Part I · Chapter 3

Core Advertising Metrics

Status: reviewed · Prerequisites: What AdTech Is and Why It Exists, AdTech Ecosystem Participants

Core Advertising Metrics

The Core Idea: a metric is a contract, not a formula

An advertising metric is not a formula but a contract about how it is counted: an agreement on which event is counted, at which point, after which filters, on which denominator, and in whose money. The arithmetic here is simple — division and multiplication by a thousand — but the semantics are hard: impression1 and click2 mean different events in different systems, and one metric name hides different contracts. That is why one campaign3 produces three simultaneously truthful reports — for the advertiser4, the publisher5, and the platform — with different numbers. The chapter follows the order of the flow: from the appearance of an ad slot to impression, click, and outcome, then the price of an event, money and return, user value, and monetization — and brings it all together in one end-to-end example.

Why a number without a definition says nothing

An advertising campaign passes through several systems. The publisher observes the creation of ad slots and ad delivery; the buying platform — the requests and deals available to it; the advertiser — spend and business outcomes. Each system aggregates its own slice of the flow. That is why a column reading Impressions: 8,000 says nothing by itself about what actually happened, or why its number should or should not match a neighboring report.

A metric contract6 is the complete definition of a metric by which a number can be reproduced and interpreted:

event definition + measurement point
+ filters and deduplication
+ numerator / denominator
+ dimensions
+ period and event/reporting/attribution window
+ currency
+ perspective and ledger boundary

For example, CTR = 2%7 is not yet a contract. You need to know which clicks and impressions are eligible, where they are recorded, whether repeats are excluded, over what period, and for which campaign. For a monetary metric, the currency, the currency conversion and refund rules, and whose money it is additionally matter. From here on, the word qualifying means an event that passed the filters and validation of a specific system under its contract, and eligible means an event that system admitted to the calculation (usually in the denominator).

The basic metric classes answer different questions:

Class What it measures Example
count How many qualifying events or units of account were recorded 8,000 impressions
rate What share of one count falls on another 160 clicks / 8,000 impressions = 2% CTR
unit/effective metric How much cost8 or revenue9 falls on a unit of outcome $96 / 160 clicks = $0.60 CPC10
return metric How the measured value compares to the investment $480 / $96 = 5.0x ROAS11

Before comparing two columns, ask seven questions:

Who counts?
What is counted and at which measurement point?
Where does the measurement boundary run?
When did the event occur and which period/window did it fall into?
Which filters and deduplication were applied?
What is the denominator?
Which currency is it in, and whose money is it?

This contract matters more than the name. Even an arithmetically simple formula answers a different question once rendered impression12 is swapped for served impression13, click for install, or advertiser spend14 for publisher revenue15.

From opportunity to a possible exposure

On a media surface the publisher first gets an ad opportunity16 — a concrete chance to show an ad, for example a full-screen placement17 after a level is completed in a game. A request is already a technical message sent to a specific system asking it to select an ad, return a creative18, or take part in an auction19. These are different levels:

  • an opportunity may not produce a request because of a local rule or filter;
  • one request may carry several ad objects;
  • one opportunity may fan out into several downstream bid requests;
  • retries increase the request count without a new opportunity;
  • a batch combines several opportunities in one message.

The IAB Tech Lab Programmatic Auction Definitions use Ad Request for information about an impression opportunity put up for auction. At the OpenRTB 2.6 protocol boundary, a single BidRequest contains one or more Imp objects. This is a sufficient reason not to treat HTTP requests as an unqualified proxy for user-facing opportunities; a specific production implementation may differ.

publisher surface
  → ad opportunity
  → 0..N request / bid request
  → response / bid / win
  → served response
  → rendered impression
  → measurable → viewable impression
  → click
  → advertiser-defined conversion
  → attributed revenue / customer value

This is a map of measurement20 points, not a universally strictly decreasing funnel. Fan-out and retries can produce more requests than opportunities; a video pod — an ad break made of several video ad units in a row (chapter 48) — yields several impressions for one opportunity; one user can make several clicks or conversions21; filters, deduplication, and different windows change the set of events between reports.

An impression is a recorded ad event according to the counting rule of a specific system. The word without a qualifier proves neither render, nor viewability, nor human attention.

State Minimal evidence What it does not prove
served impression / served response The server or platform recorded issuing an ad response, code, or creative instructions That the device loaded and started displaying the creative
rendered impression The client/device started a render under the applicable rule That a sufficient part of the creative was on screen long enough
measurable The measurement tool was able to determine the geometry/time conditions of exposure That the viewability conditions are met
viewable impression22 A valid rendered served impression met the applicable geometry/time criteria That a human actually looked at the ad or paid attention
non-viewable Measurement occurred, the criteria were not met That the impression was non-measurable
non-measurable / undetermined The available signals are insufficient to determine viewability That the impression was viewable or non-viewable

The MRC Desktop Display Impression Measurement Guidelines require, for a qualified client-side count, that the creative was downloaded and the render at least began. The MRC Viewable Ad Impression Guidelines (v2.0) tie a viewable impression to a valid rendered served impression. The classic benchmark of the Viewable Guidelines 2014/2015 is at least 50% of pixels for one continuous second for display and two continuous seconds for video; the Digital Audience-Based Standard 2017 uses the same one. This is not a universal rule for every format and environment: the MRC publishes separate guidelines for mobile, video, OTT/CTV (streaming and connected TV; chapter 49), in-game, and AR; the details belong to chapter 37.

Reach and frequency: from events to audience

Reach23 is the de-duplicated number of unique audience units (entity: a person, a device, or a household) that received at least one qualifying exposure in a period, or the share of such units in an explicitly declared population:

reach count = unique entities with ≥ 1 qualifying exposure in the period
reach rate  = reached unique entities / declared universe × 100%

The audience unit must be named: person, device, household, account, or modeled audience. 20,000 device IDs, 20,000 modeled persons, and 20,000 households are not interchangeable counts. Without the size of the universe — the declared audience against which the share is computed — the reach rate cannot be computed; without identity coverage and cross-device deduplication rules, the completeness of reach cannot be assessed.

Frequency24 is the average number of qualifying exposures per reached audience unit over the same period:

frequency = qualifying impressions / reached unique entities

An illustrative example: 60,000 qualifying viewable impressions in a week are distributed across 20,000 de-duplicated persons from the chosen population. The reach count is 20,000 persons, and the average frequency is 60,000 / 20,000 = 3. This does not mean every person saw the ad exactly three times: frequency is an average.

The MRC Digital Audience-Based Standard (2017) uses qualifying viewable impressions as its measurement basis. Product dashboards may count reach on a different impression basis, by devices, or with modeling. That is why reach and frequency are comparable only with an identical audience unit, exposure rule, population, de-dup scope, and period.

From exposure to response and outcome

A click is a qualifying user-initiated interaction with an ad. Even a single action has several measurement points:

user initiates click
  → click tracker / redirect      # measured click
  → destination receives request # received click
  → landing page resolves/loads   # resolved click

The user may close the screen, the network may drop, the redirect may not complete, and different systems may apply different filters. So initiated, measured, received, and resolved clicks need not match. The IAB/MRC Click Measurement Guidelines v1.0 (2009) are still listed by the MRC as a measurement baseline, but their browser/HTTP wording is legacy and does not exhaustively describe modern app and platform interactions; the redirect above is a typical, not the only, flow.

CTR (Click-Through Rate) shows the share of the selected impressions after which the selected clicks were registered:

CTR = qualifying clicks / counted impressions × 100%

If an advertiser report counts 160 clicks and 8,000 impressions under one contract, CTR = 2%. Swapping served impressions for rendered, all clicks for filtered clicks, or a campaign-level population for a creative-level population will change the number. A high CTR by itself does not mean quality traffic, conversions, or causal lift.

A conversion is a valuable action or outcome that the advertiser defined as meaningful and that the system credited under its measurement and attribution25 rules. For a subscription app it can be an install, a trial, the first paid subscription, or a renewal. These actions must not be summed without an explicit counting rule: one click can lead to an install, a trial, and a subscription — that is, to several conversions.

An observed purchase means the event was recorded. An attributed conversion means a chosen attribution rule assigned an ad touchpoint credit for that outcome. Counts depend on the conversion definition, the attribution/lookback window (the time after a touchpoint during which an outcome can still receive credit; chapter 25), filters, late-arriving events, and the counting mode. Neither an observed nor an attributed conversion by itself proves that the ad caused the outcome: attribution distributes credit, while causality requires proof of incrementality — an experimental estimate of the lift the ad produced (chapter 34).

CVR (Conversion Rate)26 is meaningful only with a named precursor event:

CVR = qualifying conversions / declared eligible precursor events × 100%
Full name Numerator Denominator Which transition it measures
click-to-install CVR attributed installs eligible clicks Click → install
install-to-trial CVR qualifying trials attributed installs Install → trial
click-to-subscription CVR attributed subscriptions eligible clicks Click → subscription

A bare CVR hides different funnels. Even a value above 100% is not always an error: for example, Google Ads counts conversions on trackable ad interactions, and the Every mode (count every conversion after an interaction rather than one) together with several conversion actions can yield more than one conversion per interaction. The denominator and the counting rule are part of the metric’s name.

Normalized price of an event: CPM, eCPM, and cost-per-X

Absolute spend does not allow comparing activities of different sizes. Unit metrics divide recognized cost or revenue by an explicitly defined event. In a post-fact report these are actual averages; the same names may denote a pricing model, a bid27, or an optimization target28, which does not guarantee they match the realized average.

Metric Actual formula Perspective and denominator Main ambiguity
CPM (Cost per Mille)29 advertiser cost or spend / counted impressions × 1,000 Buyer; a thousand counted impressions May mean a pricing basis30, a rate, a bid, or the observed average CPM
eCPM (effective CPM)31 recognized revenue / counted impressions × 1,000 Usually publisher/seller; a thousand counted impressions Revenue and impression scope depend on the seller contract
CPC (Cost per Click) cost / qualifying clicks Buyer; clicks under a disclosed rule A bid/target CPC is not the actual average CPC
CPA (Cost per Action)32 cost / qualifying actions Buyer; advertiser-defined action The action can be a trial, order, lead, or subscription, not necessarily a new customer
CPI (Cost per Install)33 cost / qualifying installs Mobile acquisition; attributed installs The install count depends on measurement and attribution rules
CPL (Cost per Lead)34 cost / qualifying leads Lead generation; business-defined lead Lead definition and quality vary; a lead is not yet a customer
CAC (Customer Acquisition Cost)35 allocated acquisition costs / new customers The advertiser’s business; new customers The numerator may include media, agency36, sales, and onboarding costs

Mille means a thousand. The observed CPM answers: how much did the buyer spend per thousand counted impressions? The eCPM answers: how much recognized revenue did the seller receive per thousand counted impressions? The formulas are symmetric, but the numerators belong to different ledger boundaries37. A publisher can use eCPM to compare demand38 paid on CPM, CPC, or CPA on a common realized-revenue basis. eCPM also appears on the buy side — as a normalized cost: a DSP39 paying on CPC or CPA converts its spend into a price per thousand impressions to compare it with CPM-priced inventory. So CPM and eCPM are not automatically two names for one number.

CPC, CPA, CPI, and CPL follow one pattern, recognized cost / qualifying events, but the event defines the business meaning. A low CPL is useless if the lead does not pass qualification; a low CPI does not say the install will convert to a subscription; a CPA cannot be read without the name of the action.

CAC is broader than platform CPA. If a platform report shows a $4 CPA for a subscription, it uses platform cost and the attributed subscription action. A business may count as a new customer only the first confirmed payment and include media, agency, sales, and onboarding in acquisition costs. Until the customer event, period, cost allocation, and refunds are agreed, a platform CPA cannot be renamed CAC.

Whose money and what return

One amount changes its name depending on the participant and the ledger boundary:

advertiser: spend / media cost
        → selling or buying platform:
          revenue on one boundary,
          downstream media/traffic cost on another boundary
        → publisher: recognized publisher revenue

Revenue is the amount recognized by the chosen participant under its rules. Spend is the buyer’s expenditure on media in the chosen scope. Cost is the expenditure of the participant whose economics are being analyzed; for an advertiser it can be media cost, for a platform downstream cost, for a business a broader set of expenses. Estimated dashboard value, invoiced amount, settled amount, and recognized revenue can belong to different states. To compare, you must agree on period, timezone, currency conversion, taxes/refunds, and participant. Gross/net accounting40, take rate41, and settlement42 are covered in detail in chapter 4.

ROAS (Return on Ad Spend) measures media efficiency:

ROAS = attributed conversion value or revenue / ad spend

The result can be written as 5.0x or 500%. It depends on the attribution rules and the chosen value field. ROI (Return on Investment)43 assesses the profitability of a broader investment scope:

ROI = net profit / total relevant investment cost
    = (return − total relevant cost) / total relevant cost

In the illustrative walkthrough, the advertiser receives $480 of attributed conversion value against $96 of ad spend: ROAS is 5.0x. If the $480 is treated as attributed revenue and the full agreed cost — including media, product delivery, agency, and operations — is $600, net profit is −$120 and ROI is −20%. A high ROAS is compatible with a negative ROI because the denominators and cost scopes differ. Both ROAS and attributed revenue remain attribution-dependent: without a causal design they do not prove incremental return.

User value over time: ARPU, ARPPU, and LTV

An acquisition metric says how much it cost to obtain an event or a customer; a product-value metric says how much the chosen user population brought in — or, according to a model, will bring in.

Metric Working formula Required contract
ARPU (Average Revenue per User)44 selected revenue / all eligible users User definition, revenue categories, period, cohort/population
ARPPU (Average Revenue per Paying User)45 selected purchase revenue / paying users Payer event, purchase revenue, refunds, period
LTV (Lifetime Value)46 cumulative observed or predicted value / cohort members Cohort, horizon, identity, observed/predicted status, revenue/contribution basis

An illustrative subscription cohort: in a month, 1,000 active users generate $1,000 of selected recognized revenue, 100 of whom pay; for simplicity, all revenue in the example is purchase revenue. Then ARPU = $1,000 / 1,000 = $1 and ARPPU = $1,000 / 100 = $10. ARPU includes the non-paying users of the chosen population; ARPPU divides only by payers. These numbers say nothing about lifetime without a cohort and a horizon (cohort analysis — chapter 33).

The Google Analytics Data API, for example, defines ARPU as total revenue per active user with purchase, subscription, and ad revenue, and ARPPU as purchase revenue per paying active user. This is a product-specific contract, not a universal vocabulary: another company may define active user, payer, revenue, taxes, and refunds differently.

Observed LTV sums the value actually recorded for a cohort up to a given horizon, for example 30-day revenue LTV. Predicted LTV estimates future value with a model and assumptions, for example 12-month predicted contribution LTV. Revenue LTV does not subtract expenses; contribution/profit LTV uses an agreed margin basis. GA4 User Lifetime unifies lifetime interactions and supports observed and predictive metrics, but the identity method and sampling affect the result. That is why a user’s LTV without a cohort, horizon, identity rule, value basis, and model status is not a defined number.

Where monetization is lost: fill rate and win rate

Fill and win describe different stages of the flow and belong to different owners. They are not complements of each other adding up to 100%.

Metric One common contract Stage and owner Why the number changes further
Fill rate47 filled responses / eligible ad requests Publisher/ad server48: request → fill A filled response may not render and may not become a counted impression, so fill on a response basis is higher than the share of delivered impressions
Bid rate submitted bids / eligible bid requests Buyer: request → bid/no-bid The buyer filters opportunities and does not respond to every request
Win rate49 winning bids / submitted eligible bids Buyer/deal: bid → auction win A win notice is not yet a delivery or billing event

Fill rate has no single safe vendor-neutral denominator. A product may divide filled responses, counted impressions, or another fill event by requests, opportunities, or ad units — and these are different metrics. The Google Ad Manager sell-through documentation uses Total impressions / Total ad requests; for optimized video pods, part of the requests/unfilled units is derived from durations and the configured ad opportunity duration. This is an exact product definition, not a rule for every ad server. Match rate, response rate, delivery rate, and fill rate may name adjacent transitions rather than aliases.

Win rate also requires a disclosed population. An auction may receive a bid, select it as the winner, send a win notice, and still not deliver the creative or create a billable impression50. OpenRTB distinguishes a win notice from a billing notice; billing policy may depend on delivery, viewability, or other rules. Legacy Google Ad Manager reporting had the formula Deals win rate = winning bids / bids; this is a dated example of a denominator, not a universal taxonomy.

One campaign, three truthful reports

Let us assemble the metrics on one consistent scenario. All numbers below are illustrative; their scale is arbitrary and chosen for simple arithmetic. A subscription app is advertised inside a mobile game; the period is, say, one week, and the currency is USD. Each calculation uses its own explicitly named contract rather than assuming a shared global event store.

Advertiser: outcome and acquisition efficiency

The advertiser report contains only its slice: 8,000 counted impressions, 160 qualifying clicks, 80 attributed installs, 24 attributed subscriptions, $96 of spend, and $480 of attributed conversion value.

CTR  = 160 / 8,000 × 100%        = 2%
CPM  = $96 / 8,000 × 1,000       = $12
CPC  = $96 / 160                 = $0.60
CPI  = $96 / 80                  = $1.20
CVR  = 80 / 160 × 100%           = 50%   # click → install
CVR  = 24 / 80 × 100%            = 30%   # install → subscription
CVR  = 24 / 160 × 100%           = 15%   # click → subscription
CPA  = $96 / 24                  = $4    # action = attributed subscription
ROAS = $480 / $96                = 5.0x

Here $12 CPM, $1.20 CPI, and $4 CPA are post-fact averages, not claims about a billing model or a bid. The three CVRs describe three different transitions of one funnel and are not comparable without the label. $480 is attributed value, not proven causal value.

Publisher: delivery and monetization

Over the same period the publisher aggregates all buyers, not just this advertiser: 100,000 eligible ad requests, 80,000 counted impressions, and $800 of recognized publisher revenue.

fill rate = 80,000 / 100,000 × 100% = 80%   # numerator — counted impressions under this ad server's contract
eCPM      = $800 / 80,000 × 1,000   = $10

The publisher totals are naturally larger than the advertiser’s slice. Its $800 revenue is not the advertiser’s $96 spend: it is a different set of impressions and a different ledger boundary. The gap between the buyer’s $12 CPM and the seller’s $10 eCPM is made up of intermediary fees along the money path and of a different mix of impressions; intermediary fees are the subject of chapter 4.

Platform: auction and delivery boundary

The platform in question — in the illustrative example a DSP through which only this advertiser buys — sees 60,000 eligible bid requests, submits 45,000 bids, receives 9,000 wins; 8,000 of those events become counted/billable impressions under the chosen contract. That is why its 8,000 billable impressions match the advertiser’s 8,000 counted impressions; in reality even this match is not guaranteed.

bid rate = 45,000 / 60,000 × 100% = 75%
win rate = 9,000 / 45,000 × 100%  = 20%

The difference between 9,000 wins and 8,000 counted/billable impressions need not be an error: win and billing/delivery are different event points. Likewise, the platform’s 60,000 requests need not equal the publisher’s 100,000 requests because of measurement boundaries, routing, batching, fan-out, and filters.

Perspective Observed subset Key decisions Monetary meaning
Advertiser Bought impressions, clicks, attributed outcomes Acquisition and value efficiency Spend/cost and attributed value
Publisher All eligible requests and demand sources Delivery and monetization Recognized publisher revenue
Platform Traversed requests, bids, wins, and its billing events Auction participation and the delivery boundary Its own revenue, spend, or downstream cost under a specific ledger

Here the data flow and the money flow diverge. Impression/click/conversion events may arrive asynchronously into different reports; invoicing and settlement need not follow the same network path. Arithmetic consistency within each metric contract does not prove that raw events are equal across systems. For primary diagnostics, the seven questions from the first section are needed again. Detailed reconciliation belongs to chapter 32.

Common Confusions

  • An opportunity, an ad request, and a bid request are not one event. An opportunity arises on the media surface; requests are system-bound messages subject to batching, fan-out, retries, and filtering.
  • An impression does not mean a human viewing. Served, rendered, measurable, and viewable provide different levels of evidence; viewable means a standardized opportunity to see, not attention.
  • Non-viewable is not the same as non-measurable. In the first case measurement completed and the criteria were not met; in the second the status could not be determined.
  • CPM is not eCPM, and CPA is not CAC. The first pair differs in ledger perspective; the second in action/customer definition and cost scope.
  • ROAS is not ROI. ROAS normalizes attributed value against ad spend; ROI uses net profit and the full chosen investment cost.
  • Fill rate and win rate are not one auction metric seen from two sides. Fill belongs to publisher delivery; win belongs to the bid → win transition under a disclosed auction contract.

What to Remember

  1. A metric is a contract of measurement, not just a name and a formula.
  2. Reach requires a definition of the audience unit and the universe; frequency requires a shared impression basis, reached population, and period.
  3. CTR and especially CVR are comparable only with named numerators, denominators, and filters; a CVR is named as a specific transition.
  4. CPM/CPC/CPA/CPI/CPL can be a contractual price, a bid/target, or actual averages; eCPM usually normalizes seller revenue.
  5. Conversion, ROAS, and attributed revenue depend on attribution rules; attribution is not causality.
  6. ARPU, ARPPU, and LTV require user, revenue, cohort, period/horizon, and model definitions.
  7. Advertiser, publisher, and platform can simultaneously have correct but different reports on one campaign: they see different boundaries and ledgers, and one amount (spend, cost, revenue) changes its name when it crosses a boundary.

Check Yourself

  1. One opportunity produced three downstream bid requests. Which count describes the media event, and which one the protocol messages?
  2. Why can CVR = 15% not be compared with another CVR = 12% until the denominators and conversion rules are disclosed?
  3. How can a campaign with 5.0x ROAS have a negative ROI?
  4. Which metric contracts allow an advertiser CPM of $12, a publisher eCPM of $10, and a platform win rate of 20% at the same time without contradiction?

Sources and Further Reading

  1. MRC — Standards & Guidelines
  2. MRC/IAB — Desktop Display Impression Measurement Guidelines v1.1 and MRC — Viewable Ad Impression Measurement Guidelines v2.0
  3. MRC/IAB — Digital Audience-Based Measurement Standards, Final 1.0
  4. IAB/MRC — Click Measurement Guidelines v1.0
  5. MRC — Outcomes and Data Quality Standards
  6. IAB Tech Lab — Programmatic Auction Definitions (final version dated June 24, 2026) and OpenRTB 2.x
  7. Google Ad Manager — Sell-through and Report metrics (checked August 27, 2026; the Reports tool has since been replaced by Interactive reports)
  8. Google Ads — CTR, Conversion rate, ROI, Google Analytics Data API schema and GA4 User lifetime (checked August 27, 2026)

Footnotes

  1. Impression — a delivery or display event for the selected ad, recorded by a system under a specific rule; a counting unit, not a fact of human perception. Served, rendered, and viewable are different delivery states, so the counters of different participants in the same deal may not match. ↩

  2. Click — a qualifying user-initiated interaction with an ad, recorded under a specific system’s rule. Even a single action has several measurement points: initiated, measured (click tracker), received (destination), and resolved (landing page) clicks need not match due to dropped connections, redirect chains, and filters. ↩

  3. Campaign — an organized set of an advertiser’s advertising activity with a goal, a budget, delivery rules, and creatives. Before delivery, its conditions are communicated to the buying systems; objects and events are linked by identifiers such as campaign_id. ↩

  4. Advertiser — a market participant that funds advertising for a business result: installs, subscription purchases, growth in awareness. It buys not a person but the opportunity to address an audience in a specific context; it creates demand and sets goals, budget, and creative. ↩

  5. Publisher — the owner or operator of a media environment where advertising opportunities arise: a website, a mobile app, a video service, a game. It monetizes the environment with advertising, creates supply, and decides which placements, formats, and buyers to admit, balancing revenue against user experience. ↩

  6. Metric contract — the complete definition of a metric by which its number can be reproduced and interpreted: the event definition, measurement point, filters and deduplication, numerator/denominator, dimensions, period and windows, currency, perspective, and ledger boundary. A metric’s name without a contract does not fix its meaning: the same formula on different contracts answers different questions. ↩

  7. CTR (Click-Through Rate) — the share of selected impressions after which the selected clicks were recorded: CTR = qualifying clicks / counted impressions × 100%. The number depends on the click/impression basis (served or rendered, filtered or all) and the population; a high CTR by itself does not indicate quality traffic, conversions, or causal lift. ↩

  8. Cost — the expenses of the entity whose economics are being analyzed, in the chosen scope: for an advertiser, the media cost; for a platform, the downstream cost; for a business, a broader set of expenses. Not a synonym for spend: spend is the buyer’s outlay on media, while cost can include non-payment and out-of-media components and is always tied to a perspective. ↩

  9. Revenue — the amount recognized by the chosen entity under its accounting rules on its ledger boundary. Estimated dashboard value, invoiced amount, settled amount, and recognized revenue can be different states of one chain; for comparison, the period, timezone, currency conversion, taxes/refunds, and participant must be agreed. How much of the money passing through a platform it recognizes as revenue depends on its principal or agent role (gross/net presentation), not on the size of the invoice. ↩

  10. CPC (Cost per Click) — the cost or spend per qualifying clicks under a disclosed rule: CPC = cost / qualifying clicks. A bid/target CPC is not the actual average CPC: the former is a bid or an optimization target, the latter is a post-fact average. ↩

  11. ROAS (Return on Ad Spend) — media efficiency: attributed conversion value or revenue per ad spend: ROAS = attributed conversion value or revenue / ad spend; written as 5.0x or 500%. Attribution-dependent: it depends on the attribution rules and the chosen value field, and without a causal design it does not prove incremental return. A high ROAS is compatible with a negative ROI. ↩

  12. Rendered impression — an impression counted after the creative has loaded and been rendered on the user’s device (usually on a signal from client-side code: a pixel or an SDK event after render). It proves rendering, but not that the creative entered the visible area of the screen or was noticed; the MRC standard for display counts an impression precisely on client-side load/render. ↩

  13. Served impression — an impression counted on the fact that the ad server or selling platform returned the creative (or a reference to it) in response to a request; it proves sending, but not loading, rendering, or visibility. The earliest and largest number in the served → rendered → viewable ladder. ↩

  14. Media spend — the money an advertiser allocates and spends on buying advertising. It passes through the media sellers and ultimately forms publisher revenue, shrinking along the way by fees; it is not a synonym for publisher revenue, nor for intermediary fees — these are different amounts in one chain. ↩

  15. Publisher revenue — the amount the publisher recognizes as revenue under its own accounting rules from media payments for its inventory; in the simple case it equals the publisher payout, but recognition (accrued / billed / settled) follows the publisher’s policy, not the moment of payment. It differs from advertiser spend by the sum of all fees in the chain; participants’ service and SaaS fees outside the media money do not count toward it. ↩

  16. Ad opportunity — a specific opportunity to select and show an ad that arises when a user reaches a placement. A short-lived event: in a digital environment it can vanish faster than a person can make a decision, so systems process it in milliseconds, and in programmatic it is put up for bidding as a separate unit. ↩

  17. Placement (ad slot or selectable media scope) — an overloaded product term. On the publisher side, it is a predefined ad slot or rule for ad appearance that generates ad opportunities as the product is used. On the buy side, placement can mean an explicitly selected delivery environment — a site, app, channel, video, section, or individual slot. The entity level must be clarified from the product contract. ↩

  18. Creative — the advertising material that is delivered and shown to the user: a banner, a video, a native format, a playable. Along with targeting rules it is an object of delivery and must match the format and rules of the publisher’s environment; for a campaign, several creatives are usually prepared. ↩

  19. Auction — the mechanism for choosing a winner among several bids for one ad opportunity and for determining the price (clearing price) under the auction’s rules — first-price, second-price, and their variants. An auction win is the event of choosing a winner; it is not an impression and does not by itself make the impression billable. ↩

  20. Measurement — the function of recording and aggregating advertising events (impressions, clicks, outcomes) for reporting and optimization. Measurement providers give an independent point of observation; different participants have different timestamps and counting rules, so reports do not have to match perfectly. In mobile, the dedicated category is the MMP. ↩

  21. Conversion — a valuable action or outcome that the advertiser has defined as significant (install, trial, first payment, renewal, purchase) and that a system has counted under its measurement and attribution rules. Conversions cannot be summed without an explicit counting rule: one click can lead to several conversions. An observed conversion is a recorded event; an attributed conversion is an event to which an attribution rule assigned credit; neither proves causality. ↩

  22. Viewable impression — an impression that met the applicable geometry/time criteria of viewability (the classic MRC benchmark: at least 50% of the creative’s pixels on screen for one continuous second for display and two for video). Viewable means a standardized opportunity to see, not a person’s attention; non-viewable (the measurement took place, the criteria were not met) is not the same as non-measurable (the status could not be determined). ↩

  23. Reach — the de-duplicated number of unique entities (persons, devices, households, accounts, or a modeled audience) that received at least one qualifying exposure in a period, or the share of such entities in an explicitly defined population: reach rate = reached unique entities / declared universe × 100%. Comparable only with identical entity definition, exposure rule, population, de-duplication scope, and period. ↩

  24. Frequency — the average number of qualifying exposures per reached unique entity over the same period: frequency = qualifying impressions / reached unique entities. It is an average, not a distribution: it does not mean that every person saw the ad the same number of times. ↩

  25. Attribution — the rules by which the available advertising touchpoints (impression, click) are matched to outcomes (install, registration, purchase) and receive credit for the conversion. Attribution answers the question “which source was credited with the action under the chosen rules,” not “whether the ad was the true cause”: credit does not prove causality. Attribution providers are a separate cross-cutting category; in mobile, the MMP. ↩

  26. CVR (Conversion Rate) — the share of qualifying conversions among the declared eligible precursor events: CVR = qualifying conversions / declared eligible precursor events × 100%. A bare CVR hides the transition: only named transitions such as click-to-install or install-to-trial are comparable; a value above 100% is possible with several conversion actions on one interaction. ↩

  27. Bid — a buyer’s offer to pay a certain price for a specific ad opportunity, usually expressed as a CPM. A bid is a pre-event decision: it is neither the final price (clearing price) nor the billable amount; in programmatic, the DSP forms a bid for each bid request separately. ↩

  28. Optimization target — the outcome toward which a bidding system tunes its future decisions: most often an optimization event (install, paid subscription) plus a goal for it. Not to be confused with a strategy parameter: target CPA $4 is a bid-strategy constraint, not the event the system predicts. It does not turn the contract into CPA billing: the payment model establishes the contract, the realized metrics describe what happened post-fact, and the optimization target influences only decisions before the event. ↩

  29. CPM (Cost per Mille) — the cost of a thousand counted impressions for the buyer: CPM = advertiser cost or spend / counted impressions × 1,000. The term is ambiguous without a contract: it can mean a pricing basis, a rate, a bid, or a post-fact observed average CPM; the observed CPM answers the question “how much did the buyer spend per thousand counted impressions.” ↩

  30. Pricing basis — the contractual rule for calculating a price: the object for which the buyer pays — an impression, click, action, install, lead, a share of proceeds, a period of access to software, or another agreed result. A metric’s name in an interface (CPM, CPC, CPA) does not by itself fix the basis: the same name can be a contractual price, a bid, or the actual average after the events, so only the basis named in the contract creates a monetary obligation. ↩

  31. eCPM (effective CPM) — the seller’s recognized revenue per thousand counted impressions: eCPM = recognized revenue / counted impressions × 1,000. Usually the publisher’s perspective; it makes it possible to compare demand paid on CPM, CPC, or CPA on a common realized-revenue basis. Not a synonym for CPM: the numerators belong to different ledger boundaries. ↩

  32. CPA (Cost per Action) — the cost per advertiser-defined qualifying actions: CPA = cost / qualifying actions. The action is defined by the campaign contract — a trial, an order, a lead, or a subscription, not necessarily a new customer — so a CPA cannot be read without the name of the action or compared between campaigns directly. ↩

  33. CPI (Cost per Install) — the cost of mobile acquisition per qualifying installs: CPI = cost / qualifying installs. The install count depends on the measurement and attribution rules, so a CPI is comparable only under named install-counting rules. ↩

  34. CPL (Cost per Lead) — the cost per business-defined qualifying leads: CPL = cost / qualifying leads. Lead definitions and quality differ between businesses, and a lead is not yet a customer, so a low CPL is useless without qualification. ↩

  35. CAC (Customer Acquisition Cost) — the business’s allocated acquisition costs per new customer: CAC = allocated acquisition costs / new customers. Broader than the platform CPA: the numerator can include media, agency, sales, and onboarding costs, and a new customer is defined by a business rule (for example, the first confirmed payment). A platform CPA cannot be renamed a CAC until the customer event, period, cost allocation, and refunds are agreed. ↩

  36. Agency — a contractor company that plans and executes marketing and media work on behalf of an advertiser: develops strategy, prepares creative, selects channels, launches a campaign, and reports on results. An advertiser can keep media buying in-house, hand it to an agency in full, or split the responsibility. An agency is an organization and a contractual role, as opposed to a media buyer — an operational function that can also work inside an advertiser. ↩

  37. Ledger boundary — the boundary within which an entity recognizes and records amounts and events under its own rules: the same amount is revenue on one boundary and a downstream media/traffic cost on another. Because of different ledger boundaries, the reports of the advertiser, the platform, and the publisher about the same deal naturally do not match and do not have to match. ↩

  38. Demand — the budgets and intent of advertisers to buy suitable advertising opportunities; the market side around which the buy side / demand side groups — the people and systems that help plan and buy media (agencies, DSPs). Demand is a market side, not a specific technology: platforms must not be confused with sides. ↩

  39. DSP (Demand-Side Platform) — a platform that automates the purchase of advertising opportunities for an advertiser: it aggregates access to several supply sources, evaluates each opportunity, and decides whether to bid and how much to pay. It belongs to the demand side not because it “creates demand” but because it acts in the buyer’s interest. ↩

  40. Gross/net revenue (presentation) — a variant of how the reporting entity presents revenue: gross shows the agreed payment in full, net only the fee or commission. The choice follows the contract-specific principal-agent assessment for the specific specified good/service, not the logic of “money before/after expenses”: the product’s label, the invoice, or gross billings do not by themselves determine the conclusion. ↩

  41. Take rate — the retained amount relative to a named transaction flow on the chosen boundary: take rate = retained amount / named transaction flow. Comparable only when the numerator, denominator, and included fees are disclosed: the same amount yields different rates relative to the boundary proceeds and the original advertiser spend. ↩

  42. Settlement — the actual movement of money between participants for a period, as distinct from the invoice (the issued claim) and from accrual. One amount passes through three states: accrued — recorded from events, billed — issued in an invoice or statement, settled — paid; at each step the amount may change because of adjustments, fraud filtering, credits, and FX. For one side it is a receivable (expected inflow), for the other a payable. ↩

  43. ROI (Return on Investment) — the profitability of a broader investment scope: ROI = net profit / total relevant investment cost = (return − total relevant cost) / total relevant cost. Unlike ROAS, it uses the full chosen cost scope, not just ad spend. ↩

  44. ARPU (Average Revenue per User) — the selected revenue across all eligible users of the chosen population or cohort for a period: ARPU = selected revenue / all eligible users. It requires a user definition, revenue categories (purchase, subscription, ad), a period, and a cohort/population; it includes non-paying users. ↩

  45. ARPPU (Average Revenue per Paying User) — revenue per paying user: ARPPU = selected revenue from payers / paying users. It divides only by payers and requires a payer event, the composition of revenue (purchase, subscription), refund rules, and a period. It is usually higher than the ARPU of the same population, but not always: if ARPU includes ad revenue from non-paying users while ARPPU counts only purchase revenue, the ratio can flip; which revenue enters the numerator is part of the metric’s contract. ↩

  46. LTV (Lifetime Value) — the cumulative observed or predicted value of a cohort’s members: LTV = cumulative observed or predicted value / cohort members. It requires a cohort, a horizon, an identity rule, a value basis (revenue or contribution), and an observed/predicted status; revenue LTV does not subtract costs, and contribution LTV is calculated on the agreed marginal basis. Without these definitions, “a user’s LTV” is not a defined scalar. ↩

  47. Fill rate — the share of eligible ad requests that received an ad response (a fill), for example filled responses / eligible ad requests. A publisher/ad server metric of the request → fill transition; there is no safe vendor-neutral denominator — a product may divide filled responses, matched bids, or counted impressions by requests, opportunities, or ad units, and these are different metrics. A fill is not a final outcome: a filled response may not render and may not become a counted impression, so a response-based fill rate is higher than the share of impressions actually delivered. Not a synonym for match, response, or delivery rate. ↩

  48. Ad server — a system that selects, delivers, and logs advertising under given rules. There are publisher-side ad servers (matching direct campaigns, networks, and programmatic demand, managing delivery) and advertiser-side ones (storing creatives, counting impressions and clicks); the term denotes a function, not a market side. ↩

  49. Win rate — the share of submitted eligible bids that won the auction: winning bids / submitted eligible bids. A buyer/deal metric of the bid → auction win transition; it requires the population to be disclosed. A win notice is not yet delivery and not a billing event: the auction can select a bid as the winner, but the creative may not be delivered or the impression may not become billable. ↩

  50. Billable event — a qualifying event defined by the contract, after which the payee is entitled to accrue the billable amount — the sum under the pricing basis after counts, rates, validation, and contractual adjustments; it is not necessarily invoiced, recognized as revenue, or paid. The event is set by contract policy, not automatically by the protocol: an auction win, a win notice, and a billing notice are different points, and a win does not guarantee delivery or a charge. ↩