Attribution models reference: how credit is assigned

Last-click attribution assigns 100% of a conversion's credit to the last touchpoint before it. It is simple, deterministic, and the historical default — which is exactly why it misleads: it ignores every earlier touch that created demand, systematically overrating bottom-funnel channels and underrating discovery.

Data-driven attribution (DDA) assigns credit using a model trained on a site's own conversion paths rather than a fixed rule like last-click. Done well it credits assist touches more fairly. Its caveats are real: it needs enough conversion volume, it is a model not a measurement, and it cannot see touches that were never tracked.

First-click attribution assigns 100% of a conversion's credit to the very first touchpoint a visitor had. It is the mirror image of last-click: it celebrates discovery and awareness channels while ignoring everything that nurtured and closed the journey. Useful for studying acquisition, misleading as a sole budget lens.

Linear attribution divides a conversion's credit equally among all touchpoints in the path. It is the simplest multi-touch model: every touch matters the same. That even-handedness avoids the single-touch extremes, but it also pretends a fleeting impression and a decisive demo are worth the same — which is rarely true.

Time-decay attribution weights touchpoints by recency: the closer a touch is to the conversion, the more credit it earns, usually following an exponential decay with a configurable half-life. It is a compromise between last-click and linear, but its recency bias under-credits the early demand-creating touches.

A lookback (or conversion) window is the period before a conversion in which earlier touchpoints are eligible for credit. Touches outside the window are ignored entirely. Because every attribution model only sees touches inside this window, its length quietly governs which channels can ever receive credit.

An assisted conversion is one where a channel participated in the path but was not the closing touch. The assisted-conversions view is a corrective to last-click: it reveals the supporting channels that last-click hides. It is a count of participation, not a clean measure of incremental contribution.

Marketing mix modeling (MMM) estimates how much each channel contributed to outcomes using aggregate, time-series data — spend, sales, seasonality — rather than user-level paths. It predates digital tracking, needs no cookies, and is gaining renewed interest as privacy limits user-level attribution. It is statistical inference, with real uncertainty.

Position-based (U-shaped) attribution gives most credit to the first and last touchpoints — commonly 40% each — and shares the remaining 20% among middle touches. It tries to honour both discovery and closing while still acknowledging the middle. The specific weights are a convention, not a measured truth.

Multi-touch attribution (MTA) is not one model but the whole family of models that distribute credit across more than the final touch — linear, time-decay, position-based, data-driven. What unites them is the ambition to value the full path, and the shared dependency on every relevant touch being tracked.

The attribution (lookback) window decides which past touches can earn credit for a conversion; the reporting window is the date range you are viewing. They answer different questions, and confusing them is a frequent cause of numbers that 'do not add up' between tools or between dates.

Incrementality testing measures the lift a channel actually causes by withholding it from a control group and comparing outcomes. It answers the question every attribution model dodges: would this conversion have happened anyway? It is causal where attribution is merely correlational, but it requires deliberate experiment design.

A view-through conversion credits an impression a user was served but did not click, when they later convert within an impression window. It tries to value awareness that does not get clicked, but it is among the easiest credits to over-count, because seeing is not the same as being influenced.

Cross-device attribution is the problem of a single person using multiple devices in one journey. Default cookie-based tracking treats each device as a separate visitor, so paths fracture and credit lands on the wrong channel. Closing the gap usually requires a logged-in identity — which carries its own privacy weight.

A conversion path is the ordered list of touchpoints a visitor had before converting — the raw material every attribution model operates on. Reading paths directly, before any credit rule is applied, often reveals more than a single model's tidy split, but short and single-touch paths deserve caution.

The dark funnel is the part of a buyer's journey that leaves no trackable click: private Slack and WhatsApp groups, podcasts, word of mouth, dark social. None of it appears in attribution reports, yet it shapes demand — surfacing instead as unexplained direct, branded-search, and self-reported traffic.

Self-reported attribution asks the buyer directly — usually a 'how did you hear about us?' field — instead of inferring from tracking. It captures untrackable and dark-funnel influence that analytics miss, but it trades cookie blind spots for human memory bias. The two methods are complements, not rivals.

Channel grouping is the rule set that sorts raw source/medium values into named channels — Organic Search, Paid Social, Email, Direct. Every attribution model operates on these buckets, so a mis-grouped touch is mis-attributed no matter how good the model. Grouping is upstream of, and quietly governs, attribution.

First-click and last-click are the two single-touch extremes: one credits the opener, the other the closer. Their value is not in being right — both are wrong about the middle — but in being compared. The gap between them, channel by channel, is the cheapest diagnostic of who creates versus who harvests demand.

Single-touch attribution is the family of models that hand a conversion's entire credit to one touchpoint — first-click or last-click. It is deterministic, easy to explain, and reconciles cleanly across tools. The cost is that it denies any role to every other touch in the journey.

A custom attribution model lets you define your own credit rules — adjusting weights, lookback, and channel treatment beyond the presets. The flexibility can fit a real, unusual journey, but it just as easily encodes the answer you wanted. A custom model is only as honest as the assumptions you can defend.

Walled gardens are closed ad platforms that measure and report the conversions they claim credit for, inside their own systems. Each marks its own homework with its own window and rules, so summed across platforms the attributed conversions routinely exceed the real total — double-counting is structural, not accidental.

U-shaped attribution is the position-based model in its classic form: the first interaction and the conversion-driving interaction each receive a large fixed share (commonly 40% each), and the remaining credit is split among the middle touches. It is a rules-based heuristic that values discovery and closing equally, and it is the lens many tools mean when they say 'position-based'.

W-shaped attribution extends the U-shaped idea by recognizing three milestone interactions rather than two: the first touch, the lead-creation touch, and the opportunity-creation (or closing) touch. Each milestone receives a large fixed share — commonly 30% apiece — and the remaining credit is spread across other touches. It is popular in B2B funnels with defined lifecycle stages.

Full-path attribution is the W-shaped model extended to four milestones: first touch, lead creation, opportunity creation, and the final closing interaction. Each milestone takes a fixed large share and the remaining credit is distributed across other touches. It is designed for long B2B sales cycles where the deal-closing interaction is a distinct, measurable event worth its own credit.

Shapley value attribution applies a concept from cooperative game theory: it treats channels as players in a coalition and assigns each one credit equal to its average marginal contribution across all possible orderings of channels. The result is a principled, order-independent way to split conversion credit. It underpins data-driven attribution in several analytics products.

Markov chain attribution models customer journeys as a probabilistic graph of transitions between channel states, ending in conversion or null. Each channel's credit is derived from its 'removal effect' — how much the overall conversion probability falls if that channel (and its transitions) are removed from the graph. It is a leading algorithmic alternative to Shapley-based attribution.

Attribution models split into two families. Rules-based models apply fixed, human-chosen weights to touchpoints by position — last-click, first-click, linear, time-decay, U/W-shaped. Algorithmic (data-driven) models learn credit from observed conversion paths using methods like Shapley values or Markov chains. This page contrasts the two and explains when each is appropriate.

Modeled conversions are conversions a platform estimates statistically rather than observes directly. When direct measurement is blocked — by missing consent, cross-device journeys, or privacy protections — ad and analytics platforms model the likely conversions from observable trends and aggregated data, and report them alongside observed ones. Understanding which conversions are modeled is essential to reading attribution honestly.

Enhanced conversions is a Google Ads feature that supplements cookie-based conversion measurement by sending hashed first-party customer data — such as an email address the user provided — to match conversions that cookies alone would miss. The data is hashed (SHA-256) before transmission. It is one industry response to the decline of third-party identifiers, with its own consent and configuration requirements.

Offline conversion import (OCI) connects events that happen away from the website — a sales call that closes, an in-store purchase, a qualified lead in a CRM — back to the online ad click that began the journey. It works by capturing a click identifier (such as Google's GCLID) at the start and later uploading the offline outcome keyed to that identifier, closing the online-to-offline attribution loop.

Server-side attribution moves the collection and forwarding of measurement events from the browser to a server you control — via server-side tag management or platform conversion APIs like Meta's CAPI. It can improve resilience to browser restrictions and give you governance over what data leaves your environment, but it is a data-flow change, not a way to bypass consent.

Identity resolution in attribution uses two approaches. Deterministic matching links touchpoints when they share a known, persistent identifier (a logged-in user ID, a hashed email). Probabilistic matching infers that two touchpoints belong to the same user from circumstantial signals — IP, device, behavior — without a confirmed identifier. The two differ sharply in accuracy and privacy posture.

Household-level attribution credits conversions to a household rather than an individual, grouping the devices and people sharing one home (often by a shared IP or a graph of devices). It is common in connected-TV and cross-device measurement, where pinpointing the exact person who saw an ad and the exact person who converted is impossible — and where a household unit is a deliberately privacy-conscious coarser grain.

Conversion lag is the time between an interaction and the resulting conversion. Some conversions happen minutes after a click; others take days or weeks. Because of lag, recent activity always looks under-performing at first — conversions for recent touches have not happened yet — and the lookback window must be long enough to capture them. It is a core reason attribution reports change as data matures.

In time-decay attribution, credit declines exponentially the further a touchpoint is from the conversion, and the half-life is the parameter that sets how fast. A touch one half-life before the conversion gets half the credit of one at conversion time; two half-lives back, a quarter. Choosing the half-life decides how strongly recency is rewarded — a model choice, not a measured fact.

Unified marketing measurement is the practice of combining methods rather than trusting one. It blends bottom-up multi-touch attribution (granular, user-path based), top-down marketing-mix modeling (aggregate, covering offline and untrackable media), and incrementality experiments (causal validation). The goal is a triangulated view that compensates for each method's blind spots instead of relying on a single biased lens.

App Tracking Transparency (ATT) is Apple's framework requiring an app to request user permission before tracking it across apps and websites owned by other companies, or accessing the device's advertising identifier (IDFA). When permission is denied, the IDFA is unavailable, which removed the deterministic identifier mobile attribution long relied on and pushed the ecosystem toward aggregated, privacy-preserving measurement.

SKAdNetwork (SKAN) is Apple's framework for attributing app installs and post-install conversions to ad campaigns without identifying the user or device. Instead of a deterministic identifier, it sends the ad network an aggregated, delayed 'postback' confirming a conversion happened, with deliberately limited campaign granularity and a conversion value of restricted resolution. It is the privacy-preserving backbone of iOS install attribution after ATT.

Consent is upstream of attribution: under frameworks like the EU's GDPR and ePrivacy Directive, storing or reading identifiers for tracking generally requires the user's consent. When consent is declined or withheld, the touchpoints those identifiers would have recorded never enter the data, so attribution operates on partial paths. Understanding consent is therefore inseparable from reading attribution honestly.

Google Analytics 4 (GA4) implements attribution with a data-driven model as the default for its conversion reporting, plus rules-based options, configurable lookback windows, and default channel groupings. It also distinguishes attribution used in GA4 reports from the conversions Google Ads counts. This page describes GA4's attribution posture and the settings that change how credit appears.

Each ad platform measures and attributes conversions within its own boundary: its own conversion windows, its own default model, and counts it reports for itself. Because platforms attribute independently and cannot see each other's touchpoints, their self-reported conversions overlap — the same sale can be claimed by several platforms. This page describes that data-model posture without ranking any platform.

Cross-channel attribution distributes conversion credit across all the channels a user touched — paid search, organic, social, email, referral, direct — rather than crediting only what one platform can observe. It is the antidote to siloed, self-reported platform counts: by viewing the whole path in one place, it can apportion credit coherently and reveal how channels actually work together.

Last non-direct click is an attribution rule that credits the most recent non-direct channel in the path. When the final interaction before converting is 'direct' (someone typing the URL or returning via a bookmark), the model skips it and credits the prior identifiable marketing channel instead — on the reasoning that direct traffic is often the downstream result of earlier marketing rather than a source of its own.

A geo experiment divides geographic regions into a treatment group (which sees a media change) and a control group (which does not), then compares outcomes between them. Because assignment is at the region level rather than the user level, geo experiments measure incremental effect without needing cookies, device IDs, or per-person attribution — making them a privacy-resilient complement to touch-based models.

A conversion lift study randomizes users into a group eligible to see ads and a control group held out from them, then compares conversion rates between the two. The difference estimates incremental conversions — those caused by the ads rather than ones that would have occurred anyway. Major ad platforms offer lift studies as a counterfactual check on attributed conversion counts.

Ghost ads are an experimental design for measuring ad effectiveness. Rather than showing a placebo ad to the control group, the system records which control users would have been served the test ad had they been in the treatment group, then compares only comparable users. This isolates the ad's incremental effect while avoiding the cost and bias of serving placebo creative.

A PSA (public-service announcement) control test is an incrementality design where the control group is served unrelated placebo ads instead of the test campaign. Because both groups receive an ad impression, exposure conditions are similar, and the difference in conversions estimates the test campaign's incremental effect. It is an older alternative to ghost ads.

Holdout-based attribution uses a randomized holdout — a group deliberately excluded from a campaign or channel — to estimate how much of a channel's credited conversions are genuinely incremental. By comparing the treated population against the holdout, it grounds attribution in a counterfactual rather than relying solely on observed click paths, which tend to over-credit channels that intercept already-converting users.

A Conversions API (CAPI) is a server-side interface that sends conversion and event data directly from a business's servers to an ad platform, rather than relying solely on a browser pixel. It exists because in-browser tags increasingly miss events due to tracking prevention, ad blockers, and lost cookies; a server connection can pass events the browser never reported, subject to consent and matching.

Server-to-server (S2S) conversion tracking reports a conversion from one server directly to another — typically your backend to an ad platform or affiliate network — keyed by a click ID captured at the landing visit. It removes the dependence on a browser pixel firing at conversion time, which matters for offline conversions, multi-step flows, and environments where client tags are unreliable.

The Attribution Reporting API is a Privacy Sandbox proposal that lets browsers measure ad conversions without third-party cookies or cross-site identifiers. It produces event-level and aggregatable reports; aggregatable reports are combined into noisy summary reports that give campaign-level conversion counts and values while limiting what can be learned about any individual.

Google Ads attribution determines how credit for a conversion is distributed across a user's ad interactions within the conversion window. The platform offers attribution models (including data-driven attribution) set per conversion action, and the chosen model affects reported conversions and how automated bidding optimizes — without changing the underlying real conversions.

Meta (Facebook/Instagram) attribution assigns conversions to ad interactions using an attribution setting that combines a click-through window and a view-through window. Conversions are credited within the chosen window and reported on the day the interaction occurred. Because windows and the click/view distinction differ from other platforms, Meta-reported conversions will not match GA4 or other tools.

Google Analytics 4 uses data-driven attribution (DDA) as its default model, but DDA requires sufficient data to train per conversion event. When a property or conversion lacks enough conversions and paths, GA4 cannot model credit reliably and behavior differs. Understanding the data requirements explains why channel credit can look unstable on low-volume properties.

Value-based attribution assigns the monetary value of a conversion — not just a count of one — across the touchpoints in the path. It matters because optimizing for conversion counts treats a low-value and a high-value sale identically; distributing value lets bidding and analysis favor the channels that bring more revenue, provided conversion values are passed accurately.

Paid vs organic attribution is the distinction between crediting conversions to paid channels (ads) versus organic ones (SEO, direct, referral, organic social). It matters because the two often overlap on the same path, and platform-specific attribution can claim conversions that organic also influenced — making it easy to over-credit paid media if you do not reconcile the views.

Brand vs non-brand attribution separates conversions driven by branded queries (people already looking for you) from non-branded ones (people discovering you via generic terms). The split matters because brand traffic often converts on demand that existed already, so crediting brand campaigns can overstate their incremental impact, while non-brand activity is more likely to be generating new demand.

Fractional attribution assigns each touchpoint a fraction of a conversion rather than the whole credit, so a multi-touch path distributes one conversion across several channels. It is the mechanism behind linear, time-decay, position-based, and data-driven models, and it explains why per-channel conversion counts can be decimals that still sum to the real total.

Conversion credit distribution describes how an attribution model allocates the credit for a conversion across the interactions that preceded it. Every model — single-touch or multi-touch, rules-based or algorithmic — is fundamentally a different distribution rule. Understanding distribution as the shared concept clarifies why models disagree even on identical paths.

The time-to-conversion distribution is the spread of elapsed times between an early interaction and the eventual conversion across a population. Reading it reveals how much of your conversion volume is fast versus slow, which is essential for choosing lookback windows, interpreting recent-period reports, and avoiding the mistake of judging campaigns before slow conversions land.

Attribution bias is the systematic, predictable way a given model mis-assigns credit relative to true causal contribution. Last-click over-credits closing and demand-harvesting channels; first-click over-credits discovery; view-through can over-credit cheap impressions. Recognizing each model's bias is essential because no observational model recovers causation on its own.

Observed conversions are directly recorded from events that the system actually saw. Modeled conversions are statistical estimates that fill gaps left by consent declines, cross-device journeys, or blocked tags. Modern reports blend both, so understanding which conversions are measured versus estimated is essential to reading a total honestly and not treating an estimate as a count.

Attribution, incrementality testing, and marketing-mix modeling (MMM) are three distinct measurement approaches often confused. Attribution distributes credit across observed touches; incrementality experiments measure causal lift versus a control; MMM uses aggregate, often top-down regression on spend and outcomes. They answer different questions and should be used together, not treated as interchangeable.

Duplicate conversion counting happens when a single real conversion is recorded more than once — for example by both a browser pixel and a server event, by a tag firing twice, or by two platforms each claiming it. It silently inflates reported conversions and value, distorts ROAS, and misleads bidding unless deduplication via shared event IDs and clear ownership is in place.

Attribution data discrepancies are the routine mismatches between conversion numbers reported by different tools — an ad platform versus site analytics, or two analytics products. They arise from different attribution models, lookback and reporting windows, time zones, deduplication rules, bot filtering, and consent handling. Most discrepancies are structural and expected, so the goal is to explain them, not eliminate them.

Google announced that, starting in 2023, Google Analytics 4 would deprecate several rule-based attribution models — first-click, linear, time-decay, and position-based — in the Attribution settings and reports. After the change, GA4 offers data-driven attribution (the default for new properties) and a paid-and-organic last-click model. Knowing exactly which models survived prevents teams from chasing reports that no longer exist.

Alongside data-driven attribution, Google Analytics 4 retains last-click models that assign 100% of conversion credit to the final touch. GA4 documents a paid-and-organic last-click model that credits the last channel and, like last-non-direct logic, avoids crediting a direct visit when a prior campaign click is available. Understanding this default-eligible model clarifies how GA4 reports single-touch credit after the 2023 model changes.

Web-to-app attribution connects a marketing touch that happened on the web to a conversion that happens inside a native mobile app. The handoff is hard: app store installs, OS privacy limits, and the loss of browser identifiers sit between the click and the in-app event. Platforms use deferred deep links, install referrers, and matching to reconnect the journey. This page explains the mechanics and the measurement gaps.

Deep link attribution relies on links that open a specific screen inside an app rather than a generic home view, while carrying the campaign data needed to attribute the resulting conversion. Universal Links, App Links, and custom schemes route the tap; deferred variants apply the context after an install completes. This page explains how deep links preserve attribution signal across the web-to-app and app-to-app boundary.

Store visit conversions are an ad-platform measurement that estimates how many people visited a physical store after seeing or clicking an ad. Google documents that store visits are modeled and aggregated, derived from anonymized, consented location data and statistical extrapolation rather than tracking specific individuals into a shop. This page explains the modeled nature of the metric and how to read it responsibly.

Phone call conversions count phone calls as conversions and attribute them to the ad or campaign that drove them. Ad platforms use call extensions with dynamic forwarding numbers, or track clicks on a phone-number link on the website, to connect a call back to its source. This page explains how call tracking works, what it can attribute, and the privacy considerations around recording or measuring calls.

Conversion modeling fills gaps where direct observation fails, but platforms only report modeled figures when they have enough data to model reliably. These thresholds mean a low-volume campaign may show no modeled conversions at all, not because none occurred but because the estimate would be too unstable. This page explains why thresholds exist and how they shape what you can and cannot read from modeled reports.

Difference-in-differences (DiD) is a quasi-experimental method that estimates the causal effect of an intervention — like turning a campaign on in some regions — by comparing how a treated group changed against how an untreated control group changed over the same time. By differencing out both pre-existing gaps and shared time trends, DiD isolates the incremental effect. This page explains the method, its key assumption, and where it fits in measurement.

The synthetic control method estimates causal impact by constructing a 'synthetic' version of the treated unit — a weighted blend of comparison units that closely matches its pre-intervention behavior. The gap between the real treated outcome and its synthetic counterfactual after the intervention is the estimated effect. It is widely used in geo-experiments where a single market is treated. This page explains the construction and its assumptions.

Every conversion total contains a baseline — what would have happened without the marketing — and an incremental portion driven by it. Incremental lift is that incremental portion: conversions a campaign actually caused, over and above the baseline. Confusing the two leads to crediting marketing for sales it did not cause. This page defines baseline and incremental lift and explains how experiments estimate the split.

In measurement, a halo effect occurs when activity in one channel, campaign, or product drives demand that converts elsewhere. Brand advertising lifting branded search, or a hero product lifting a whole catalog, are classic examples. Last-touch attribution credits the downstream channel and misses the halo. This page explains the halo effect, why it understates upstream activity, and how experiments surface it.

Cannibalization in measurement is the opposite of a halo: a channel captures conversions that another channel — often organic or direct — would have delivered anyway. Branded paid search bidding on terms users would have clicked organically is the canonical case. Attribution credits the paid click, but the incremental value may be small. This page explains cannibalization and how incrementality testing exposes it.

Propensity score matching (PSM) is an observational method for estimating causal effects when randomization is not available. It models each unit's probability of being exposed (the propensity score) from observed characteristics, then compares exposed and unexposed units with similar scores. By balancing the groups on measured confounders, PSM approximates an experiment. This page explains the technique and its key limitation — unmeasured confounding.

Media mix modeling (MMM) and multi-touch attribution (MTA) often disagree because they measure differently: MMM is top-down and aggregate, capturing offline and brand effects; MTA is bottom-up and user-path-based, granular but blind to unobservable touches. Reconciliation treats them as complementary lenses to be aligned, not rivals to be ranked. This page explains why they diverge and how teams triangulate between them.

Return on ad spend (ROAS) and marketing return on investment (ROI) are often conflated but measure different things. ROAS is revenue divided by advertising spend — a top-line efficiency ratio. Marketing ROI is profit (or net gain) divided by the full cost of the marketing — a bottom-line return. A campaign can have a high ROAS yet a poor ROI once margins and total costs are included. This page defines both formulas and when each applies.

Blended attribution takes total business results — say, all orders — and relates them to total spend across every channel, ignoring per-platform claims. Platform-reported attribution is what each ad platform credits itself using its own model and self-reporting. Because platforms can double-count and credit non-incremental conversions, summed platform numbers often exceed reality. This page contrasts the two views and where each is useful.

Conversion import from a CRM connects later, offline business outcomes — a qualified lead, an opportunity, a closed-won deal — back to the marketing click that started the journey. The click is captured with an identifier (such as Google's GCLID), stored in the CRM, and uploaded back to the ad platform when the outcome occurs. This lets optimization target real revenue events, not just form fills. This page explains the flow and its requirements.

Automated bidding optimizes toward the conversions it is told to value — and the attribution model determines how that credit is assigned across the path. Switching from last-click to data-driven attribution, for example, changes which keywords and audiences get credit, which in turn changes how Smart Bidding allocates budget. This page explains the tight coupling between attribution model choice and bid strategy behavior.

Some attribution tools historically used device fingerprinting — combining browser and device signals to re-identify users without cookies. Browser vendors and privacy frameworks increasingly restrict or block fingerprinting because it identifies users covertly. This page explains why fingerprinting-based attribution is constrained and points toward consented, first-party, and aggregate alternatives. It does not endorse fingerprinting.

Organizations that run multiple ad accounts — agencies, multi-brand companies, or manager hierarchies — often need conversions defined once and shared across accounts. Cross-account conversion tracking lets a manager account hold conversion actions that linked sub-accounts use, so every account optimizes toward consistent outcomes and avoids divergent or double-counted definitions. This page explains the setup and the consistency and double-counting considerations.

Attribution assumes the touches it credits came from real people. Bots — crawlers, click fraud, automated agents — break that assumption: invalid clicks can inflate a channel's apparent contribution, and bot sessions can muddy conversion paths. Platforms filter known invalid traffic, but no filter is perfect. This page explains how bot traffic distorts attribution and how filtering and first-party bot detection mitigate it.

Attribution data is not final the moment a conversion happens. Conversion lag, late-arriving offline and CRM imports, modeling that backfills over time, and platform processing delays all mean recent numbers keep moving. Reading the last day or two as settled leads to false conclusions. This page explains why attribution data matures and how to wait for stability before judging performance.

The GA4 model comparison report (under Advertising > Attribution) places two attribution models next to each other for the same conversion events, exposing how much credit each channel gains or loses when you change the rule. It does not change billing or optimization — it is a diagnostic to understand model sensitivity before acting.

A brand lift study estimates the causal effect of advertising on attitudinal outcomes — ad recall, awareness, consideration, favourability — by surveying an exposed group and a control group that did not see the ad. The difference in survey responses is the lift. It measures perception change, not clicks or conversions, so it complements conversion attribution rather than replacing it.

B2B attribution is harder than B2C because a single purchase involves a buying committee of several people, a sales cycle of weeks to quarters, and a close that happens in a CRM rather than on the website. Touchpoints scatter across people and time, much of the decision happens off-site, and the final 'conversion' is a deal stage, not a checkout. This page explains why standard models break and what to track instead.

CRM closed-loop attribution connects the top of the funnel (web visits, campaign clicks, lead forms) to the bottom (qualified opportunities and won revenue in the CRM) by carrying an identifier from the first touch into the lead record. It 'closes the loop' so marketing credit follows actual booked revenue rather than stopping at the form submission. This is the backbone of B2B and high-consideration measurement.

Post-click attribution credits a conversion to an ad the user clicked; post-impression (view-through) attribution credits a conversion to an ad the user saw but did not click, within a window. Post-click rests on a deliberate action; post-impression rests on an ad being served and (sometimes) viewable. They measure different strengths of evidence, and mixing them without labels inflates totals.

Affiliate attribution is the rule that assigns a sale (and the commission) to a referring partner, conventionally the last affiliate click within a tracking-cookie window. Because money changes hands, the model is contractual, not just analytical: it determines who is paid, invites last-click stuffing, and clashes with the brand's own multi-touch view of the same sale. This page explains the mechanics and the conflicts.

The GA4 conversion paths report (Advertising > Attribution > Conversion paths) lists the channel sequences users followed before converting and splits data-driven credit across early, middle, and late positions of those paths. It answers 'what touched users before they converted, and in what order?' — turning multi-touch theory into an inspectable table rather than a single last-click number.

GA4's attribution settings (Admin > Attribution settings) define the property-wide reporting attribution model and the lookback windows for acquisition and other conversions. Changing them re-attributes credit across the property's reports going forward and, for the model, retroactively in attribution reports. Understanding these settings is prerequisite to reading any GA4 attribution number correctly.

Reporting identity is the GA4 setting that decides how events are joined into individual users for reporting: User-ID, Google signals, device, and modeling, applied in a chosen order (Blended, Observed, or Device-based). Because attribution depends on knowing which touches belong to the same person, the identity space directly affects path lengths, deduplicated user counts, and credit distribution.

A search lift study estimates how much additional searching — for the brand or related terms — an advertising campaign causes, by comparing search behavior between an exposed group and a randomized control. It captures a demand-generation effect that conversion attribution misses: ads that prompt people to search rather than click straight through. Like brand lift, it is a randomized-experiment measure, not a click count.

Foot traffic (store visit) attribution estimates how many people visited a physical location after seeing or clicking an ad, using aggregated and modeled location signals from consenting panels rather than tracking individuals. Because it relies on sampling and modeling, it is reported as a modeled estimate above a privacy threshold, not a precise headcount. It bridges digital ads to offline visits where on-site conversion tracking cannot reach.

Measuring influencer impact is hard because most of it is unattributable by clicks: audiences see content, remember the brand, and convert later through search or direct. Practical influencer attribution combines unique discount codes, dedicated tracking links, and self-reported 'how did you hear about us?' surveys, accepting that view-through and dark-social effects will always be undercounted by last-click models.

Coupon (promo) code attribution assigns a sale to the partner, creator, or campaign whose code the buyer entered at checkout. It is deterministic and cross-device by nature — the code is typed regardless of cookies — which makes it popular for influencers and affiliates. But it only captures buyers willing to use a code, and shared or leaked codes can be claimed by buyers a partner never reached.

A vanity URL is a short, memorable address (like brand.com/show) that a person can hear and type, used in podcasts, radio, TV, and print where no link is clickable. When typed, it redirects to a landing page carrying campaign parameters, so an otherwise untrackable offline exposure becomes an attributable visit. It trades reach for measurability: only listeners who remember and type it are captured.

Attributing podcast advertising is constrained by the medium: ads are read aloud, there is no click, and a download is not proof of a listen. Practical podcast attribution combines unique promo codes, spoken vanity URLs, post-purchase surveys, and download-prefix analytics, plus pixel-based methods where the host platform supports them. Each captures a different slice, and none alone is complete.

TV, radio, and print have no click, so their attribution is built from indirect evidence: correlating exact spot airtimes with spikes in site traffic and search, dedicated vanity URLs and promo codes, self-reported surveys, and — most rigorously — geo or matched-market experiments that compare regions with and without the buy. Each method trades precision for the reach these channels uniquely deliver.

When a purchase takes months, attribution windows become the binding constraint: cookies expire, click lookbacks lapse, and the first touches that created the opportunity are gone by the time it closes. Standard digital attribution then over-credits whatever happened near the close. Measuring long cycles means moving the system of record to the CRM, extending or replacing windows, and accepting modeling for the unrecoverable early touches.

Lead scoring assigns a quality or readiness score to each lead from fit and engagement signals; attribution credits the marketing sources that produced the lead. They are distinct but complementary: scoring weights quality, attribution weights origin. Joining them shifts measurement from 'which channel drives the most leads' to 'which channel drives the most qualified leads' — the question that protects budget from high-volume, low-fit sources.

Total (attributed) conversions are every conversion a channel gets credit for under some model. Incremental conversions are the subset that would not have occurred without that channel — the causal effect measured by a holdout. The difference matters because a channel can be credited with conversions it merely rode along on. Total answers 'how many did we attribute here?'; incremental answers 'how many did this channel actually cause?'

Blended ROAS is total revenue divided by total advertising spend across every channel, with no attribution model applied. Because platforms each claim overlapping conversions, summing platform-reported ROAS overstates performance; the blended ratio sidesteps that by working from one revenue figure and one spend figure. It is honest at the top line but cannot tell you which channel earned the return — that still needs attribution or incrementality.

Privacy-safe attribution is the design goal of measuring marketing without tracking individuals across sites. It favors aggregation, consent-gated first-party data, on-device and server-side processing, differential-privacy-style noise, and modeling to fill consent gaps — explicitly rejecting fingerprinting and covert cross-site identifiers. It accepts coarser, modeled results as the price of measurement that respects users and regulation.

Durable measurement is the strategy of building attribution that keeps working as third-party cookies disappear and consent tightens. Rather than one fix, it layers a first-party data foundation, consent signaling, server-side collection, conversion modeling for gaps, and incrementality testing as ground truth. The aim is resilience: measurement that degrades gracefully instead of collapsing when a single identifier vanishes.

When conversions happen in different currencies, attribution must convert each conversion value into a single reporting currency before credit and ROAS can be compared. Platforms apply exchange rates (often daily) at value capture, so the same sale can report slightly different amounts depending on when and at what rate it was converted. Getting currency handling right is a prerequisite for trustworthy cross-market attribution and ROAS.

In CRM-driven funnels, a single 'conversion' is too blunt: a deal moves through stages (created, qualified, proposal, closed-won), and different touches influence different stages. Opportunity stage attribution assigns credit by the stage a touch helped reach — for example crediting content that created the opportunity separately from the demo that progressed it — giving a stage-aware view of which marketing moved deals along, not just which closed them.

Matched market testing measures causal impact by pairing geographic markets with similar historical behavior, running a campaign in the test market while holding out its matched control, and attributing the post-period difference to the campaign. It is the practical workhorse for offline and channel-level incrementality where user-level randomization is impossible — closely related to geo experiments and the synthetic control method.

A tracking template is a rule that constructs the final click URL — appending campaign parameters and click identifiers — when an ad is served, instead of relying on manually tagged destination URLs. Defined once at the account, campaign, or ad level, it standardizes attribution parameters across many ads and reduces the tagging errors that silently miscategorize traffic. It is the upstream guarantor of clean channel data.

Conversion window overlap is what happens when multiple ad platforms each track their own click-to-conversion window for a buyer who touched several of them. A single sale can fall inside Google's window and Meta's window at once, so both count it. The overlap is structural, not a bug: walled gardens measure independently. Recognizing it explains why summed platform conversions exceed the real total and why de-duplication is required.

GA4's BigQuery export delivers raw, event-level data — every event with its parameters and user/session identifiers — enabling teams to compute attribution outside the platform's built-in models. With the full path available as rows, analysts can implement custom rules, Markov or Shapley models, or reconcile against CRM and spend data, rather than accepting only the models the reporting UI offers. It is the foundation for bespoke, auditable attribution.