Glossary

What is A/B test? An A/B test compares two or more variants (A vs B) by randomly assigning traffic or users to each version and measuring which performs better

is the decline in click-through rate, conversion rate, or efficiency as the same audience sees the same creative repeatedly. Short-window metrics look fine until frequency rises and performance decays. Platforms may keep

What is anchor event? An anchor event is the first meaningful user action you use to clock a user-level pLTV score and send a value signal to ad platforms: inst

is Apple's iOS framework that requires user opt-in before apps can access the Identifier for Advertisers (IDFA) for cross-app tracking. Low opt-in rates drive

is the click, view, and campaign metadata that links an ad exposure to downstream events in your product or store: GCLID, gbraid/wbraid, fbc/fbp, ttclid, MMP click IDs, UTM parameters, and network campaign or ad set IDs.

What is attribution window? An attribution window is the lookback period an ad platform or analytics system uses to credit a conversion to a prior ad touchpoint

is mean revenue per order: total order revenue divided by order count over a period. Merchants and performance teams use AOV to track basket size and merchandising impact. AOV is a useful snapshot for ecommerce, but a

is when customers buy only the hero SKU (razor handle, coffee machine, core device) without the profitable attachments, refills, or consumables the business model assumes. The hero purchase counts as a conversion in ad p

is return on ad spend calculated across multiple channels, campaigns, or an internal attribution model, rather than inside a single ad platform's reporting. It answers how efficiently total paid media spend converts to r

is when a shopper orders multiple sizes, colors, or variants of the same product intending to try them at home and return what does not fit or match. Standard purchase tracking fires

is the control conversion setup you keep unchanged while testing a new value signal, bid strategy, or optimization change. It is the baseline against which you measure

are refunds driven by regret after an impulse purchase: the customer bought quickly, then reversed the decision before or after delivery. The order still counted as a conversion with positive value in ad platforms;

What is calibration? Calibration is how closely predicted lifetime value (pLTV) scores match realized customer outcomes in both rank (who is higher value) and s

What is causal modeling (pLTV)? Causal modeling asks whether a value signal or bidding change actually caused better outcomes, or whether correlated users would

A cohort-based LTV model estimates lifetime value for

Sum all revenue (or margin) from customers acquired in a cohort, divided by customers in that cohort. Example: January cohort of 1,000 buyers generates $120,000 by D90 → D90 cohort LTV = $120 per customer. Extend the win

Cohort maturity is the time an acquisition cohort needs before its revenue, retention, refund, and repeat-purchase patterns are stable enough to judge lifetime value and experiment outcomes. Performance teams use a

(Google Consent Mode v2) adjusts how Google tags and partner integrations behave based on user

Revenue minus variable costs directly tied to the order (COGS, shipping, payment fees, returns). Example: $80 order − $35 COGS − $8 fulfillment − $2 fees → $35 contribution margin. Use this margin in payback and LTV:CAC,

What is Conversion API? Conversion API is the generic industry term for server-side interfaces that send conversion and value events from your website, app, or

What is conversion lift study? A conversion lift study is a platform-run incrementality experiment (for example, Meta Conversion Lift or Google Conversion Lift)

has two layers. On ad platforms, it is bidding that maximizes

Conversion signal loss is the gap between conversions that happen in your business and conversions ad platforms can

is average ad spend required to generate one converting user or customer: total spend divided by attributed conversions. Platforms report CPA on the conversion event you optimize for (purchase, lead, trial start, install

is ad spend divided by attributed app installs over a period. It is the default efficiency metric for mobile user acquisition (UA) when platforms optimize toward install volume or install-linked events. CPI is a

What is custom conversion? A custom conversion is a platform-defined conversion rule built from standard events, URL rules, or parameters (for example, purchase

Divide total sales and marketing spend by new customers acquired in the same period. Example: $500,000 fully loaded S&M ÷ 2,500 new customers → CAC = $200.

Customer lifetime value (LTV) is the total economic value a customer generates over their relationship with your brand, usually measured as cumulative revenue, margin, or profit after discounts, refunds, and churn. Perfo

What is data readiness? Data readiness is whether your first-party data and data warehouse history are complete, consistent, and timely enough to model predicte

A data warehouse is the centralized system of record where your business stores

What is delayed conversions? Delayed conversions are outcomes that arrive after the platform can still learn from them for delivery and targeting . Repeat purch

is when customers learn to purchase only during promotions, waiting for codes, sales, or cart offers instead of buying at full margin. First promo conversions still fire as positive value in ad platforms; repeat revenue

is the retail pattern of limited-quantity or time-boxed releases that spike demand, urgency, and first-order conversions. Drop campaigns often deliver exceptional short-window ROAS and platform efficiency. Repeat value f

is when a paid subscriber cancels within the first 30–90 days after converting. The

What is early value users? Early value users are users who show high predicted or realized value soon after acquisition . They may pay quickly, engage deeply, o

Event Match Quality (EMQ) is Meta's 0–10 score for how effectively customer information on

What is experiment readout? An experiment readout is the structured analysis and decision phase after a test ends (or reaches predefined checkpoints). Teams com

describes the tradeoff between trying new options (exploration) and scaling known winners (exploitation). In performance marketing, platforms explore audiences and bids during

What is feedback loop (pLTV)? A feedback loop (pLTV) is when bidding changes who you acquire, which changes the data your model learns from . Value-based biddin

First-party data is customer and behavioral information you collect directly from people who interact with your brand: purchases, signups, app sessions, support contacts, and consented identifiers. For performance market

are refunds driven by honest mismatch: wrong size, poor fit, quality below expectations, or product unlike the photos and copy. The purchase still counts as a conversion with positive value in ad platforms;

is a period of no-charge access before paid billing begins. Performance teams often optimize on trial start as the conversion event because it fires inside the ad platform

is when a customer who received the product files a chargeback or payment dispute claiming non-receipt, unauthorized purchase, or item-not-as-described without legitimate cause. The order still counted as a conversion wi

GA4 Measurement Protocol is Google's server-side API for sending events into

What is geo experiment? A geo experiment splits geographic markets into test and control groups to measure incremental lift from a marketing change (budget, cha

is a buyer who purchases for someone else (holiday gift, care package, corporate gifting) and is not the end user of the product. The gift order counts as a conversion and positive value in ad platforms. Future

What is Google Ads Conversion API? The Google Ads Conversion API is Google's server-side path for sending web conversion and value data into Google Ads through

What is Google Ads Experiments? Google Ads Experiments is Google's in-platform split-test framework : you duplicate a campaign (or apply a draft change) and Goo

A holdout test is an experiment design where a

Subtract control-group conversions (scaled) from treatment-group conversions. Example: treatment 1,200 purchases, holdout 1,050 → incremental conversions = 150. Express as lift %: 150 ÷ 1,050 ≈ 14.3% incremental lift.

asks whether a campaign, bid strategy, or value signal caused

Divide incremental revenue from a test by incremental spend (or treatment spend net of baseline). Example: +$40,000 incremental revenue on +$10,000 incremental spend → incremental ROAS = 4.0. Do not use platform-attribut

What is leading indicators? Leading indicators are early behaviors that correlate with long-term value before revenue matures . Session depth, feature adoption,

in pLTV modeling means training features or labels include information that would not exist at the moment you score a user and send a value to an ad platform. Classic examples: future refunds, full subscription tenure, r

What is learning phase? Learning phase is the period after significant campaign or signal changes when an ad platform's delivery is unstable while its algorithm

What is lifetime value (LTV) reporting? LTV reporting is how teams measure and report customer lifetime value (LTV) after users are acquired: by cohort, channel

What is lookalike audience? A lookalike audience (Meta Lookalike, Google similar segments, TikTok lookalike) is a targeting audience the platform builds to rese

Divide matched server events by total server events sent in a period. Example: 8,000 matched purchases ÷ 10,000 purchase events sent to the ad network → 80% match rate. Definitions vary by network; use each platform's di

What is maturity window? A maturity window is the agreed time period after acquisition or conversion when cohort outcomes (repeat purchase, subscription renewal

Media mix modeling (MMM) is a statistical approach that estimates how

Meta Conversions API (CAPI) is Meta’s server-side channel for sending conversion and value events from your website, app, or backend into Meta Ads. It complements the browser pixel, improves data durability as cookies an

What is mobile measurement partner (MMP)? A mobile measurement partner (MMP) is a third-party platform (for example, AppsFlyer , Adjust , or Singular ) that att

What is model drift? Model drift is when predictor performance degrades as customer mix or behavior shifts . Promo calendars, catalog changes, privacy updates,

Multi-touch attribution (MTA) is a method for

measures how much recurring revenue you retain and expand from an existing customer cohort over a period, after churn and contraction, usually expressed as a percentage of starting ARR or MRR. NRR above 100% means expans

is the value you send to ad platforms (or score in pLTV models) after subtracting returns, refunds, discounts, and sometimes fees from gross purchase revenue. Gross order value fires at checkout; net revenue reflects wha

tracks the mix between first-time purchasers and returning buyers in revenue, orders, or acquisition reporting. Prospecting campaigns optimize on

is industry shorthand for running Meta

is when digital ads influence or drive in-store purchases that ad platforms cannot attribute to the online click or view. Online campaigns look underperforming or overperforming depending on which slice you trust. True c

are customers who complete one purchase (or one paid conversion) and never buy again within your measurement window. The first order fires as a positive conversion in ad platforms; the absence of repeat revenue only beco

Churney's delivery work observes a sharp drop in

is when a customer buys in bulk during a promotion or flash sale, then goes silent for months before the next order. The large first purchase counts as a high-value conversion in ad platforms. The long gap before

is the time it takes for a customer's cumulative

What is pilot? A pilot is a limited, time-bound rollout of a new capability (such as user-level pLTV , enhanced Conversion API values, or value-based bidding )

What is platform learning? Platform learning is how ad algorithms update delivery (audience, bid, placement, creative rotation) based on the conversion and valu

is return on ad spend as reported inside an ad platform (Meta, Google Ads, TikTok, etc.): platform-attributed conversion value divided by platform-reported spend. It reflects that network's attribution window, match rate

What is postback? A postback is a server-to-server callback that carries attribution or conversion data from a mobile measurement partner (MMP) or ad network to

Predicted lifetime value (pLTV) is the practice of forecasting customer economic value

is optimizing ad delivery and bidding for

What is prediction horizon? Prediction horizon is how far ahead a model forecasts value (D7, D30, D90, lifetime, etc.). Horizon choice trades early signal usabi

What is predictive events? Predictive events are conversion or value events sent to ad platforms before the full business outcome is known , carrying modeled sc

is when shoppers game discounts, coupon codes, referral credits, or stackable offers to maximize short-term savings without intent to build a profitable relationship. Conversions and order value fire in ad platforms at c

What is proxy metric? A proxy metric is a short-window conversion or value signal ad platforms use to learn when true economic outcomes are not yet visible. Exa

is auction-based ad buying where demand-side platforms bid on individual ad impressions in milliseconds, using audience, context, and performance data to set prices. Social and search platforms use RTB-style auctions und

is the share of orders or revenue returned after purchase, usually measured over a defined post-order window. Refunds are a

is when a customer abandons a intended replenishment routine (skincare stack, supplement protocol, pet food cadence) after the first kit or bottle. The first purchase counts as a strong conversion in ad platforms. The mi

is the share of subscribers who renew at billing period end (or equivalent continuation). Ad platforms often optimized earlier on install,

is the share of customers who make at least one additional purchase after their first order within a defined time window (for example, D90 or 12 months). It is a core

is the share of users or customers still active (or still paying) after a defined period: D7, D30, month two, etc. It is a core building block of

is systematic exploitation of a retailer's return policy: wearing or using products without intent to keep them, returning empty boxes or swapped items, or cycling purchases for resale. The initial purchase still counts

is revenue attributed to advertising divided by ad spend. A ROAS of 4.0 means $4 of attributed revenue per $1 spent. The number changes depending on whether you use

is when shoppers exploit free samples, trial sizes, or gift-with-purchase (GWP) offers to trigger a conversion without intent to buy full-price product again. The qualifying purchase or signup counts as a positive event

sends conversion and customer events from your backend (or tag server) to ad platforms and analytics tools, instead of relying only on browser pixels or SDK callbacks. It improves capture when cookies break, ad blockers

What is signal engineering? Signal engineering is the end-to-end discipline of designing, building, and maintaining the events and values ad platforms learn fro

What is signal freshness? Signal freshness describes how current a pLTV or value event is when an ad platform receives it relative to the user's anchor action (

What is signal health? Signal health is the overall quality of conversion and value data reaching ad platforms . It combines signal quality (accuracy, completen

Signal optimization is the practice of designing, testing, and refining the

What is signal orchestration? Signal orchestration is the coordinated system that turns first-party data into platform-learning inputs and proves they work: mod

What is signal quality? Signal quality is the accuracy, completeness, and timeliness of events used for bidding. High-quality signals send the right value, to t

What is signal transformation? Signal transformation is mapping raw predictions into platform-ready value magnitudes and timing . Models output scores; platform