber's, because a subscription is an observed cadence rather than a hoped-for repeat purchase, and a more certain estimate justifies bidding closer to the theoretical limit even when the point estimate is similar.

How to start this week

•    Pull one 30 day Sponsored Products search term report and build the cumulative spend versus orders curve for your top campaign. This works for any seller with basic ad spend history — no special access required.

•    If you have Amazon Marketing Cloud access, pull new-to-brand share by search term and swap it in as your acquisition denominator, the way we did above. If you don't have AMC (most sellers don't — it requires a minimum ad spend threshold and isn't self-serve for everyone), you won't get this at the query level, full stop. The fallback is coarser but still useful: pull your account-wide repeat-purchase rate from Brand Analytics and apply it as a single blended NTB assumption across all tiers, with a clear caveat that it's an account average, not a per-query measurement, so it won't catch specific broad or misleading terms that skew toward existing customers.

•    If your product is Subscribe & Save eligible, calculate your actual monthly cancellation rate from your own subscription reporting instead of borrowing a category assumption. If it isn't — most non-consumables aren't S&S eligible at all — skip this step entirely and build LTV off repeat orders per customer and average order value instead, which is the right substitute metric for apparel, home goods, electronics, and anything else people don't reorder on a schedule.

•    Where S&S applies, segment your conversion rate by campaign type so you can tell high-LTV traffic from cheap traffic. Where it doesn't, do the equivalent with repeat-purchase rate instead: segment by campaign to see which traffic sources bring back buyers versus one-and-done customers.

•    Set a marginal CAC ceiling per campaign and review it monthly, rather than managing to a single account-wide ACOS target. This holds regardless of what your LTV model is built on.

If you would rather see this run on your own account data than build it yourself, this analysis is something we do routinely. Contact us