Glossary
Know what kind of word you're reading.
A metric is a number. A report or data source is where numbers come from. Our own terms are ways of reading them. Mixing the three up is how two people end up arguing past each other, so every term on this site sits in exactly one of these groups.
01 · METRICS
Numbers
ACOS, CTR, conversion rate and the rest. Each is a ratio or a count, worked out from one report. A metric answers “how much”, never “why”.
Jump to metrics → 02 · REPORTS AND SOURCESWhere numbers come from
The ad reports, the Business Report, Search Query Performance, Amazon Marketing Cloud. Each counts a different group of people over a different window.
Jump to sources → 03 · OUR OWN TERMSWays of reading
Shadow, the mountain, the ticket, intent groups. These are Finite Samples terms, not Amazon's. They are models, and some are still hypotheses.
Jump to our terms →
The one rule this page exists for. Compare a metric with a metric, a source with a source. “ACOS versus Amazon Marketing Cloud” is not a comparison: one is a number, the other is a place you can go to compute numbers. And before you compare two numbers, check they come from the same source, count the same thing and cover the same days.
01 · Metrics
Metrics: numbers you read
Each metric below names the report it is usually taken from. The same name can mean different things in different reports; where that happens, it is spelled out.
| Metric | What it is, and where it comes from | What it can and can't tell you |
|---|---|---|
| Impressions | How many times something was shown.Ad reports count impressions of your ads. Search Query Performance counts impressions on search results pages, organic and ads together. The two are different counts with the same name. | Can: how often you were in front of shoppers. Can't: how many different people saw you, or whether they noticed. |
| Clicks and CTR | CTR (click-through rate) is clicks divided by impressions.Ad reports for your ads; Search Query Performance for a search query as a whole. | Can: how well the listing, as shown in that spot, draws a click. Can't: on its own, separate the effect of position from the effect of the listing. |
| CPC (cost per click) | Ad spend divided by clicks. The “actual CPC” is what you were charged, as opposed to your bid.Ad reports. | Can: what a click costs you, and how close that sits to your bid. Can't: what a click is worth. |
| Conversion rate | Orders, or units, divided by clicks or visits. The name hides two different ratios.In ad reports: attributed orders ÷ clicks. In the Business Report: units ÷ sessions (“unit session percentage”). In Search Query Performance: purchases ÷ clicks for one search query. | Can: how readily the people who arrived bought. Can't: tell you whether the product got better or the mix of people arriving changed. Never compare the ad-report version with the Business Report version. |
| Add-to-cart rate | Cart adds divided by clicks.Search Query Performance, per search query. | Can: where between the click and the purchase shoppers drop out. Can't: tell you why. |
| Ad spend | What your ads cost over a period.Ad reports, by campaign, ad group, target, product or placement. | Can: the cost side of every ad ratio. Can't: anything about results on its own. |
| Attributed sales and attributed orders | Sales and orders that Amazon credits to an ad because the shopper interacted with it within a set window before buying (the attribution window, which depends on the ad type).Ad reports. | Can: which purchases followed an ad touch. Can't: which purchases happened because of the ad. Some would have happened anyway. |
| ACOS | Advertising cost of sales: ad spend divided by attributed sales.Ad reports. Both parts come from the same report. | Can: what each credited sale cost in ads. Can't: whether those sales were new, whether they would have happened without the ad, or whether the account is growing. |
| ROAS | Return on ad spend: attributed sales divided by ad spend. The same information as ACOS, upside down.Ad reports. | Same limits as ACOS. “Return” here means credited sales, not profit and not incremental sales. |
| TACoS | Total advertising cost of sales: ad spend divided by total sales, ad-attributed and not. Usually computed by the seller rather than read off one screen.Two sources: spend from the ad reports, total sales from the Business Report. | Can: how heavy ad spend is relative to the whole business. Can't: prove ads caused growth. Promotions, seasons, prices and competitors move the denominator too. |
| Cost per order | Ad spend divided by attributed orders.Ad reports, including the placement report. | Can: what a credited order cost. Can't: whether the buyer was new or one of your existing customers. |
| New-to-brand share | The share of attributed purchases made by people who had not bought from your brand within a lookback period. “New-to-brand” is Amazon's own label, with Amazon's own lookback.Reported by Amazon for some ad types; in Amazon Marketing Cloud you can compute your own version with a lookback you choose. | Can: whether an ad reaches people outside your existing customers. Can't: be compared directly with a new-customer share built on a different lookback or a different group of buyers. |
| Cost per new customer | Ad spend divided by the number of new customers it reached.Spend from the ad reports; the new-customer count from new-to-brand figures or from Amazon Marketing Cloud. The count you use decides the number you get. | Can: judge ads whose job is finding new customers. Can't: judge ads whose job is collecting existing demand. |
| Sessions and units | Sessions are visits to your product pages; units are items ordered.Business Report. Recent periods keep being revised for weeks. | Can: the whole business, ad and non-ad together. Can't: be split by keyword. Any “sessions from this keyword” figure is a join of two different grains. |
| Impression, click, cart-add and purchase share | Your brand's share of everything that happened on one search query: impressions, clicks, cart adds, purchases.Search Query Performance. | Can: where down the funnel you fall behind the market. Can't: separate your ads from your organic results, or tell new shoppers from returning ones. |
| Organic rank | Where your product appears in the unpaid search results for a term.Not in Amazon's standard reports; it has to be tracked separately. | Can: where you stand for a term on the days it was checked. Can't: be read week to week without accounting for its normal noise. |
| Suggested bid | Amazon's recommended bid for a target. Amazon's help pages say it is calculated from the winning bids of similar ads over recent days.The advertising console and the Ads API. | Can: a reference point Amazon computes for your ad group. Can't: be read as a measured market price. It is an estimate. |
02 · Reports and data sources
Reports and data sources: where numbers come from
None of these is a metric. Each one counts a different group of people over a different window, so the same metric name can come out differently in each. Brand Analytics, including Search Query Performance, is open only to brand owners enrolled in Brand Registry.
| Source | What it is, and what it counts | What it can and can't tell you |
|---|---|---|
| Sponsored ads reports | The advertising reports, downloaded from the console or the Ads API. The ones used most here: the campaign report, the search term report (what shoppers actually searched before seeing or clicking your ad), the advertised product report and the placement report (top of search, rest of search, product pages).Counts only shoppers who met your ads. Days follow the advertiser account's time zone, and recent days can still change as attributed orders come in. | Can: spend, impressions, clicks, attributed orders and sales, at fine grain. Can't: see shoppers who never met your ads, or say which sales the ads caused. |
| Business Report | Seller Central's report of traffic and sales: sessions, page views, units ordered, ordered product sales, by date and by product.Counts all activity on your product pages, ad and non-ad. Figures keep being revised for weeks. | Can: the whole business, the denominator for TACoS. Can't: split by keyword or by ad versus organic. |
| Brand Analytics | A set of dashboards for brand owners, inside Seller Central. Search Query Performance is one of them; others include the top search terms and repeat purchase behavior views.Aggregated shopper behavior, not individual shoppers. | Can: what the market searches and how your brand fares on it. Can't: be mixed freely between its own dashboards. Shares from one view and from another are counted differently. |
| Search Query Performance (SQP) | A Brand Analytics dashboard: for each search query, total impressions, clicks, cart adds and purchases across the market, and your brand's share of each, by week, month or quarter.Counts activity on search results, organic and ads in the results together. Its purchases are a subset of what the Business Report counts. | Can: your funnel against the market's, query by query. Can't: separate ads from organic, or new shoppers from returning ones. |
| Repeat purchase view | A Brand Analytics view of how many customers bought from your brand more than once.Brand or product level. | Can: how much repeat buying a brand has. Can't: be lined up with a single search term. |
| Amazon Marketing Cloud (AMC) | A data environment, not a report and not a metric. Advertisers write their own queries over event-level, pseudonymous advertising and purchase records, and get back only aggregated results.Privacy thresholds hide groups that are too small, and history goes back only a limited window. | Can: questions the standard reports can't ask, such as new versus returning buyers by keyword, by placement or by sequence of ad contacts. Can't: show small groups, and its outputs can be biased by what the thresholds hide. |
| Ads API | Amazon's programmatic access to the advertising account.The same data as the console, plus things like bid recommendations, pulled automatically. | Can: pull reports and estimates on a schedule and line them up. Can't: give you anything the underlying reports don't contain. |
| Payments and settlement reports | Seller Central's financial records: what was sold, the fees taken, and what was paid out.The finance view, which differs in timing and scope from the sales reports. | Can: profit that adds up to the money that actually arrives. Can't: be compared line for line with the Business Report's sales. |
| Bid and impression estimates | The estimate tables Amazon shows when you set or change a bid.Amazon's projections, not measurements. | Can: the shape Amazon expects between bid and volume for your ad group. Can't: stand in for what actually happened after a change. |
Also useful
Ad settings and testing terms
Not metrics and not sources: the settings that shape the numbers, and the standard testing words used on this site.
| Term | What it means | Why it matters here |
|---|---|---|
| Placement | Where an ad appears: top of search, rest of search, or product pages. A placement bid adjustment raises your bid for one placement by a percentage. | A placement bid adjustment works like a higher bid. Placements also differ in who tends to buy there. |
| Match type and targeting type | How a keyword target is matched to searches (exact, phrase, broad), and how the ad targets more generally (keywords, products, automatic). | They change which shoppers you can reach. Compare results within one targeting type before blaming the placement or the bid. |
| Attribution window | The period after an ad interaction during which a purchase is credited to that ad. It depends on the ad type. | Sources with different windows can be placed side by side but not added up. |
| Incremental | Sales that would not have happened without the ad. | Attribution does not measure it. Only comparisons against a world without the change do. |
| Holdout and control | A holdout switches ads off (fully or partly) to see what is lost. A control is something left unchanged to compare against. | The strongest evidence on whether ads cause sales. |
03 · Our own terms
Our own terms: Finite Samples' ways of reading
These are not Amazon terms. They are the models and labels used on this site and in the course. Some are tested, some are still hypotheses, and the ledger says which.
| Term | What we mean by it | Status and limits |
|---|---|---|
| Shadow | Every report or data source flattens the account onto one plane: the ad reports on one wall, Search Query Performance on the floor, Amazon Marketing Cloud on another wall. Each is a shadow of the same account. A metric like ACOS is a reading taken off one shadow, not a shadow itself. | A framing, not a claim to test. It says: read several sources before you name the shape. |
| Reading in three dimensions | A report gives you two dimensions: the metric and the period. The third is who the number is made of. | See “time as the third axis”. |
| Time as the third axis | Shopping is a journey: shoppers see, consider, compare, buy and come back. Where the shoppers behind a number stand on that journey changes the number, even if no one behaves differently. | ⚠️ Partly. That the mix changes the reading is measured; how much of a swing is mix rather than reaction is not yet separated. |
| Customer mix | The share of shoppers behind a number who are new, recent or long-standing customers, or early or late in their journey. | Needs user-level data, such as Amazon Marketing Cloud, to measure by term. |
| Dashboard vs. map | Metrics such as ACOS and TACoS are dashboard readings: they tell you something moved, after it moved. The map is the structure behind them: who can be matched to you, and where your customers are in their cycle. | A framing. It decides what you look at first. |
| The mountain | Picture shoppers already on a mountain. How high they stand is how ready they are to buy. Whether they can be matched to you, and how high they stand, is on their side. Your bid is on your side: it decides which ring of the mountain your ad reaches, and how often. | ⚠️ Partly; the model as a whole is a hypothesis. In one auction a bid moves no one; repeated contact may warm a shopper, which is not the same as carrying them up. |
| Ticket (being matched) | Whether a given shopper's search can be matched to your product at all: being matched, or eligible to be shown for that search. The keyword is the entry ticket, not the match itself. This is about the shopper and the product, not Amazon's separate use of “eligible” for whether a product can be advertised. | A model of the first stage. How to get more tickets is an open question in the ledger. |
| Ticket, auction, reaction | Three stages that should be read separately: whether you are matched (ticket), whether your ad wins and where it shows (auction), and what shoppers do when they see it (reaction). | A way to locate a problem before fixing it. |
| Intent group | The set of search terms behind one buying motive, read together instead of one keyword at a time. | A reading unit we use with Search Query Performance and the search term report. Groups are drawn by us, not by Amazon. |
| Acquisition, harvest and defense | Roles we give campaigns. Acquisition (also called prospecting) finds new customers; harvest collects demand that already exists, much of it from your own customers; defense holds your own brand terms. | Each role is judged by a different pair of metrics: acquisition by cost per new customer, harvest by cost per order. ACOS is cost per order divided by order value, so it stands in for cost per order only while prices hold steady. |
| Warm and cold | Shorthand for how close a shopper already is to you: warm shoppers know or have bought from you, cold ones don't. | Measured with new-versus-returning data; the exact cut-off is stated where it is used. |
| New, recent and settled customers | Our split of buyers by how long ago they last bought from the brand, with the lookback measured from the account's own repurchase curve instead of assumed. | Not the same as Amazon's “new-to-brand”, an attribution label Amazon assigns using its own brand lookback. Ours uses a lookback window we set. |
| Four labels on every number | Before comparing two numbers, label each one: how it is counted, its scope (which shoppers), its time (which days, which time zone, when pulled) and its causal level (attributed, observed comparison, or tested increment). | Together they are a number's definition: what it counts. Numbers with different labels are not subtracted from each other. |
| Rulers (for incrementality) | A stack of checks, from weakest to strongest, for whether ads are taking credit for sales that would have happened anyway: buyer mix, organic response, touch versus attribution, controls with placebo arms, and, with user-level data, whether ads only claimed buyers you already had. Only a holdout speaks to causation. | An operating method. The note on attributed sales lists all five; only the last needs user-level data. |
| Placebo arm | A campaign or product deliberately left untouched during a test, to show what would have changed anyway. | Borrowed from experiment design; our use. |
| Residual | Total sales minus ad-attributed sales, often called “organic”. We treat it as a ceiling that holds only if the ad side isn't over-counted, not as a measure of organic sales. | A reading rule. |
| One seller, one entry | Different match types on the same keyword don't bid against each other. Across campaigns, when several of yours target the same search, one of them appears to be picked for the auction. Raising one campaign's bid can just move volume between your campaigns. | No self-bidding across match types: Amazon states it for Sponsored Products (targeting guide). How one campaign is picked across campaigns: observed in our own data, not documented; untested. |
| Suggested bid as a score | Our reading of the suggested bid: an estimate about your ad group as it is now, not the keyword's market price. | ⚠️ Partly. Goes beyond Amazon's definition; the reach explanation is untested. |
| Reading trap | Not a claim to judge but a way a report misleads a careful reader, such as a time zone boundary or a revised figure. | Marked — in the ledger. |
| The ledger and its verdicts | The public record of every claim we have judged. ✅ Holds: survives the tests we know how to run. ⚠️ Partly: true under stated conditions, misleading outside them. ❌ Doesn't hold: the evidence points the other way. 🔶 Untested: an open question. — Reading trap: a way the numbers mislead. 💡 Viewpoint: an argument, not tested. It marks premises inside a note and is never a ledger verdict. The same symbols mark each premise inside a note. | Verdicts change only when the evidence does; reversals stay on the record. |
| Experience, Hypothesis, Reader-contributed | Experience: measured on my own accounts. Hypothesis: the framework is formed and still being tested. Reader-contributed: a reader's case, not verified by us. | Every claim carries one of these labels. |
| Disproof condition | The result that would prove a claim wrong, stated in advance. | Every note names one. |
| Finite samples | The idea behind the name: you never get all the data, only a finite set of samples. Connected across sources, a finite set can still show the real shape. | The site's premise, not a claim to test. |
Definitions of Amazon's own reports and metrics follow Amazon's public help pages as we understand them, and Amazon changes them from time to time. If you spot one that is out of date, tell us.