Glossary

Metrics

Key performance metrics and KPIs

Term	Definition
Spend	The total amount of money spent on advertising campaigns. Pulled directly from your connected ad platforms.
Revenue	The total revenue generated from conversions attributed to your ads. Calculated based on conversion values tracked by the Adverfly pixel.
ROAS	Return on Ad Spend. The ratio of revenue to ad spend. Calculated as Revenue / Spend. A ROAS of 3.0 means $3 earned for every $1 spent.
ROI	Return on Investment. The percentage return on your advertising investment. Calculated as ((Revenue - Spend) / Spend) × 100.
CPA	Cost per Acquisition. The average cost to acquire one customer or conversion. Calculated as Spend / Conversions.
CAC	Customer Acquisition Cost. The cost of acquiring a new customer. Often used interchangeably with CPA.
CPM	Cost per Mille. The cost per 1,000 impressions. Calculated as (Spend / Impressions) × 1000.
CPC	Cost per Click. The average cost for each click on your ad. Calculated as Spend / Clicks.
CTR	Click-Through Rate. The percentage of impressions that resulted in a click. Calculated as (Clicks / Impressions) × 100.
CVR	Conversion Rate. The percentage of sessions that resulted in a conversion. Calculated as (Conversions / Sessions) × 100.
AOV	Average Order Value. The average value of each order/conversion. Calculated as Revenue / Conversions.
LTV	Lifetime Value. The total revenue expected from a customer over their entire relationship with your business.
Impressions	The number of times your ad was displayed to users.
Clicks	The number of times users clicked on your ad.
Conversions	The number of desired actions completed (purchases, sign-ups, etc.) attributed to your ads.
Sessions	A single visit to your website tracked by the Adverfly pixel. A session groups all events from one visitor within a continuous browsing period. Sessions are the denominator for Conversion Rate (CVR).
Hook Rate	The percentage of viewers who watched the first 3 seconds of a video ad. Calculated as (3-Second Video Views / Impressions) × 100. Measures how well a creative captures attention.
Hold Rate	The percentage of viewers who watched a video ad to the end (ThruPlay). Calculated as (ThruPlays / 3-Second Video Views) × 100. Measures how well a creative retains attention.
ThruPlay	A video metric counting when someone watches a video to the end or for at least 15 seconds.
Net ROAS	Net Return on Ad Spend. ROAS adjusted for costs beyond ad spend (e.g., COGS, shipping). Provides a profitability-focused view of ad performance.
MER	Marketing Efficiency Ratio. Total revenue divided by total marketing spend across all channels. Unlike ROAS which is per-channel, MER measures overall marketing efficiency.
CPO	Cost per Order. The average cost to generate one order. Calculated as Spend / Orders.

Attribution

Attribution models and terminology

Term	Definition
Cart Attribute Attribution	How a purchase is tied back to an Adverfly landing page when the checkout happens on your shop's own domain. The page stamps which page and which variant the visitor saw into the Shopify cart; Shopify carries those onto the order, and the order sync reads them back. This is what survives the move between two domains, where cookies and browser storage do not.
Attribution	The process of assigning credit to marketing touchpoints that contributed to a conversion.
Attribution Window	The time period during which a touchpoint can receive credit for a conversion. E.g., a 7-day click window means clicks within 7 days before conversion get credit.
Click Attribution	Credit given to ad clicks that led to a conversion.
View Attribution	Credit given to ad impressions (views) that may have influenced a conversion, even without a click.
Last Click	An attribution model that gives 100% credit to the last touchpoint before conversion. Useful as a baseline comparison.
Any Click	Last click per platform: every platform clicked inside the attribution window receives 100% credit for the order, assigned to its own last click. Because each platform is credited independently, channel rows add up to more than real revenue — by design. Use it to compare a channel, campaign or ad with its own ad-platform dashboard, not for totals or financial reporting.
First Click	An attribution model that gives 100% credit to the first touchpoint in the customer journey. Useful for understanding awareness drivers.
Linear	An attribution model that distributes credit equally across all touchpoints in the customer journey. Provides a balanced view when every interaction is considered equally important.
U-Shaped	An attribution model that emphasizes the first and last touchpoints (typically 40% each), with less credit distributed among the middle interactions. Also known as position-based attribution.
Total Impact	An attribution model that measures the incremental impact of marketing across channels by evaluating each touchpoint's contribution to the final outcome, combining click and impression signals.
MTA (Multi-Touch Attribution)	A user-level attribution approach that distributes conversion credit across multiple touchpoints in a single customer journey. MTA provides the most granular view of marketing performance and acknowledges that multiple interactions collectively influence purchasing decisions.
MMM	Marketing Mix Modeling. A statistical modeling technique that quantifies the incremental impact of each marketing channel on revenue using aggregated data, saturation curves, and external factors. See the dedicated MMM glossary for detailed terms.
UMM	Unified Marketing Measurement. A holistic approach combining multiple measurement methodologies (MTA, MMM, incrementality testing) into a single framework. Instead of treating each method in isolation, UMM cross-validates insights to reduce conflicting answers and produce defensible budget decisions. See the Measurement glossary for more details.
Incrementality	A measurement approach that isolates the portion of performance that would not have happened without advertising. By comparing exposed and control groups, incrementality reveals the true causal lift of marketing efforts. See the Measurement glossary for detailed terms.
Prism	Adverfly's evidence-based attribution engine. It reconciles every order into one canonical record, splits its revenue across the evidence behind it — observed clicks and what happened in those sessions (pages, product views, cart, checkout, recordings), self-reported survey answers, redeemed codes — and always reports what it could not place. Its weights are learned per workspace from converting and non-converting visits, validated on later data, and fall back to a stated heuristic until they beat the baseline. Named for the invariant: a prism splits light into its spectrum and loses none of it.
Canonical Order	One purchase, counted once, after shop orders, payment records, browser events and server events describing the same sale have been reconciled. Cancellations are excluded and refunds are subtracted, so the canonical order is what attributed revenue must add up to.
Recognized Order Revenue	The gross value of a canonical order minus the refunds booked against it. Every attribution share plus the unattributed share equals exactly this amount.
Evidence Record	A single piece of information about how an order came about: an observed click, a self-reported survey answer, or a redeemed creator or referral code. Several technical records of the same interaction collapse into one evidence record so they cannot earn several votes.
Self-Reported Channel	A channel a customer named in a post-purchase survey without a matching tracked interaction. It can hold credit at channel level, marked as self-reported, but never implies a specific ad, campaign, timestamp, or that the exposure was paid.
Discovery Question	A post-purchase survey question asking how the customer first found the brand. Distinct from a purchase-trigger question, so two different answers are not automatically a contradiction.
Purchase-Trigger Question	A post-purchase survey question asking what made the customer buy now. Recorded as evidence and shown in the order explanation, but it does not move channel credit.
Allocation Policy	The versioned set of weights that turns evidence into credit shares. Deterministic by design: the same evidence and the same policy version always produce the same split, which is what makes an explanation reproducible.
Unattributed	Revenue from orders the available evidence cannot place. Reported as its own bucket rather than distributed, so a small unattributed share is a claim about data coverage, not a target to optimise.
Unresolved Entity	An interaction whose channel is known while the campaign or ad behind it is not. Kept as its own bucket in campaign and ad views instead of being dropped or guessed.
Reconciliation	The check that attributed revenue plus unattributed revenue equals the order ledger, in minor units. A report that does not reconcile is reported as broken rather than shown.

Measurement

Unified measurement, incrementality testing, and cross-method validation

Term	Definition
Unified Measurement	A holistic approach that combines attribution (MTA), Marketing Mix Modeling (MMM), and incrementality testing into a single measurement framework. Instead of relying on one method, unified measurement cross-validates insights across all three to reduce conflicting answers, uncover hidden growth opportunities, and produce defensible budget recommendations.
Triangulation	The practice of validating marketing insights by comparing results across multiple measurement methods — such as MMM, MTA, and incrementality. When all three methods agree, confidence in the insight is high. When they disagree, it signals the need for further testing.
Causality	The ability to prove that a marketing action directly caused an outcome, rather than merely being correlated with it. Incrementality testing is the primary method for establishing causality in marketing measurement.
Incrementality	A measurement approach that isolates the portion of performance that would not have happened without advertising. By comparing an exposed group to a control group, incrementality reveals the true lift caused by marketing — as opposed to conversions that would have occurred organically.
Incremental Lift	The difference in outcomes between a treatment group (exposed to ads) and a holdout group (not exposed). Represents the additional conversions or revenue directly caused by advertising efforts.
Lift (%)	The percentage share of outcomes caused by ads during a test window. Calculated as (Treatment − Holdout) / Holdout × 100. A higher lift percentage indicates a greater causal impact of the advertising.
iROAS (Incremental Return on Ad Spend)	The revenue generated specifically because of advertising, divided by the incremental ad spend. Unlike standard ROAS, iROAS excludes organic conversions and measures only the true return from ad investment.
Treatment vs. Holdout	An experimental design where one group (treatment) is exposed to advertising while another group (holdout) is intentionally excluded. Comparing outcomes between the two groups reveals the incremental impact of the ads.
Conversion Lift Test	A user-level holdout test typically managed by ad platforms (Meta, Google, TikTok). The platform splits audiences into exposed and holdout groups using its identity graph, then measures the difference in conversions. Also known as a holdout test.
Geo Lift Test	An incrementality test that measures impact by comparing performance across geographic regions (cities, states, or postal codes) with and without advertising exposure. Instead of user-level data, it uses regional sales trends to assess incremental impact — making it privacy-friendly and scalable.
DMA (Designated Market Area)	A geographic region used as the recommended granularity for geo lift tests. DMAs define distinct media markets, making them ideal for isolating advertising impact by region.
Matched Markets	In geo lift testing, control markets selected because they closely match the treatment markets on historical performance and trends. For example, if Berlin is a treatment market, Hamburg might serve as a matched control.
Synthetic Control	A statistical method used in geo lift tests that creates a weighted blend of multiple control regions to closely approximate the treatment region's pre-test behavior. Provides a more precise comparison than a single control market.
View-Through Conversion (VTC)	A conversion credited to an ad that was viewed (impressioned) but not clicked. VTCs capture the influence of brand awareness and visual advertising that standard click-based attribution misses.
Modeled Conversions	Estimated conversions inferred by ad platforms when direct tracking is unavailable due to privacy restrictions, cookie deprecation, or signal loss. Platforms use machine learning to fill gaps in observable conversion data.
Point of Diminishing Returns (PDR)	The spend level at which additional investment in a channel yields progressively less incremental return. Unified measurement helps identify PDR across channels to prevent over-investment and guide budget reallocation.
Triangulation Weights	The confidence-based weighting used to combine measurement methods into a unified ROI. With geo-test: 50% Experiment + 30% MMM + 20% MTA. Without geo-test: 60% MMM + 40% MTA. Fallback: 100% MTA. Weights reflect each method's causal reliability.
Confidence Score	A numeric score (0–1) indicating how trustworthy a unified measurement result is. Ranges from 0.95 (significant geo-test available) to 0.40 (MTA only). Higher scores mean more causal evidence supports the result.
Ground Truth Feedback	When a geo-test reveals a channel's true causal impact, that measured lift is fed back into the MMM as a Bayesian prior — forcing the model to converge toward experimentally validated truth. This closes the loop between experiments and statistical modeling.
Difference-in-Differences (DID)	A statistical method used in geo-tests that compares the change in outcomes between treatment and control regions before and after an intervention. Uses a frequentist t-test to determine whether the treatment effect is statistically significant.
Halo Effect (Channel)	The indirect revenue and organic traffic uplift caused by paid marketing activity beyond direct tracked conversions. This effect applies to all channels (not just creators) and is inherently captured by MMM: because the model correlates total spend with total revenue over time, any indirect lift — brand searches, word-of-mouth, organic traffic spikes — is naturally attributed to the channel whose spend drove it. No separate measurement phase or feature is required; the halo effect is a built-in property of the Marketing Mix Model.
A/B Test	A comparison in which visitors are split between a control and one or more variants, so the difference in outcome can be attributed to the change rather than to the period, the audience or the weather. Adverfly assigns every visitor with a stable per-browser hash, so the same person always sees the same variant.
Rollout	Shipping one change to a share of visitors without a control arm. There is nothing to compare against, so a rollout reports delivery and conversion but never a lift, a p-value or a winner. Use it when the decision is already made and you want to limit the blast radius; use an A/B test when you want to know whether the change helped.
Fixed Horizon	The classic experiment design: plan the sample size in advance, look once at the end. Anything read earlier is exploratory, because repeated peeking at a growing sample inflates false positives dramatically.
Sequential Testing	A design whose p-values and confidence intervals stay valid at every look, so a test can be stopped as soon as the result is conclusive instead of at a planned sample size. Adverfly uses a mixture sequential probability ratio test (mSPRT) and only counts visitors whose conversion window has already elapsed, so an early look is never biased by half-observed visitors.
Sample Ratio Mismatch (SRM)	When the observed split between variants differs from the planned split by more than chance allows. It usually means delivery is broken rather than that one variant is winning, so a test flagged with SRM should be fixed and restarted, not read.
Revenue per Visitor	Total revenue divided by all enrolled visitors, non-buyers counted as zero, net of refunds and with cancelled orders excluded. The right metric when a change moves order value rather than order count. It needs far more traffic than conversion rate, because order values vary much more than a yes-or-no purchase does.
Holm Correction	A step-down adjustment applied when a test has more than one treatment variant. Comparing three variants against one control is three chances to find a difference by luck; the correction raises the bar for each so the overall false-positive rate stays at 5%.
Exposure	The record that a visitor actually received a variant, written once per test and pageview, control included. A visitor is only enrolled when the variant was really delivered: a redirect that could not happen must not count as treatment, because that visitor is about to see the control.

Marketing Mix Modeling

MMM concepts, saturation modeling, and budget optimization terms

Term	Definition
Marketing Mix Modeling (MMM)	A statistical modeling technique that quantifies the incremental impact of each marketing channel on revenue. Unlike attribution, MMM works with aggregated data and can account for external factors like seasonality and weather.
Base Revenue	The revenue a business would generate without any marketing spend — driven by brand equity, organic demand, and repeat customers. Also called baseline or organic revenue.
Incremental Revenue	The additional revenue directly attributable to marketing activities. Calculated as Total Revenue minus Base Revenue.
Seasonality Effect	The cyclical variation in revenue caused by recurring patterns like holidays, weather, or industry cycles. MMM isolates this so it doesn't get falsely attributed to marketing channels.
Revenue Decomposition	The process of breaking total revenue into its component parts: base sales, seasonality, and each channel's contribution. Typically visualized as a waterfall chart.
Hill Function	A mathematical function used to model saturation in marketing response curves. Formula: response = L * (spend^S) / (K^S + spend^S), where L is the maximum response, K is the half-saturation point, and S is the shape parameter.
Saturation Curve	An S-shaped curve showing the relationship between spend and response for a channel. As spend increases, each additional euro generates less return — this is the diminishing returns effect.
Diminishing Returns	The economic principle that each additional unit of spend in a channel generates progressively less revenue. Visualized as the flattening portion of the saturation curve.
Half-Saturation Point (K)	The spend level at which a channel reaches 50% of its maximum possible response. A key parameter in the Hill function — channels with higher K values need more spend to reach their potential.
Maximum Response (L)	The theoretical upper limit of revenue a channel can generate regardless of how much is spent. The asymptote of the saturation curve.
Shape Parameter (S)	Controls the steepness of the saturation curve. Higher values mean the curve transitions more sharply from growth to saturation. Also called the Hill coefficient. In Adverfly's implementation (LogisticSaturation), S is fixed at 1.0 — the saturation behavior is controlled entirely by the half-saturation point K.
Operating Point	The current position on a channel's saturation curve based on actual spend. Shows how close a channel is to saturation and how much room exists for scaling.
Historical ROI	The average return on investment over a period: total revenue attributed to a channel divided by total spend. Reflects overall efficiency but not marginal efficiency.
Marginal ROI (mROI)	The return generated by the next euro of spend in a channel. Calculated as the derivative of the Hill function. A channel can have a high historical ROI but a low mROI if it's near saturation.
Marketing Efficiency Ratio (MER)	Total revenue divided by total marketing spend across all channels. A blended efficiency metric that shows the overall return on marketing investment.
Budget Optimization	The process of reallocating budget across channels to maximize total revenue. Uses saturation curves to find the optimal spend level for each channel where the combined marginal returns are maximized.
R² (R-Squared)	The coefficient of determination — measures how well the MMM model explains the variance in actual revenue. Values range from 0 to 1, where 1.0 is a perfect fit. Values above 0.85 are considered strong for MMM.
MAPE	Mean Absolute Percentage Error — the average percentage deviation between actual and predicted values. A MAPE below 10% is considered excellent, below 20% is good.
Adstock	The carryover effect of advertising — the idea that a marketing message continues to influence consumers after it was shown. Modeled as a decay function over time.
Saturation Threshold	The spend level beyond which marginal returns drop significantly (typically below 20% of the initial mROI). Spending beyond this point yields rapidly diminishing returns.
Point of Diminishing Returns (PDR)	The spend level at which additional investment in a channel generates progressively less incremental revenue. MMM applies guardrails to prevent investment beyond the PDR, ensuring efficient budget allocation.
Data-Driven Budget Allocation	The process of using MMM outputs — saturation curves, marginal ROI, and revenue decomposition — to provide intelligent budget recommendations across every marketing channel, down to the campaign level.
Predictive Capabilities	The ability of MMM to forecast future marketing performance by understanding historical patterns, saturation dynamics, and seasonal trends. Enables scenario planning before committing budget.
Competitive Insights	MMM can incorporate competitor activities and market dynamics into its models, helping you understand how external competitive pressure affects your marketing performance.
Model Calibration	The process of refining MMM predictions using incrementality test results. When a geo lift or conversion lift test reveals a channel's true causal impact, those results recalibrate the model's channel coefficients, improving accuracy over time.
Feedback Loop	A continuous improvement cycle where incrementality results calibrate MMM, MMM identifies high-value tests, and MTA provides granular optimization signals. This iterative process makes the measurement system smarter with every decision.
Endogeneity	When a variable is determined inside the system being measured rather than outside it. Endogenous variables (CTR, CPC, clicks, impressions, ROAS, CPA) are results of the marketing process — including them in MMM creates circular reasoning. Only exogenous inputs (spend) and control variables (weather, holidays) may enter the model.
Bayesian Updating	The process of using yesterday's model posteriors as today's priors, creating a daily learning chain. This stabilizes the model (no wild daily swings), enables rapid adaptation to genuine changes, and supports anomaly detection when results diverge beyond 3 standard deviations.
PyMC Marketing	An open-source Bayesian MMM engine (Apache 2.0, maintained by PyMC Labs) built on PyMC. Supports hierarchical modeling across multiple content dimensions, geometric adstock, logistic saturation (Hill function), and full prior control for geo-test calibration.
Hierarchical Priors	A Bayesian modeling technique where channels sharing common attributes (e.g., same angle or format) pool information through shared prior distributions. This improves estimates for channels with limited data by borrowing strength from related channels.
Posterior Distribution	The updated probability distribution of a model parameter after observing data. In daily Bayesian updating, today's posteriors become tomorrow's priors — creating a chain that converges toward the true parameter values over time.
Control Variable	An external factor included in MMM to prevent false attribution. Weather, holidays, seasonality, and news sentiment are control variables — they affect revenue but are not marketing levers. Without them, the model would incorrectly credit weather-driven sales to marketing.
Geometric Decay (Adstock)	A specific adstock model where the advertising effect decays by a fixed proportion each day: effect(t) = spend(t) + decay × effect(t-1). Decay values range from 0.1 (search, short memory) to 0.9 (brand/YouTube, long memory).
ArviZ Diagnostics	A suite of Bayesian model diagnostic checks including R-hat convergence (must be < 1.05), divergent transitions (must be 0), and Effective Sample Size (must be > 400). These ensure the MCMC sampling has converged and the model results are reliable.

Tracking

Tracking and data collection terms

Term	Definition
Pixel	A small piece of JavaScript code installed on your website that tracks user behavior and conversions.
Event	A specific user action tracked by the pixel, such as a page view, add to cart, or button click.
Conversion	A completed desired action, typically a purchase or sign-up, that represents a successful outcome.
Pageview	An event fired when a user loads a page on your website.
Add to Cart	An event fired when a user adds a product to their shopping cart.
Initiated Checkout	An event fired when a user begins the checkout process.
Purchase	A conversion event fired when a user completes a transaction.
UTM Parameters	URL parameters used to track traffic source, medium, campaign, term, and content. Includes utm_source, utm_medium, utm_campaign, utm_term, utm_content.
First-Party Data	Data collected directly from your own website or app, owned and controlled by you.
Third-Party Data	Data collected by external parties and shared or sold to advertisers.
Cookie	A small file stored in a user's browser used to identify and track users across sessions.
Server-Side Tracking	Tracking that sends data from your server to analytics platforms, rather than from the user's browser.
CAPI	Conversions API. Server-side APIs provided by ad platforms (Meta, Google, etc.) to send conversion data directly from your server.
Data Bridge	A feature that syncs first-party pixel data back to ad platforms (Meta, Google, TikTok, Microsoft Ads) with one click. Improves platform optimization and match rates by sending enriched conversion signals server-side.
Adverfly Layer	The global window.adverfly object used to configure the pixel and pass customer context. Properties include store_id, store_currency, store_timezone, customer_id, and transaction_id.
Visitor Identification	The process of associating anonymous pixel events with a consistent visitor ID across sessions. Uses first-party cookies and server-side matching to maintain identity continuity.
Autocapture	Automatic tracking of user interactions (clicks, form submits, input changes, rage clicks, copy/cut) by the v3 pixel without requiring manual advPxl calls. Modeled on PostHog's open-source autocapture with client-side privacy filters that block password/hidden/file inputs and sensitive field names, and scrub credit-card, SSN and long-digit tokens from captured text.
Rage Click	An autocapture event ($rageclick) fired when the same element is clicked 3 or more times within 1 second — a signal of user frustration, broken UI, or an unresponsive element.
adv-no-capture	Opt-out marker for autocapture. Any element with class 'adv-no-capture' or attribute 'data-adv-no-capture' — including all its descendants — is excluded from autocapture entirely.
Events Table	The unified ClickHouse table where all autocapture events land. Designed with projections (proj_by_visitor, proj_by_source) to support multi-touch attribution via argMin(adv_source, dt) GROUP BY session_id instead of joining a separate sessions table.
URL Rule	A per-workspace mapping from a landing-page query parameter to attribution, configured under Channels. Example: ?ref=marie → source creator, campaign marie. Rules are evaluated by the tracking beacon when the visit is recorded, top to bottom, first match wins. When the campaign equals a creator's tracking code, the visit and its purchases count for that creator.
URL Params	Every query parameter of the landing page, stored with each event (url_params). Lets URL rules be applied to past traffic and shows which parameters a shop actually receives.
Dead Click (Autocapture)	An autocapture event ($dead_click) fired when a click on a link or button changes nothing at all within 2.5 seconds: no DOM change, no navigation, no scroll, nothing selected. The classic signature of a button that looks interactive and is not wired up. Only genuinely interactive elements are judged, and downloads, new-tab links, canvases, shadow-DOM components and form fields are exempt, because all of those work without changing the page. Capped at 20 reports and 60 watched clicks per page load. A page with a countdown or a rotating carousel looks alive constantly, so dead clicks are not detected there at all: the check prefers silence to fabricated reports. Session Replay detects dead clicks separately, inside a recording.
Bot Filtering	The pixel decides before its first request whether it is talking to a person, and a detected bot produces no events, no session, no A/B enrollment and no recording. Signals: a named crawler in the user agent, the "<name>bot/<version>" shape, navigator.webdriver, globals injected by Selenium, Puppeteer or Playwright, and the headless shape of zero plugins together with no languages. Deliberately conservative, because a false positive deletes a real customer's whole journey: a bare match on "bot" would delete every shopper on a CUBOT phone, and matching company names would delete the DuckDuckGo, Yandex, Baidu, Sogou and Perplexity browsers, so each crawler is named exactly. Extend it with extra_bot_patterns, switch it off with disable_bot_detection.
SPA Pageview	A pageview fired on a client-side route change in a single-page app, marked navigation: "spa". Only a path change counts unless spa_track_query is set, the same path is never counted twice in a row, changes are debounced by 300 ms, and the hook stops itself after 100 views in one page load. Every guard exists because an unguarded hook is the fastest way to produce tens of thousands of pageviews from one visitor. Opt-in: activate_spa_pageviews.
Core Web Vitals	Google's three page-experience measurements, collected from the browser's own performance entries and sent as one web_vitals event when the page goes away: LCP (Largest Contentful Paint, when the main content appeared, frozen at the first interaction, good under 2.5 s), CLS (Cumulative Layout Shift, the worst five-second window of shifts rather than the sum of the visit, good under 0.1) and INP (Interaction to Next Paint, the slowest real interaction, good under 200 ms). FCP and TTFB come along with them. There is no timer: both CLS and INP are defined over the whole visit, so flushing early would report a better score than the visitor got. Opt-in: activate_web_vitals.
Exception Event	An uncaught JavaScript error or rejected promise, captured as an exception event with type, message, truncated stack, source and path. The same error is sent once per pageview (fingerprinted on type, message and first stack FRAME, so two bugs sharing a message stay two bugs), at most ten distinct errors per pageview, and cross-origin "Script error." reports are dropped because they carry nothing actionable. The pixel filters out its own errors by stack and never by message, so a theme throwing "advPxl is not defined" still reaches us. Automatic capture is opt-in via activate_error_tracking; reporting one yourself with advPxl("exception", err) needs no flag, because writing the call is the opt-in.
Offline Queue	Opt-in (activate_offline_queue). An event that failed while the browser reported itself OFFLINE is parked in localStorage and replayed on the next pageview; bounded to 25 entries, 64 KB and 24 hours, dropping ordinary events before conversions. Only the offline case is queued, because ingestion does not deduplicate: an event that may already have arrived must never be sent again. It is opt-in because a replayed event is stamped with the time it arrives, so it can land on the wrong day and in a session with no ad click on it.

Customers & Segments

Customer data, segmentation, lifetime value, and loyalty

Term	Definition
Customer Segment	A group of customers defined by shared characteristics — purchase behavior, geographic location, acquisition channel, or custom rules. Segments are created using a rule builder and automatically updated as new customers match the criteria.
Segment Rule	A condition that determines segment membership. Rules consist of a dimension (e.g., country, first purchase date, acquisition channel), an operator (equals, greater than, contains), and a value. Multiple rules can be combined for precise targeting.
Segment Member	A customer who matches all rules defined for a segment. Membership is dynamic — as customers make new purchases or their attributes change, they may join or leave segments automatically.
Customer Lifetime Value (LTV)	The total revenue a customer is expected to generate over their entire relationship with your business. In Adverfly, LTV is calculated from pixel conversion data and can be segmented by acquisition channel, campaign, or customer segment.
RFM Analysis	A customer segmentation framework based on Recency (how recently a customer purchased), Frequency (how often they purchase), and Monetary value (how much they spend). In Adverfly, RFM dimensions are available as individual segment rules — you can build segments using recency, frequency, or monetary filters — but there is no standalone RFM scoring feature.
Cohort Analysis	Grouping customers by their acquisition date (or another shared event) and tracking their behavior over time. Reveals whether customer quality is improving or declining and how different acquisition channels produce different retention patterns.
Loyalty Program	A rewards system that incentivizes repeat purchases through points, tiers, and exclusive perks. Customers earn points on purchases, reviews, and referrals, and advance through membership tiers (Bronze, Silver, Gold, Platinum) to unlock increasing benefits.
Loyalty Tier	A membership level within a loyalty program. Higher tiers unlock better rewards (exclusive discounts, early access, free shipping). Tiers create aspirational goals that drive repeat purchases and increase customer lifetime value.
Redemption Rate	The percentage of earned loyalty points that customers actually redeem. A healthy redemption rate (40-60%) indicates the rewards are valuable enough to drive engagement without being too easy to earn.
Referral Program	A friends-refer-friends system where every customer gets a permanent personal referral code. When a friend's first qualifying order arrives, both sides are rewarded — either as loyalty credit or as coupon codes. Attribution is coupon-based: redeeming the personal code at checkout links the friend to the referrer, no cookies required.
Personal Referral Code	A durable, shareable code owned by one customer (e.g. K7Q2M9PX). Any number of different friends can redeem it once each. With coupon rewards enabled, the code doubles as a live Shopify discount so the friend saves at checkout while the redemption attributes the referral.
Self-Reported Attribution	Qualitative data collected through post-purchase surveys asking customers how they discovered your brand. Captures channels that digital tracking misses — word-of-mouth, podcasts, TV, creator mentions — providing a complementary signal to quantitative attribution methods.
Post-Purchase Survey	A short questionnaire shown to customers after checkout, typically asking 'How did you hear about us?' The most common method of self-reported attribution. In Adverfly, surveys are configured in the Surveys app and tracked via pixel events.
New vs. Returning Customer	Classification based on purchase history. A new customer has no previous transactions; a returning customer has purchased before. Tracked via the pixel's is_new_customer flag on conversion events. Critical for understanding whether campaigns acquire new customers or re-engage existing ones.

Advertising

Advertising platforms and campaign terms

Term	Definition
Campaign	A set of ad groups organized around a specific marketing objective or theme.
Ad Set / Ad Group	A collection of ads within a campaign that share targeting, budget, and schedule settings.
Ad / Creative	The actual advertisement shown to users, including images, videos, copy, and call-to-action.
Audience	A defined group of users targeted by your ads based on demographics, interests, behaviors, or custom data.
Lookalike Audience	An audience of users who share similar characteristics with your existing customers or website visitors.
Retargeting	Showing ads to users who have previously interacted with your website or app. Also called Remarketing.
Prospecting	Targeting new potential customers who haven't interacted with your brand before.
Frequency	The average number of times each user has seen your ad.
Reach	The total number of unique users who saw your ad.
Ad Fatigue	A decline in ad performance that occurs when users see the same ad too many times.
Creative Fatigue	Declining performance of a particular creative asset due to overexposure.
Creative Diversity	How evenly ad spend is spread across the creative dimensions the AI labels describe: ad angles, creative types, hook tactics, ad formats, target audiences, emotional triggers, funnel stages, proof types and production styles. Each dimension scores 0–100 (100 = spend spread evenly over every value the labels know, 0 = all spend on one value); the headline score on the Summary is their average. Based on labeled image and video ads that spent in the selected period; text-only search ads are left out, and no score is shown while labeled creatives carry less than 25% of ad spend.
A/B Testing	Comparing two versions of an ad or landing page to determine which performs better.
Bid	The maximum amount you're willing to pay for a specific action (click, impression, conversion).
Budget	The total amount allocated to spend on a campaign or ad set over a specific time period.
Optimization	The process of improving campaign performance through adjustments to targeting, bidding, creatives, or other settings.
Influencer	A person with a social media following who promotes products or services to their audience. In Adverfly, an creator record tracks a unique affiliate code/link.
Creator	A content creator or creator who has access to the Creator Dashboard to view their assigned affiliate codes and track performance.
Affiliate Code	A unique URL parameter or code assigned to an creator/creator to track traffic and conversions from their promotions.
Commission	The payment an creator/creator earns for each conversion or sale generated through their affiliate code. Can be percentage-based or a fixed amount.
Creator Dashboard	A dedicated portal in Adverfly where creators can view all their assigned affiliate codes, track performance, and see commission information across multiple workspaces.
Responsive Search Ad (RSA)	A Google Ads format where you provide up to 15 headlines and 4 descriptions. Google automatically tests combinations and serves the best-performing mix. In Adverfly, RSA assets are stored as text-only creatives (no image URL) with all headline and description variants in the properties field.
Performance Max (PMax)	A Google Ads campaign type that serves ads across all Google channels (Search, Display, YouTube, Gmail, Maps) from a single campaign. Assets are organized in Asset Groups rather than individual ads. In Adverfly, the campaign_id acts as the ad_id unit for spend attribution.
Asset Group	The creative container within a Performance Max campaign. Each asset group contains a mix of headlines, descriptions, images, logos, and videos. Google automatically assembles these into ads. Adverfly collects all unique assets from all asset groups within a campaign.
Responsive Display Ad (RDA)	A Google Display Network format where you provide multiple images, headlines, and descriptions. Google assembles and optimizes the combinations. In Adverfly, marketing images and YouTube video thumbnails are extracted and stored as individual assets.

AI & Automation

AI-powered features, automation, and intelligent optimization

Term	Definition
AI Visibility	A feature that tracks how your brand appears in AI-powered search engines and assistants like ChatGPT, Claude, Gemini, and Perplexity. Daily automated queries measure your Visibility Score, Mention Rate, Citation Rate, and Sentiment — giving you a quantitative view of your brand's presence in the AI-driven discovery layer.
Visibility Score	A score from 0 to 100 that measures how prominently your brand appears in AI system responses. Calculated based on whether the brand is mentioned, how early in the response, and whether it's recommended or just listed. Tracked daily to show trends over time.
Mention Rate	The percentage of AI queries that mention your brand in the response. For example, if you track 20 prompts and your brand appears in 12 responses, your Mention Rate is 60%.
Citation Rate	The percentage of AI mentions that include an explicit link to your website. A high Citation Rate means AI systems are not only mentioning your brand but actively driving traffic to you.
AI Suggestion	An AI-generated recommendation for a specific app. Suggestions are context-aware and can be accepted or regenerated for fresh ideas.
Recommendation	An AI-generated optimization suggestion based on your workspace performance data. Recommendations can be accepted (marking them for implementation), dismissed (acknowledging but not acting), or left pending for review. Accepted recommendations are tracked for outcome measurement.
Recommendation Digest	A periodic email summary of new and pending AI recommendations for your workspace. Ensures optimization opportunities are not missed even when the dashboard is not actively monitored.
Outcome Tracking	The measurement of accepted recommendation performance over time. After accepting an AI recommendation, outcome tracking monitors whether the suggested change improved the target metric.
Automation Rule	A configurable AI-driven action that triggers automatically when certain conditions are met. Rules can adjust budgets, pause underperforming campaigns, or send alerts based on thresholds defined in Workspace Settings.
Agent Automation	A scheduled or manually-triggered run of an AI agent — either a preset (Adverfly AI Digest, Recommendations, Creative Director) or a custom template-based agent you built in Agents & Skills. Managed in System & Automations alongside system schedules, with per-agent cron, timezone, enable/disable toggle, and a 'Run Now' button that can optionally email the result.
AI Automations Settings	A section in Workspace Settings where you configure AI-driven automation rules — set thresholds, enable or disable specific automations, and control how aggressively the AI optimizes on your behalf.
AI Chat	A conversational interface for querying your marketing data using natural language. Ask questions like 'What was my best-performing campaign last month?' and receive answers backed by your actual data across all measurement methods.
AI Forecasting	A feature that uses historical data and MMM model outputs to predict future revenue, conversions, and spend performance. Forecasts include credible interval bands showing the range of likely outcomes.
AI Labeling	Automated classification of creative assets using AI vision models. Assets are tagged with labels describing their format, visual execution, hook, marketing angle, awareness stage, emotion, tone, products, benefits, problems, audience, creator, copywriting framework and audio — enabling performance analysis by creative attributes. Runs per workspace once Auto-label is switched on, or on demand for single creatives.
Creative Library	Every creative of a workspace in one sheet — each ad with extracted media or spend — with the AI Labeling tags as columns. Your team can change any tag: a change replaces the AI's value for that one column and can be reset to it. Filter by tags, search names, copy and tags, export to CSV, open a creative to see its ad, transcript and full analysis, or select creatives and ask the AI about them.
Ad Deconstruction	A report per creative tag — visual format, hook, angle, awareness stage, emotion, tone, product, benefit, problem, audience, creator, copywriting framework or audio — that ranks the tag's values by a metric you pick (purchases, cost per purchase, ROAS, hook rate …) and shows how each changed against the period before, as cards, a table or a daily trend. A creative counts in every value it carries; values with under 10% of the top value's spend are never ranked.
Copywriting Framework	The structure an ad's script or copy follows, such as PAS (problem, agitate, solution), AIDA, BAB (before, after, bridge) or FAB (features, advantages, benefits). Detected by AI Labeling and shown as a column in the Creative Library.
Reflex	The self-optimizing traffic split. When a landing page has more than one variant, Reflex shifts a little traffic toward whichever variant performs better every night, and keeps shifting as performance changes. There is no test end and no decision to make: no variant ever drops below 10 percent, so a losing variant can recover when the market moves, and no variant moves more than 15 percentage points in one night. Reflex does nothing at all while the evidence is thin — below a traffic threshold, or for a variant that has not been shown often enough to judge. Every night's decision is written to the Logbook with the numbers behind it.
Landing Page Variant	An alternative version of a landing page. All variants ship inside the same published page and each visitor is assigned one, so the split can change without republishing anything the visitor can notice. Variants are what Reflex compares.
Execution Queue	An approval queue where AI-proposed actions (budget changes, campaign pauses, scaling decisions) await human review before execution. Actions are classified by risk tier (low/medium/high) and can be approved or rejected.
Marketing Angle	A strategic messaging approach used in creative assets — such as Social Proof, Urgency, Fear of Missing Out, or Testimonial. Adverfly automatically labels creatives by angle and tracks which angles perform best per channel and audience.
Geo-Lift Test	An incrementality test that compares performance across geographic regions with and without advertising. Measures the true causal effect of ads by analyzing conversion differences between treatment and control cities.
Candidate Finder	An analysis tool that scans your pixel conversion data to identify cities suitable for geo-lift testing. Evaluates stability (CV), trend, and spike ratio to find regions with clean, predictable baselines.
Own API Key	Available only with Enterprise: connecting your own Anthropic, OpenAI, Gemini or Perplexity key in Settings. Calls using your key are billed directly by your provider, without consuming Adverfly credits or an Adverfly markup.
Adverfly AI	Running AI features on Adverfly's provider keys instead of your own. Usage is charged at the provider's public list price plus a 30% service fee and appears as a single 'Platform API Usage' line on your monthly invoice. Opt-in per workspace in Settings.
List Price	The provider's published price per million tokens, which is the base every AI cost in Adverfly is calculated from. Each call is priced the moment it happens using the model that actually ran, including prompt-cache discounts.
Prompt Caching	Reusing an identical prompt prefix across calls. Providers charge a cached token at a fraction of the normal input price (Anthropic 10%, OpenAI 50%), which Adverfly accounts for separately so cached usage is never billed at the full rate.
Monthly AI Budget	A EUR ceiling on billable platform AI usage for a workspace. AI features pause once the month's usage reaches it, so an unattended workflow cannot produce an invoice you did not expect. Replaces the token limit once Adverfly AI is enabled.

Analytics

Analytics and reporting terminology

Term	Definition
Breakdown	Segmenting data by a specific dimension such as campaign, ad set, creative, date, or device.
Dimension	A categorical attribute used to segment and analyze data (e.g., campaign name, country, device type).
Metric	A quantitative measurement of performance (e.g., spend, revenue, conversions, ROAS).
Filter	A condition applied to data to show only records matching specific criteria.
Date Range	The time period for which data is displayed or analyzed.
Comparison Period	A secondary date range used to compare performance against the primary date range.
Trend	The direction of change in a metric over time (increasing, decreasing, or stable).
Cohort	A group of users who share a common characteristic, often used for analyzing behavior over time.
Funnel	A visualization of the stages users go through from initial interaction to conversion.
Customer Journey	The complete path a customer takes from first touchpoint to conversion and beyond.
Touchpoint	Any interaction between a customer and your brand, including ad views, clicks, website visits, and purchases.
Channel	A marketing platform or medium used to reach customers (e.g., Meta, Google, TikTok, Email).
Source	The origin of traffic or conversions, often referring to the specific platform or campaign.
Medium	The general category of marketing channel (e.g., paid, organic, referral, email).
Dashboard	A visual interface displaying key metrics and data for monitoring performance.
Report	A structured presentation of data and insights, often exported or scheduled for regular delivery.
Custom Report	A user-built report with custom metric selections, formula calculations, and visualization types. Custom reports can be saved, reused, and shared. The report builder supports AI-assisted natural language queries for faster report creation.
Report Chat	An AI-powered conversational interface within Custom Reports that lets you describe what you want to see in natural language. The system generates the appropriate query, metrics, and visualization automatically.
Dynamic Renderer	The engine that interprets a custom report's schema and renders the appropriate visualization — tables, charts, comparisons, or combined views — based on the selected metrics and format.
Workspace Usage	Metrics tracking how much of the platform's resources a workspace consumes — API calls, data volume, model runs, and feature usage. Used for monitoring and billing purposes.
Logbook	An audit trail that records all significant platform actions — model runs, budget changes, automation executions, recommendation decisions, and configuration changes. Provides accountability and historical context for every decision.
Session Replay	A frame-by-frame reconstruction of a real visitor session — the page as they saw it, plus every click, scroll and form interaction. Rebuilt from a DOM snapshot and a mutation stream captured by the pixel; input values are masked in the browser and never leave it.
Rage Click	A burst of three clicks, each landing within one second and 30 pixels of the one before it. A visitor clicking repeatedly is telling you an element looks interactive and is not responding — the strongest single signal of a broken interface. Counted once per burst, so six angry clicks read as two.
Dead Click	A click the page did not answer. A scroll or text selection counts as an answer only if it starts within 100ms; a DOM change, a form input or a navigation counts for a full 2.5 seconds, so a button that fires a request and renders a second later is not mistaken for a broken one; and a click that sends the visitor away within a second — a new tab, a download, a native dialog — is not counted at all. Unlike a rage click it happens once — a visitor quietly giving up rather than fighting.
Frustration Signal	Umbrella term for the behaviours that mark a session as going wrong — rage clicks, dead clicks, JavaScript errors and failed requests. Computed per session so recordings can be filtered by them instead of watched at random.
Replay Timeline	The readable index of a recording: landings, clicks, typing, errors, failed requests and pauses, each an offset you can jump to. Turns a replay from something you watch into something you can search.
Data API	The REST endpoints a workspace can PULL its own data from, authenticated with an Adverfly API key created in the Data Exports app. Serves report, ads, conversions, sessions and events.
Data Warehouse Export	The push counterpart to the Data API: Adverfly delivers the same datasets into BigQuery, Snowflake, Amazon S3 or Google Cloud Storage, as a one-time load or on a schedule. Set up together with the Adverfly team on request from the Integrations page, available from the Scale plan up.

Activity & Operations

Daily activity tracking across every app, plus the registry, digest, and threshold trigger that share the same plumbing

Term	Definition
Activity	A workspace-wide view that surfaces daily counts of meaningful events per app — new survey answers per day, new labels created, new vectors generated, new customers added to segments, conversions dispatched, and so on. Lives at /workspace/{id}/activity.
Activity Registry	The single declarative source of truth that maps each app to one or more daily metrics. Each entry specifies the source table, the date column to bucket on, an optional restricted-grammar filter, and an optional distinct-count column. Both the dashboard and the backend resolver read from this registry.
Activity Metric	A single (app, id, label, source) tuple in the Activity Registry. The backend turns one metric into one parametrized ClickHouse query that returns daily counts for the requested date range.
Activity Resolver	The generic backend resolver (getAppActivity) that compiles ClickHouse queries from the Activity Registry on the fly. There is no per-app Lambda — adding an app to the dashboard is a registry edit, not new code.
Activity Email Digest	An opt-in daily morning email summarizing the previous day's activity per app and ranking the top movers (biggest absolute and percentage changes). Configured in Outbox. Delivered at 08:00 by the same scheduler that runs the Adverfly AI, Recommendations, AI Visibility, and Automation digests.
activity_threshold Trigger	An Automations trigger type that fires when an Activity metric, summed over a configurable lookback window, satisfies an operator condition (>, <, >=, <=, =) against a numeric threshold. Cooldown prevents repeat firing within a window.
Activity Recommendations Provider	A Recommendations provider that compares yesterday's bucket of every Activity metric against the trailing 7-day average. Drops below 50% or spikes above 200% surface as activity_drop / activity_spike recommendations alongside MMM, Pixel, AI Visibility, and other patterns.
Top Movers	The shortlist of Activity metrics whose latest bucket changed most relative to the previous bucket — surfaced in both the dashboard delta indicators and the Activity Email Digest.
Shipment Status	The carrier state of a parcel belonging to an order: fulfilled (Shopify fulfillment created), pre_transit, in_transit, delivered, failure. Seeded from Shopify fulfillment tracking numbers and updated by polling the DHL Unified Tracking API (Integrations → Shipping & Logistics). Each transition is an append-only status event on the order and can fire the 'Shipment status changes' Workflow trigger — e.g. to email the customer on delivery.
Status Event	An append-only record of an entity's state change — order cancelled, shipment delivered, and so on. Original facts (like the transaction itself) are never mutated; current state is derived from the latest event. This is why cancelled orders disappear from revenue without their order rows being edited.

Upsell

Storefront upsell prompts — the mined opportunity types, the rule anatomy, and how a rule reaches a shopper

Term	Definition
Upsell	A prompt bound to a specific moment in the purchase flow whose purpose is order-value uplift — moving the buyer to a higher-value version of what they already want. Distinct from an Offer, which is an incentive redeemed at some later point and has no storefront runtime.
Upsell Rule	One configured prompt. Carries the placement it renders at, the trigger that decides when it fires, the offer it presents, the copy it shows, and a priority. Stored in the enduser upsell_rules table; only rules with status active are published.
Placement	Where the pixel renders a rule: product_page, cart, checkout, post_purchase, or thank_you. A rule belongs to exactly one placement; several rules can compete for the same one, and the lowest priority number wins.
Trade-up	A mined upsell type: buyers of the cheaper product A later end up owning the markedly pricier product B, so B is put in front of them while they are still on A. Requires a price step of at least 25% — below that it is a variant swap, not a trade-up.
Quantity Upgrade	A mined upsell type: a product customers keep re-buying at quantity 1 in short intervals. Selling the multi-pack up front collapses the repeat order into one larger first order.
Order Bump	A mined upsell type: a low-priced item that already rides along on a large share of multi-item orders. Priced relative to the store's own AOV rather than an absolute threshold, so it scales with the catalogue.
Post-purchase Upsell	A mined upsell type: a product bought within about two weeks of another one. The short window is what separates it from a cross-sell — at that distance the shopper would plausibly have completed the purchase in one go, so the thank-you page catches them first.
Uplift	The uniform ranking metric across all upsell miners: estimated monthly order-value uplift in minor units. It is what the Opportunities tab sorts by and what the card's footer stat shows.
Take Rate	Accepted prompts divided by prompts shown, for one rule. Computed from the upsell_accepted and upsell_impression pixel events.
Upsell Snapshot	The public JSON object (upsell/{workspace_id}/latest.json) holding every active rule of a workspace. Published on each write, cached for 60 seconds, and read by both the browser pixel and the public /upsell/v1/rules endpoint — which is why those two can never disagree.
Accept Handoff	The pixel cannot add a line item to an arbitrary storefront's cart, so on accept it calls window.adverfly.upsell_add_to_cart(offer, rule) if the merchant defined one, or fires an adverfly:upsell:accept event. The shop owns the cart mutation; everything else is handled by the pixel.