What counts as wasted spend, and what is just noise

Every audit tool will tell you how much you spent on search terms that produced no sales. It is the easiest number in advertising to calculate and one of the easiest to be wrong about, because most of the terms on that list never had a chance to sell anything in the first place.

The problem with "no sales"

A search term with two clicks and no order has not failed. It has barely been tested. If your store converts one visitor in twenty, a term with two clicks was never likely to convert; listing it as waste and telling someone to negate it is advice with no information in it. Do that across an account and you produce a list of four hundred terms, most of them noise, which nobody reads and nobody acts on.

The useful list is short and every row on it survives being questioned by a client.

The bar we use

Before a search term with no sales is called wasted, it has to have had a fair chance, which we define as two things at once:

• It has cost at least what an average order costs you in this report — total spend divided by total orders.
• It has been clicked at least twice as often as an order usually takes — two divided by the account's conversion rate, floored at 8 clicks.

Both thresholds come out of the account's own numbers, so they move with it. A store converting at 20% has a low bar: five clicks is an order's worth of traffic, so ten clicks without one is meaningful. A store converting at 2% needs a hundred clicks before the same silence means anything.

When the report contains no orders at all, there is no conversion rate to work from, so the bar falls back to 20 clicks. That is a judgement call, not a calculation, and it is deliberately conservative.

Why both conditions, and not either

Using money alone puts every high-bid term on the list after a handful of clicks. Using clicks alone puts cheap, harmless terms on it that have cost almost nothing. Requiring both means a term has to have consumed real budget *and* had real traffic before we will tell you to block it. The list gets shorter. Every remaining row is worth someone's time.

One instruction per search term

A term that never sold gets negated. A term that sells but loses money gets a lower bid. A term that sells well inside a broad or auto campaign gets promoted to an exact keyword. A term that is none of those, but is being bought by more than one of your own campaigns, gets a note about which campaign should keep it.

Each search term appears once, in exactly one of those lists. A report that tells you to negate a term in one table and bid on it in another is not a report, it is a quiz.

Words, not just terms

A single word can run through dozens of searches that each look too small to bother with. Picture "cheap" across thirty searches, none with enough clicks to clear the bar on its own, all of them together costing real money and none of them selling. One negative phrase blocks every one of them, including the ones that have not been searched yet.

So the word list looks at every search at once. Each search contributes its words and two-word phrases, and a word is listed when it appears in at least three searches, none of those searches has sold, and together they have cost at least an average order. A two-word phrase is dropped when one of its words is already on the list, so you are not told to negate "cheap garlic" as well as "cheap". Before adding any of them, skip the ones that describe your product: a word can be dead weight in one account and the product name in another.

Where your own numbers override ours

All of the above decides whether a search term is *dead*. Whether a selling term is *profitable* is a different question, and it needs a number only you have: your break-even ACoS. Give the audit that and it can separate terms that sell at a profit from terms that sell at a loss, which is usually a bigger pot of money than the dead ones. There is more on working that number out in Break-even ACoS, worked out.