There is a guide for almost every way to improve a product: lower the bid, rewrite the title, change the main image, negate the weak search terms. There are very few guides for deciding that the improving is finished.
So sellers keep going. Every week there is one more thing to try, and the money already spent feels like a reason to keep spending.
This article gives criteria for that decision and a way to run them; only your own numbers can tell you what to conclude. One thing to settle first: a product that does not work is information you paid for, and the capital it holds is parked, not gone.
01Why is the money I have already spent not part of this decision?
Suppose you have put $6,000 into a product: the purchase order, the freight, four months of ad spend. None of it can be brought back by anything you choose today. Continuing does not recover it; stopping does not destroy it. It is gone in both futures, which makes it useless as evidence: a number identical in every future cannot tell you which future to pick.
Money already spent is called sunk cost, and it is the only kind of cost that never belongs in a decision. What belongs is the next dollar: money you have not spent yet and could spend somewhere else. Sellers answer the first question when they meant to answer the second.
Ask it in the form that has an answer. Not “have I spent too much to stop now?”, which has none. Ask: if I had this money today and no history with this product, would I put it here?
This is not an argument for stopping. Plenty of products deserve the next dollar. It is an argument for deciding with the numbers that apply.
02What do I need to rule out before I decide the product is the problem?
A poor result says nothing about the product until you know the product was tested. Often it was not, and the chain from ad to sale broke earlier.
- Did the ads serve at all? Check impressions in Campaign Manager. Almost none means the bid was too low or the campaign was ineligible, and nothing was tested.
- Did clicks arrive? Impressions with very few clicks is a question about what the shopper sees before clicking: main image, price, star rating, title.
- Did the listing get a fair chance? A main image that does not show what the thing is, or a price well above the visible alternatives, gives a listing result, not a product result.
- Did it have a rating? A listing with no reviews converts worse than the same listing with twenty, and that gap is not the product.
- Was it in stock the whole time? Any period containing a stockout is not evidence.
Do not use unit session percentage from your Business Reports as the conversion rate behind any of this. Its denominator is all sessions, organic visits included, and ad clicks usually convert worse, so the number reads high and you conclude the product converts fine when your ad traffic does not. Use the ad conversion rate: ad-attributed orders divided by clicks, from Campaign Manager, ad traffic only.
03What are the criteria for stopping?
Four criteria. They are not a score and they do not all have to be true. Any one of them, measured over a fair period, moves the decision from optimizing to allocating.
| Criterion | How to measure it | Where the numbers live |
|---|---|---|
| Contribution per unit is at or near zero before any advertising | Selling price, minus the fees actually deducted, minus landed cost per unit, minus a returns allowance | Fees: the transaction detail for a settled order, in the Payments section of Seller Central. Landed cost: your own supplier, freight, customs and inland transport invoices. Return rate: your FBA customer returns report |
| Ad conversion rate stays far below the rate this product needs, across enough clicks | Ad-attributed orders divided by clicks, against your average CPC divided by contribution per unit | Campaign Manager, ad traffic only, for a date range you choose |
| Total sales have not grown across months of sustained ad spend | Units ordered per month, month against month, alongside ad spend for the same months | Units: Business Reports, the detail page sales and traffic report by child ASIN, all traffic. Ad spend: Campaign Manager, ad traffic only |
| The capital in the inventory has a better use | The next-dollar comparison in the worked example below | Your own figures |
No Amazon report tells you what ad conversion rate your category supports, and any figure from outside your account belongs to somebody else’s product. Compare instead against the rate your own product requires: average CPC divided by contribution per unit. That is the rate at which an ad-driven sale breaks even. Below it, each sale costs you money.
Units ordered covers all traffic; ad spend covers ad traffic only. Compare each with its own earlier self, month against month, and never subtract one from the other: they are counted on different dates, so the difference would not be a clean organic figure.
04How do I compare spending on this product against spending on the next one?
Every figure is illustrative and every input is on the page, so you can substitute your own. The product is imported, so landed cost includes the unit price, ocean freight, customs and inland transport.
- Selling price$24.99
- Referral fee deducted on a settled order$3.75
- FBA fulfillment fee deducted on the same order$5.10
- Landed cost per unit, imported goods$8.40
- Contribution before returns and advertising (24.99 − 3.75 − 5.10 − 8.40)$7.74
- Share of orders refunded10%
- Not recovered on a refunded order (5.10 + 8.40)$13.50
- Returns allowance per order sold (10% × 13.50)$1.35
- Contribution per unit before advertising (7.74 − 1.35)$6.39
That $6.39 is the whole budget advertising has to fit inside.
- Average CPC$0.90
- Ad conversion rate (ad-attributed orders ÷ clicks)2.5%
- Ad cost per ad-attributed order (0.90 ÷ 0.025)$36.00
- Result on each ad-driven sale (6.39 − 36.00)−$29.61
- Ad conversion rate needed to break even (0.90 ÷ 6.39)14.1%
Is 2.5% a real result, or too few clicks to tell? Over 400 clicks it is 10 orders, where the required 14.1% would have produced about 56. That gap is too large to be chance.
Now the same money in two places. Take $4,000 you have not spent yet, and three months.
- Money not yet spent$4,000.00
- Ad-attributed orders that buys (4,000 ÷ 36.00, rounded down to whole orders)111
- Ad spend on those 111 orders (111 × 36.00)$3,996.00
- Contribution from those 111 orders (111 × 6.39)$709.29
- Units sold per month with no advertising running8
- Units over three months (8 × 3)24
- Contribution from those 24 units (24 × 6.39)$153.36
Keep advertising and you get the unaided sales as well, so contribution is $709.29 + $153.36 = $862.65 against $3,996.00 of ad spend.
- Cash after three months if you keep advertising (4,000.00 − 3,996.00 + 862.65)$866.65
- Cash after three months if you stop advertising (4,000.00 + 153.36)$4,153.36
- Difference$3,286.71
What that $4,000 earns in the next product is unknown. But it is available, and it is not being spent at a known $29.61 loss per sale.
The 8 units a month is not a guess. It is units ordered during a month when the advertising was switched off, the one period in which every sale is unaided by definition.
05Do I have to choose between continuing exactly as I am and stopping completely?
No, and the two-way framing is what makes the decision feel heavy. There are four positions in between, and all four are ordinary.
- Reduce advertising to what the product can carry. Keep only the targets whose ad cost per order is below your contribution per unit, below $6.39 in the example. Lower bids rather than budgets: a lower bid changes the price you pay, a lower budget only changes how early you run out.
- Sell through the remaining stock with no advertising. The listing keeps whatever organic ranking it has, working capital returns as units sell, and no new money goes in.
- Lower the price to clear the inventory faster. You give up contribution per unit in exchange for the cash arriving sooner, a fair purchase when that cash has somewhere better to be.
- Keep the listing alive at low cost. Stop advertising, hold a small quantity, and leave it as a listing you can return to. Amazon charges more for inventory held a long time, so a large quantity that is not moving is not free to keep.
Before lowering the price, recompute contribution per unit at the new price. The referral fee is a percentage of the selling price, so it falls when the price falls. The FBA fulfillment fee is based on size and weight, so it does not. Cut deep enough and each sale costs you money rather than returning some, which can still be right if the aim is turning stock into cash, as long as you know which of the two you are doing.
06Can I make this decision while the product is out of stock?
You can, but you would be reading data the stockout created rather than data the market gave you. When a listing goes unbuyable, sales stop and ranking decays for the keywords it held, so every recent window shows a product that collapsed. None of that describes demand, and after a restock conversion and ranking begin from a worse position, so the period afterwards is not a clean comparison either.
A stockout quietly changes the question. Out of stock, you are not deciding whether to keep optimizing; you are deciding whether to place another purchase order, a larger commitment, at the moment your data is least usable.
Run the criteria on the last period in which the listing was buyable every day and had its current images, price and rating. If no such period exists, the product has not been tested yet, which is a different answer from the one you were about to give.
Frequently asked
How long is long enough before I run these criteria?
Long enough is defined by clicks and by stability, not by weeks: enough clicks for the ad conversion rate to mean something, over a period in which price, main image, rating and stock position did not change. Four settled weeks tell you more than four months of constant changes.
My product sells, but advertising takes all the margin. Is that the same decision?
It is a different one, and it usually has a better answer. If the product converts when shoppers find it and the problem is only the price of the clicks, the first three options above apply. Contribution per unit is the test in both cases; only the conclusion differs.
Does stopping mean I chose a bad product?
It means this product, at this price, with this listing, in this market, did not return the next dollar better than the alternative. That is narrower than it feels, and the useful part is what you learned about demand, pricing and cost.
Final thoughts
Two things can still go wrong if you follow all of this exactly.
The first is measuring during a period that was not representative: a seasonal low, a competitor running a deep discount, a category-wide slowdown. Compare the same months across years where you have them, and be suspicious of a decision resting on one unusual month.
The second costs more. You can move the capital without moving the lesson. If a product did not work because of a demand estimate made before the purchase order, that estimate travels with the capital to the next product. The exit decision is worth much less than the research that follows it.
Run the criteria on a fair period, compute contribution per unit from your own fees and costs, compare the next dollar against the alternative, and pick one of the four positions rather than one of the two extremes. That is all the decision is.