Tracking and measurement glossary

AI-ready data is marketing data an AI tool can query and interpret correctly without a person translating it first, which is a higher bar than putting it somewhere a model can reach. It needs a collection layer that is not quietly dropping conversions, a written definition for every metric and dimension so one question has one answer, and no personal data sitting in fields nobody has inspected. Availability is a storage problem, readiness is a definitions problem.

Anomaly detection is a statistical technique that compares a metric against the range a model predicts from that metric's own history, then flags any datapoint whose observed value falls outside that range. Because the prediction carries the weekday and seasonal pattern of the account it learned from, the boundary moves with the metric instead of sitting at a number somebody chose. What it reports is that a value is unusual for this account, never what caused it.

An attribution model is the rule a reporting or advertising platform uses to divide credit for one conversion across the touchpoints that came before it, from giving all of it to the last click to spreading it across several channels. It governs how credit is distributed rather than how much was measured, so two models can disagree about which channel earned a sale without either having seen a different set of events. Changing the model changes the story a report tells about the same events.

A Consent Management Platform, or CMP, is the software that asks a website's visitors for permission to use cookies and their personal data, records each answer as proof, and passes it on to the tags and tools that depend on it. In practice it is the banner, the preference screen behind it, the log of who agreed to what, and the signals other tools read. A CMP collects and broadcasts consent, it does not enforce it: whether a tag actually stops firing after a refusal depends on how the site is wired to those signals.

Consent Mode v2 is Google's mechanism for passing a visitor's consent choices to Google tags, using four separate signals: one for advertising storage, one for analytics storage, one for sending user data to Google for advertising, and one for personalized advertising. Tags read those signals and adjust what they collect and send, so a refusal changes how a tag behaves instead of being ignored. It is not a cookie banner and does not collect consent, it only carries the choice a banner recorded.

Consent rate is the share of website visitors who grant a specific category of cookie or tracking consent, most often analytics or advertising, out of everyone a consent banner reached. It is calculated as accepted decisions for that category divided by total visitors shown the banner, and because EU law treats a non-response the same as a refusal, a visitor who never clicks anything falls on the declined side of that calculation rather than out of it. Sessions, conversions and audience sizes downstream only ever include the visitors on the accepted side, so consent rate sets the ceiling on how much of a report is built from directly measured events at all.

Content scoring is a method for ranking the pages on a website by how much each one contributes to pushing a visitor toward a conversion, such as a form or a booking request, rather than by how much traffic the page gets. It looks at the full sequence of pages a visitor viewed before converting and spreads credit across that route, so a page in the middle of a journey can score higher than the page a visitor happened to be on last. The score measures contribution to an outcome, not popularity.

Contribution margin is what is left from a sale after subtracting only the costs that move with that sale, such as product cost, payment fees, shipping and commission, and before subtracting costs that stay the same, such as salaries, rent or software licenses. It is usually stated as an amount per order or unit, or as a ratio once divided by revenue. Because it isolates variable costs, contribution margin shows how much of a sale is left to cover fixed costs and profit, a different number from gross margin or net profit margin, which subtract different costs.

Conversational analytics is a chat interface that lets someone ask a plain-language question about marketing data, such as which channel drove the most leads last month, and get a generated answer back instead of building a query or reading a dashboard. It sits on top of an existing data platform and translates the question into a query against that platform, rather than running the analysis itself. The interface is new, the data and the definitions behind it are exactly what they were before.

Conversion modelling is a machine learning technique that estimates conversions or user behaviour a platform could not directly observe, usually because a visitor declined analytics or advertising cookies, by learning the patterns of consenting visitors and applying them to the ones who declined. Google Ads and Google Analytics 4 each run their own version, training only on the traffic inside that account rather than sharing one model across advertisers. The result is a statistical estimate blended into reports next to directly measured numbers, not a recovered record of what one specific denied visitor did.

A Conversions API (CAPI) is a server-to-server route an advertising platform exposes so conversion events can be sent to it directly from the advertiser's own systems, instead of from a script in the visitor's browser. Meta's is the one usually meant by CAPI, and other platforms run their own under their own names and rules. Because the request does not depend on the platform's script loading in the browser, the event survives a blocked or failed page, though it is only credited to a campaign when the identifying data sent with it is enough to match a person.

Customer acquisition cost (CAC) is the total sales and marketing spend required to win one new customer over a given period, divided by the number of customers acquired in that period. A full calculation adds the cost of the people, tools and agency fees involved, not media spend alone. CAC is calculated against confirmed customers, a different number from the cost per lead or cost per conversion an ad platform reports on its own dashboard.

A data layer is a structured JavaScript object on a web page that holds information about the page, the visitor and what they just did, under a fixed set of named keys. Google Tag Manager and gtag.js read that object to decide which tags fire and which values they send, instead of reading them out of the page's HTML. It is the handover point between a website and the tools that measure it.

A data warehouse is a central system that holds copies of data taken out of the tools that produced it, organised for analysis and reporting rather than for running those tools. It keeps current and historical records from several sources in one place, so website events, ad spend and CRM revenue can be queried together and compared over time. It is a storage and query layer, so it reflects the quality of whatever is loaded into it rather than improving it.

Enhanced conversions is a Google Ads feature that sends hashed first-party customer data, such as an email address or phone number, with a conversion the website already recorded. Google matches that hashed data against accounts a person was signed in to when they engaged with an ad, so a conversion the ad click alone could not tie to a campaign can still be credited to one. It raises the match rate on conversions already measured, it does not collect the ones a site never measured.

Event deduplication is how a measurement or advertising platform recognises that two copies of the same conversion have arrived, usually one sent from the visitor's browser and one sent from a server, and counts them once instead of twice. It works by matching an identifier that both copies carry, such as a shared event ID or the order's transaction ID. Each platform applies its own key, scope and time window, so deduplication is a per-destination behaviour, never one setting on a tracking setup.

A first-party cookie is a small piece of data a browser stores under the same domain as the site the visitor is currently on, and sends back with each later request to that domain. A cookie is third-party instead when its domain differs from the site in the address bar. Sites use first-party cookies to keep someone logged in, hold a basket, remember a preference and recognise a returning visitor.

First-party data is information a business collects directly from its own audience through channels it controls, such as its website, app, CRM and email list. What makes it first-party is the relationship rather than the technology: the people it describes dealt with that business directly, and it is held under that business's own terms and legal basis. Third-party data, by contrast, is bought or licensed from an outside provider that collected it somewhere else.

Intelligent Tracking Prevention (ITP) is the set of privacy restrictions built into WebKit, the browser engine behind Safari, and it is on by default, not something a visitor switches on. It blocks all third-party cookies with no exceptions, and it also limits first-party storage, deleting cookies and other storage written by JavaScript after seven days of browser use without interaction on the site. Those limits apply to every site Safari loads, not only to domains ITP classifies as tracking capable.

Markov attribution is a data-driven method for crediting marketing channels in a multi-touch customer journey, built by modelling visitor paths as a Markov chain, a graph connecting channels by the probability of moving from one to the next. Credit for each channel is its removal effect, the drop in conversion probability the model calculates when that channel is removed from the graph and every path is recalculated without it. The graph comes from an account's own recorded paths, so it needs a large, varied set of multi-touch journeys to produce a stable answer.

Offline conversion import is the process of sending a conversion that happened away from a website, such as a signed contract or a deal a CRM marks won, back into an ad platform after the fact. The platform matches it to the original ad click using an identifier stored at that click, most often a Google Click ID (GCLID), rather than anything read from the website itself. Once matched, the offline outcome counts toward the campaign's reported conversions and, where enabled, toward what the bidding algorithm optimizes against.

POAS, or profit on ad spend, divides the profit an advertising channel produced by the ad spend that produced it, where ROAS divides revenue by that same spend. It is not one number: what counts as profit depends on which costs are subtracted first, and most reported POAS figures subtract only the cost of the goods. An ad platform cannot calculate it on its own, because it knows what an order was worth and nothing about what fulfilling that order cost.

Sampling is the practice of estimating a total from a subset of the underlying events instead of counting every one, applied by an analytics or advertising platform when a query would otherwise have to process more data than it is built to handle at that size. The output is a scaled estimate with a margin of error rather than an exact count, and that error tends to grow as the query is sliced into smaller segments or narrower date ranges. Most platforms flag a sampled result somewhere in the interface, though the flag rarely survives into an exported report or a slide.

Server-side Google Tag Manager, often written sGTM, is a Tag Manager container that runs on a server you host, most often on Google Cloud, instead of running in the visitor's browser. It receives the HTTP requests generated by the web container's tags, then processes and forwards that data on to Google Ads, GA4, Meta and other endpoints, using the same tag, trigger and variable model as a standard container. It is one specific tool for doing server-side tracking, not the technique itself, and other server containers do the same underlying job.

Server-side tracking collects marketing events on a server you control instead of in the visitor's browser, and that server forwards them on to platforms such as Google Ads, GA4 and Meta. The browser makes one request to your own domain rather than a separate request to every vendor, so collection no longer depends on what each visitor's browser and extensions allow. It does not remove the need for consent, and it cannot recover an event a visitor never agreed to send.

Value-based bidding is a Google Ads Smart Bidding approach that optimizes campaigns toward the total value of conversions rather than simply the number of them. It requires the account to report at least two different conversion values, whether real revenue or a proxy such as a lead score, so the algorithm has something to tell a valuable outcome apart from a cheap one. Google runs it through two strategies, Maximize conversion value with or without a Target ROAS, both bidding toward value instead of volume.