The Amazon bestseller trap: how to allocate ad spend for long-term growth
ROAS shows which products generate sales. It doesn’t show which ones create valuable customers. Here’s a practical framework for deciding what to fund, build, monetize, or stop.
Every brand has hero SKUs, usually defined as the products that sell the most.
That’s not the same as the products that build the business.
We saw the difference clearly on a repeat-purchase account we manage. A small trial pack produced roughly $95 in six-month customer value for each new customer it acquired. A bulk multipack a few rows above it on the bestseller list produced about $22 - and 86 percent of its first-time buyers never returned.
Both products looked strong in the sales report. But only one was a good place to invest the next acquisition dollar.
We’re all guilty of leaning too hard on ad metrics. Conversion rate and ROAS tell you whether the ad produced a transaction. They answer “did it sell?” But what they can’t answer: “did we acquire a customer?”
Use ROAS to measure the transaction. Use customer economics to decide which transactions deserve more investment.
Some best sellers acquire customers, others don’t
The problem is: at first glance, you can’t tell the difference between the two.
In our example case, ad performance had been slipping. A six-month view made the decline look broad and mild, the kind of trend that results in the team adjusting bids, increasing spend, and waiting for the next promotional event.
The picture changed when we widened the window to 15 months.
Total orders were down 15 percent. New customers were down by roughly a third. Repeat revenue had barely moved - down only about a point and a half.
At the same time, CAC (customer acquisition cost) had increased by more than half, while first order value didn’t budge. The first sale had gone underwater.
It turned out the business had a damaged acquisition engine hidden by a healthy repeat base.
That distinction matters because the repeat engine and the acquisition engine can move in opposite directions. A resilient group of existing customers can keep total revenue looking respectable while the brand quietly stops replenishing new ones.
Once we separated the two, we asked the next question: which products are bringing new customers into the brand, and which are just generating orders?
Two numbers tell you what a SKU is actually doing
For each SKU that receives ad or promotional support, two measures tell the story.
1. First-order contribution
Did the initial order make or lose money after the costs required to produce it?
At minimum, include:
- Net revenue after discounts
- COGS
- Amazon referral and fulfillment fees
- Returns or expected returns
- Advertising spend
- Promotional funding (the deal fees and co-op cost of running the event - the price cut itself is already netted out of revenue)
First-order contribution = net revenue - COGS - Amazon and fulfillment fees - returns - ad and promotional cost
This calculation is more demanding than ROAS. A product can have a solid ROAS and still lose money on every new customer it acquires.
2. Customer value after the first purchase
What happened after the initial order?
What’s the contribution from later purchases over a sensible timeframe (category dependent)? In frequently replenished consumables, that may be 60 or 90 days. In apparel or another occasional-purchase category, it may be six or 12 months and should probably include adjacent purchases, not only a reorder of the same SKU.
The important point: measure total sales from the customer the SKU acquired, not just repeat revenue of that same SKU.
On the example account above, the small trial pack worked as an entry point. It lowered the commitment required for a first purchase, then converted a meaningful share of those buyers into full-price repeats. The bulk multipack did a different job. It moved volume, but most of the buyers it acquired never came back.
Same catalog, same customer profile even. But very different economics.
The SKU Investment Matrix: two numbers, four actions
Based on first order and customer value metrics alone, we end up with a pretty useful frame for evaluating SKU value.
Cross first-order contribution with customer value after the initial purchase and four actions emerge.
SKUs to Fund
Protect these budgets. These products deserve priority in acquisition campaigns and promotional planning, and watch inventory closely - going out of stock on one of them damages more than immediate sales.
They’re the rare products that support current contribution and future growth at the same time.
The trial pack in our example belonged here.
SKUs to Build
Treat these as investments. A Build SKU acquires valuable customers before the initial order pays for itself. A new product may need reviews and velocity. A relaunch may require temporary over-investment. An entry SKU may be designed to recover acquisition cost through later purchases rather than on order one.
This is dangerous territory. It’s too easy to invest longer than you should. To avoid that, it helps to track:
- Why does this SKU deserve investment?
- How much first-order loss will we accept?
- What return justifies the initial investment?
- Critically - when do we make the keep-or-kill decision on the spend?
We had a relaunch SKU that looked inefficient on a standard ad report. We continued funding it to build reviews and sales velocity, but we treated it as a named investment with a written thesis and a sunset date. That kept a deliberate bet from becoming an open-ended excuse.
SKUs to Monetize
Scale these while the incremental contribution remains attractive - and don’t call it growth. These products are profitable on the first transaction but don’t create much additional customer value. Use them to generate profit. Just don’t mistake that profit for customer acquisition, and don’t allow them to absorb so much budget that stronger customer-building products get starved.
SKUs to Stop
Reduce the spend, remove the product from acquisition campaigns, OR fix the underlying offer before putting more money behind it. These products lose money on the first order and don’t build much value thereafter. There’s no second act coming to rescue the economics.
A SKU can survive in this position for years because conversion reports keep presenting each sale as a success. An understanding of the customer economics makes the cost visible.
One more bucket: SKUs to Harvest
Technically, there’s one more group to define to make the system complete: SKUs to Harvest. These are products purchased almost entirely by customers the brand already owns - think large stock-up formats and subscription favorites. On our example account, the largest of these sold for more than $200 per unit, with ad spend at roughly 0 to 1 percent of sales. Beautiful efficiency. Almost all of it existing, loyal customers restocking.
These products are valuable. They just aren’t acquisition products, so they don’t belong in the matrix. Keep them available, protect the subscription and repeat business, support them organically - and stop paying to reacquire demand you already own.
What about deals? Do they build any long-term customer value?
We were skeptical, so we looked.
We compared customers whose first-ever order with the brand occurred during a deal window with a matched group whose first purchase occurred at full price during an ordinary period. Across several thousand first-time buyers, we were surprised to see that the deal-acquired customers repeated at roughly the same rate as the full-price group - about 17 percent versus 18 percent. When they returned, they purchased at full price.
To be fair, it probably depends on the vertical and brand as well.
The deal also produced close to twice the normal volume of new customers. And the next event re-activated the earlier cohort on top of the new one. Deal cohorts compound - a finding big enough that we’ll cover it in its own piece later.
At the SKU level, we were able to see that roughly half the deal orders on the trial-size entry product came from genuinely new customers. The bulk multipack moved substantial volume during the same promotion too, but almost none of it was net-new. Existing customers were stocking up at a discount. Both readings of the multipack point the same way - the first-time buyers it found rarely stayed, and during deals most of its volume wasn’t acquisition at all.
So the right rule isn’t “deals work” or “deals waste cash.” It’s:
Treat a deal as an acquisition channel and measure it like one. Don’t run deals that chase unprofitable volume.
(Unless you have inventory you need to liquidate. There’s a caveat to everything.)
Score the event on:
- Net-new customers acquired
- First-order contribution after the discount and media cost
- Contribution from later purchases
- Cannibalization of full-price or organic demand
Gross units alone will call a lot of subsidized repeat demand a success.
What to do when SKUs are too new and customer value is TBD?
Customer value takes time to develop. And, that creates a real operating problem: the outcome you care about may not be visible for 60 or 90 days (consumables), or longer (almost everything else). So, what to do?
There are a few options that’ve worked for us. If you believe the new SKUs will behave similarly to more established ones, then model after them.
Use these more mature historical cohorts as the baseline for the matrix. Then monitor earlier signals - 30-day repeat, new-customer mix, first-order contribution, subscription enrollment, or another category-relevant behavior - for indications of how the new products should be bucketed and managed.
DO NOT rebuild the strategy every time yesterday’s ROAS moves. At the same time, don’t ignore developing evidence either. It’s a balance - find the goldilocks point: don’t be too rash, don’t wait too long to make a call either.
Use educated guesswork to make deliberate, balanced decisions.
Does this work outside consumables?
This analysis is easiest in consumables, where the next purchase can happen quickly and often involves the same product. Subscription and repeat revenue is magic.
But the same basic math also works in apparel, beauty devices, outdoor products, accessories, and mixed catalogs. The return behavior just looks a little different.
On an outdoor-apparel account we manage, most products are occasional purchases. One line of lightweight layering pieces repeats much more strongly because satisfied customers return for new colorways. Surprisingly often. The relevant signal isn’t monthly replenishment. It’s another economically meaningful purchase over a longer period.
For durables, low near-term repeat may be normal. Don’t punish a mattress, appliance, or piece of furniture for failing to behave like a supplement. Segment the catalog by purchase pattern before scoring it.
The question isn’t always, “Did the customer buy the same SKU again?”
It’s, “Did this first purchase create a valuable customer relationship?”
Where to start
Take the ten SKUs getting the most ad and promotional investment. Rank them by ROAS, then re-rank by first-order contribution and customer value. If the order holds, your allocation may be right.
On the accounts we’ve run this against, it rarely holds.
FAQ
Is ROAS a good metric for Amazon advertising?
ROAS is good at the job it was built for - measuring whether an ad produced a transaction. What it can’t tell you is whether that transaction acquired a valuable customer, and that’s the question that determines where the next dollar should go. Use ROAS to manage campaigns day to day; use first-order contribution and post-purchase customer value to decide which products deserve the budget at all.
How should a brand allocate Amazon ad spend across its catalog?
Score every SKU receiving meaningful ad or promotional support on two numbers: whether the first order makes money after all costs, and how much contribution the acquired customer creates afterward. Crossing the two assigns each SKU one of four actions - Fund, Build, Monetize, or Stop. And keep Harvest SKUs - products bought overwhelmingly by existing customers - out of the acquisition comparison entirely; don’t pay to reacquire demand you already own.
Do Amazon deals attract low-quality customers?
Not automatically - measure it rather than assume it. On the account in this piece, customers acquired during a deal window repeated at roughly the same rate as full-price first-time buyers (about 17 versus 18 percent), and they came back at full price. The real risk isn’t customer quality; it’s counting subsidized stock-ups by existing customers as acquisition.
What is first-order contribution?
It’s the profit or loss on a customer’s first order after every cost required to produce it: net revenue minus COGS, Amazon referral and fulfillment fees, expected returns, and the ad and promotional spend behind the order. It’s a more demanding test than ROAS - a product can post a respectable ROAS and still lose money on every new customer it acquires.
Want this run against your catalog? Let’s talk.