The PPC Audit That Actually Finds the Wasted Spend

An audit that lists spend, clicks, orders and return on ad spend has described an account rather than examined it. Every one of those figures is available on the first screen, and none of them answers the question somebody is actually asking, which is whether the money is working and where it is not. The useful framing is narrower. Find the spend that produced nothing, establish why it is still running, and fix the reason rather than the symptom. Most accounts carry a surprising amount of it and almost none of it is there deliberately.

Search terms with spend and no orders, over a window long enough to mean something. This is the largest single pocket in most accounts and the easiest to fix, and it accumulates because nobody looks: a broad or automatic campaign keeps finding new terms, and each one is small enough individually to ignore. Terms that convert into the wrong product. A click that buys something is not the same as a click that buys the thing it was advertising, and a campaign taking credit for sales of a cheaper variant is reporting a return it did not earn. Placements, which are reported separately and rarely read. Top of search usually converts best and costs most, and a product where it does not is a product paying a premium for the worst placement it has. Duplicate targeting, where the same term is bought in three campaigns and the account competes with itself. It shows as a rising cost per click that nobody can explain, because the competitor driving it is you.

Return on ad spend at account level, first and worst. It is an average of campaigns that are working and campaigns that are not, and the average moves when the mix moves, so it can improve while the account gets worse. The figure worth reading is per campaign and per term. Attributed sales, because a campaign takes credit for a sale that organic would have made anyway. A brand defence campaign bidding on your own name looks superb and is often buying traffic that was already yours. That does not always mean stopping it, but it means reading it differently. Performance marketing beyond PPC covers that in full. Impressions and click through rate, which describe the advertisement rather than the business. A campaign with a superb click through rate and no orders has an excellent advertisement pointed at a listing that does not convert, which is not an advertising problem: a campaign that spends without converting is almost always the page.

Every figure in this audit comes from a handful of reports, and each answers a different question. Amazon describes them in its reporting guide, and the guide now says to download them from the reports icon in Amazon Ads Agent. For Sponsored Products, five do most of the work, with a sixth for the brand question.

Search terms report. The searches shoppers typed that led to at least one click on an ad. This is where the clicks that bought nothing sit, and Amazon names both uses itself: the customer searches worth adding as keywords, and the under-performing ones worth adding as negatives.

Targeting report. Every keyword, product and category target that received at least one impression, across all campaigns. Sorted by target rather than by campaign, it is where duplicate targeting shows: the same term under three campaigns is the account bidding against itself.

Placement report. Top of search compared with the other placements, campaign by campaign. It answers whether the premium paid for the first row is earning its keep on each product, which is the question a placement bid adjustment depends on.

Advertised product report. Sales and spend by the product being advertised. It shows spend against a product that went out of stock or changed price, which the campaign view hides inside an average.

Purchased product report. For sellers only: the products shoppers bought after clicking an ad that were not the product advertised. A campaign whose sales sit mostly here is selling something other than what it is paid to promote, and this is the report that proves it.

Search term impression share report. Your share of all ad impressions for your search terms. It is the check on a brand defence campaign: a large share of the impressions on your own name, where the listing already ranks first, is the traffic that was already yours.

An account running more than Sponsored Products asks the same questions of its other ad types. Sponsored Brands has its own keyword and keyword placement reports and an attributed purchases report, the products sold and attributed to the campaign whether promoted directly or on its landing page. Display ads have a matched target report: the products, and their detail pages, the ads appeared against that drew at least one click. That is the display version of the search terms report, and it is audited the same way, by the spend that produced clicks and no orders.

None of these reports keeps its history indefinitely. An audit that wants to compare one quarter with the last downloads them every month and keeps the files, rather than counting on going back for them later.

An audit is only as good as the window it reads, and most audits choose that window by accident.

End it a week ago. Each sales column in the advertising reports names the window it counts, seven days after the click for a seller's Sponsored Products and longer for some other ad types, so the most recent days of any report are still collecting the orders their clicks will produce. A term that shows spend and no orders on Monday may have two orders by the following Monday. Stop the window at least one full attribution window before the download date.

Make it long enough for the slow terms. Thirty days is enough for a busy term and too short for most of the long tail, where a term may collect ten clicks a month. Sixty to ninety days lets those terms reach the number of clicks at which no order starts to mean something, the arithmetic in how many clicks a term needs before it is judged.

Keep out the days that are not normal. A window that includes a deal event, a stockout or a price change mixes two accounts in one set of figures. Move the window or read the event separately, and note the days left out so the next audit uses the same rule.

Start with total advertising cost of sales for the account against the same figure a quarter ago, because the direction tells you whether this is a leak or a level. Then break the spend into three: terms that produced orders, terms that produced clicks and no orders, and terms that produced neither. The third group is a targeting problem and is usually harmless in cost. The second is the money. Work down it by spend, and for each one decide whether the term is wrong for the product or the product is wrong for the term, because the fixes are opposite: a negative keyword against the first, a listing change against the second. Then check the structure, because most persistent waste is structural rather than a bad keyword. Duplicate targeting, budgets capping the campaigns that work while the ones that do not run free, and automatic campaigns with no negatives harvesting into nothing. How automatic and manual campaigns work together is the shape that should be there.

Clicks with no orders are the waste everybody looks for. There is a quieter kind: terms that do produce orders, at a cost per order the product cannot carry.

The test is the product's break-even advertising cost of sales, the share of the price left after the cost of goods, the fees, storage and returns, before any advertising. A term spending more than that share of the sales it brings loses money on every order, however good its conversion rate looks.

Worked through. A product sells at $30 and keeps $9 a unit before advertising, so its break-even is 30 percent. A term spent $400 last quarter and brought 30 orders, $900 of sales, an advertising cost of sales of about 44 percent. Those orders made $270 before advertising and cost $400 to buy, so the term lost $130 while sitting in the report as one of the account's converters. The figures are illustrative; the method is not.

The fix is a lower bid on that term, aimed at a cost per order under the $9 a unit can pay, not a negative, because these shoppers do buy. The exception is a term bought at a loss on purpose, for a launch or to hold rank on a product's main search, which belongs in the audit's notes as a decision with an end date. ACoS, TACoS and ROAS sets out break-even in full.

The negatives, because they are reversible, immediate and cost nothing to get slightly wrong. Then the budget allocation, moving money from campaigns buying clicks to campaigns buying orders rather than cutting proportionally, which takes the same percentage from the best performer as from the worst. Bids last, and gently. Bid changes take time to show and it is easy to attribute a seasonal movement to a change made the same week, which is how accounts end up with a bid strategy nobody can explain. One change at a time, with a fortnight between, on any account with normal volume. An audit that produces forty changes in one afternoon produces an account nobody can reason about a month later, and reasoning about it is the whole point. That discipline is most of what Amazon advertising management consists of week to week.

A list of changes made is useless a quarter later, because nobody can tell which change did what. The useful output is a short record the next audit starts from.

The baseline. The window read and the days left out, advertising cost of sales by campaign, advertising spend against all sales, and the spend in each of the three groups: terms with orders, terms with clicks and no orders, and terms with neither.

Each change, with its reason. The date, the campaign, what changed, why in a few words, and the spend it was expected to save. A negative added because a term names a different product is a different kind of decision from a bid cut on a term that lost money, and whoever reads the record next needs to know which was which.

What was left alone, and why. Brand terms kept on purpose, launch campaigns allowed to run at a loss until a set date, terms still short of enough clicks to judge. These are the decisions most often reversed by somebody who did not know they were decisions.

The date to look again. Rarely sooner than a month, with a fortnight between changes.

Monthly for the search terms, because that is where waste accumulates continuously rather than in events. A broad campaign finds new terms every week and each one is small enough to ignore individually, which is exactly how a large number becomes invisible. Quarterly for the structure, because campaign architecture only goes wrong when something is added: a new product, a new campaign type, a launch that was meant to be temporary and is still running. Between those, structure does not drift on its own. Immediately after any of three events. A launch, because launch campaigns are built to spend and are rarely turned down afterwards. A price change, because it moves conversion and therefore moves what every bid is worth. And a stockout, because campaigns kept running against an unavailable product spend against nothing and the spend is invisible in an aggregate report. What is not worth doing is auditing weekly. The data does not settle that fast, and an account audited faster than it can produce a signal ends up being managed on noise, which is a more expensive habit than the waste it was looking for. If any of the numbers this audit asks for are unfamiliar, how Amazon pay per click works sets out what each of them is measuring before you go looking for waste in it.