Almost every Amazon Ads account has a rule shaped like this: if ACOS is above 45%, cut the bid. It is the first automation anyone writes, it is easy to reason about, and it will spend most of its attention on the targets that matter least.
A ratio has no size
Take two targets. Both sitting at 120% ACOS, both breaching the same threshold, both triggering the same rule with the same urgency:
| Target | Sales | Spend | ACOS |
|---|---|---|---|
| A long-tail query, barely runs | $6.67 | $8.00 | 120% |
| A core keyword on your best ASIN | $666.67 | $800.00 | 120% |
To the threshold these are the same event. To your bank account one of them is a rounding error and the other is the reason the month missed. A rule that fires equally on both will generate mostly noise, and the usual response to noise is to widen the threshold until the rule stops being annoying — at which point it also stops catching the $800.
ACOS tells you the shape of a problem. It never tells you the size of one.
The other failure: the target that is fine
The reverse is worse, because it is silent. A product with a 55% margin can be perfectly profitable at 50% ACOS. A 45% threshold cuts its bid anyway, the target loses position, and the orders it was producing go somewhere else. Nothing in the account reports that as a loss. It shows up as your ACOS improving.
A ratio threshold is a proxy for a question you never actually asked: how much money am I willing to lose on this target, and has it lost that much yet?
Ask the money question instead
Two settings, doing two different jobs. The first says how much spend a sale justifies. The second says how much loss is worth acting on.
The arithmetic
Allowed spend = Sales × Guard Target ACOS
Excess spend = max(0, Spend − Allowed spend)
Act when Excess spend ≥ Loss budget
Run the two targets through it, with a Guard Target ACOS of 50% and a loss budget of $20:
| Target | Allowed | Spent | Excess | Against a $20 budget |
|---|---|---|---|---|
| Long-tail query | $3.34 | $8.00 | $4.66 | Left alone |
| Core keyword | $333.34 | $800.00 | $466.66 | Acted on |
Notice what happened to the first row. It is 140% past its allowance — a catastrophic number as a ratio — and the correct action is still to do nothing, because the entire problem is worth $4.66. That is the judgement a percentage cannot express and the reason experienced operators quietly ignore their own ACOS alerts.
Two knobs, not one
- Guard Target ACOS sets how much loss is allowed per unit of sales. It is the economics of the product: margin, contribution, what you are willing to pay for a sale.
- Loss budget sets how much loss is worth a change. It is the size of the problem, in currency, and it is what stops small breaches on small spend from generating alerts nobody acts on.
Separating them is what lets one rule be strict about economics and relaxed about noise at the same time — something a single ACOS number structurally cannot do, because it is being asked to encode both at once.
Where this still needs a person
This is arithmetic, not a strategy. It sizes the response to the money, and it does not know why the money moved. A target can pass a loss budget because a competitor started bidding, because your listing went out of stock, because a season ended, or because the target was always wrong. Those want four different responses.
The rule's job is to stop the bleeding inside limits you set, on the day it happens, without waiting for you to open the account. Working out which of the four it was is still yours.
What to do with this
- Go and find a target your ACOS rule acted on last month where the total spend was under $20. That is what your threshold is spending its attention on.
- Work out, per product, the ACOS at which it stops contributing. That number — not a round account-wide 45% — is the allowance.
- Decide the smallest loss worth a change. Below it, you want silence, not a smaller threshold.