Web analytics metrics reference: what each number means

A pageview is recorded when a page is loaded (or a virtual page is rendered in a single-page app). It is the oldest web-analytics metric and the easiest to misread: pageviews count loads, not people, and modern apps and prefetching can inflate or hide them. This page defines the metric and its caveats.

Bounce rate is the percentage of sessions with only one interaction. Its definition shifted: classic tools counted single-pageview sessions; GA4 derives it from engaged sessions instead. A high bounce rate is not inherently bad — for a single-answer page it can mean success — which is why context matters more than the number.

A session is a group of interactions from one visitor within a bounded time window. It starts on the first event and ends after a period of inactivity (commonly 30 minutes, configurable). The reset rules differ by tool — and historically Universal Analytics also restarted sessions at midnight and on a new campaign — so the same traffic produces different session counts in different products.

Engagement rate is the percentage of sessions that were 'engaged'. In GA4 an engaged session is one that lasted longer than a threshold (10 seconds by default), had a key event/conversion, or had at least two pageviews. Engagement rate is the inverse of GA4 bounce rate, and its threshold is configurable — so the number depends on a setting most people never check.

Average session duration is the mean length of sessions. Its core blind spot: duration is measured from the timestamps of events, so the time spent on the final page of a session — the exit page — typically counts as zero because no later event marks its end. This systematically undercounts real reading time, and GA4 replaced it with average engagement time, which is measured differently.

Scroll depth tracks how far down a page a visitor reaches, usually as percentage thresholds (25/50/75/90%) or a single 'reached bottom' event. GA4's enhanced measurement fires a scroll event at 90% vertical depth. It is a useful proxy for whether content was seen, but scrolling is not reading, and dynamic or short pages can trigger or suppress the event in misleading ways.

Click-through rate is clicks divided by impressions, expressed as a percentage. The catch is what counts as an impression: Google Search Console counts a result appearing in search, while ad platforms count an ad being served or viewed. Because the denominator differs by platform, CTR figures are only comparable within the same system — and a low CTR can mean wrong audience or simply low intent.

New vs returning classifies a visitor by whether the analytics tool recognizes them from a prior visit, usually via a client identifier. The split is fragile: cleared cookies, multiple devices, private browsing, and privacy-driven storage limits all make returning visitors look new. So the 'new' share is systematically overstated, and the dimension says more about identifier persistence than loyalty.

The users metric estimates how many distinct visitors a site had, but it actually counts distinct identifiers, not individuals. GA4 reports several user metrics — Total users, Active users (its headline), and New users — that mean different things. Because a person on three devices is three identifiers, and a cleared cookie is a new one, the count diverges from the real number of people.

Unique pageviews count how many sessions included at least one view of a given page, collapsing repeat views of the same page within one session into a single count. It was a Universal Analytics metric; GA4 does not report it and uses 'Views' (closer to raw pageviews) instead. Knowing the difference avoids comparing a de-duplicated UA number to a non-de-duplicated GA4 one.

Time on page estimates how long a visitor spent on a page, but classic tools infer it from the gap between consecutive pageview timestamps. That means the last page in a session — and every single-page session — records zero time, because there is no later event to subtract from. It systematically undercounts, which is why GA4 switched to foreground engagement time.

Average engagement time is a GA4 metric for how long your site or app was in the foreground and focused, averaged per active user or per session. Unlike classic session duration, it is measured directly from visibility — the timer pauses when the tab is hidden or backgrounded. That makes it a more honest attention signal, but it is GA4-specific and not comparable to older duration metrics.

Exit rate is the percentage of pageviews of a page that were the last pageview in their session — the point where visitors left the site. It is often confused with bounce rate, but they answer different questions: bounce is about single-interaction sessions, while exit is about where any session ended. A high exit rate matters most on pages that are not meant to be endpoints.

Pages per session (also pages per visit) is the average number of pageviews divided by sessions. It is read as a depth-of-engagement signal, but it is easily distorted: single-page apps fire virtual pageviews that inflate it, prefetching can add views nobody read, and a site designed to answer in one page will always look 'shallow'. It is comparable only against a page or site's own intent.

An impression counts a piece of content being shown — a search result, an ad, a social post. The trap is that 'shown' has no single definition: Search Console counts a listing appearing in results, ad servers count an ad being delivered, and the IAB/MRC viewable-impression standard requires a portion of pixels visible for a minimum time before it counts. Impressions are only comparable within one definition.

Event count is the number of events recorded. In an event-based model like GA4, almost everything — pageviews, scrolls, clicks, conversions — is an event, so the raw event count is large and mixes very different actions. Automatically collected and enhanced-measurement events add to the total without any explicit tagging, which is why event count must be read per event name, not in aggregate.

Events per session is the average number of events recorded per session. It can read as an interaction-intensity signal, but it is dominated by your measurement plan: enabling more event types (enhanced measurement, custom events) raises it without any change in visitor behavior. Useful for tracking interaction depth only when the set of tracked events is held constant.

Entrances count the number of times a page was the first pageview in a session — the doorway through which visitors entered the site. It differs from total pageviews because a page can be viewed mid-session without being an entrance. Entrances define which pages act as landing pages, and pairing entrances with bounce or engagement shows how well each doorway performs.

Active users is the count of distinct users with an engagement signal in a window. The window is the whole story: 1-day, 7-day, and 28-day active users (DAU/WAU/MAU) count different things, and GA4 reports rolling versions of each. They overlap rather than add up, and the DAU/MAU ratio is read as a 'stickiness' signal — but all of it inherits the identifier limits of any user count.

Sampling is when a tool computes a metric from a subset of sessions and scales the result up, instead of processing every event. It is used to return complex, high-volume queries quickly. GA4 applies sampling above per-query event thresholds in exploration reports, and the resulting numbers are estimates with a margin of error — small effects and rare segments are the least reliable under sampling.

Direct traffic is the bucket for sessions where no source could be determined — no referrer header and no campaign tags. It is meant for genuine type-ins and bookmarks, but in practice it absorbs stripped referrers, untagged links, app and email clicks, and redirects. A large direct share is therefore often a data-quality warning about lost attribution rather than a sign of strong brand recall.

Cost per click (CPC) is the amount an advertiser pays for each click, calculated as total cost divided by clicks. In an auction-based system the actual CPC is set by competing bids and ad quality, not just your max bid. CPC measures the price of a click, not its worth — a cheap click that never converts is not a bargain — so it is read alongside conversion and value metrics, never alone.

Cost per mille (CPM) is the cost of one thousand impressions — 'mille' is Latin for thousand. It is the standard pricing unit for awareness and display buying, where advertisers pay for exposure rather than clicks. CPM depends entirely on how an impression is defined (served vs viewable), and it says nothing about whether anyone clicked or converted, so it is an exposure-cost metric only.

Cost per acquisition (CPA), also called cost per action, is total cost divided by the number of conversions — the price of buying one desired action. It is more outcome-focused than CPC or CPM because it counts results, not clicks or impressions. But CPA is only as solid as the conversion definition and the attribution window behind it, and a low CPA is not the same as profit.

Return on ad spend (ROAS) is the revenue attributed to advertising divided by the cost of that advertising, usually expressed as a ratio or percentage. It answers 'how much revenue did each unit of ad spend bring back'. ROAS is not ROI — it ignores product margins and other costs — and its numerator depends entirely on the attribution model, so the same campaign can show very different ROAS under different rules.

Effective cost per mille (eCPM) expresses earnings as revenue per thousand impressions, regardless of how the inventory was actually priced. It lets a publisher compare a CPC deal, a CPA deal, and a CPM deal on one common scale by back-calculating what each earned per thousand impressions. eCPM is a normalization metric — it measures yield, not the contract terms — and it depends on the same impression definition issues as CPM.

Viewability rate is the percentage of measured ad impressions that qualified as viewable under an industry standard, rather than merely served. The IAB and MRC define a viewable display impression as at least 50% of the ad's pixels in view for at least one continuous second (two seconds for video). The rate exposes the gap between ads delivered and ads actually given a chance to be seen.

Video view rate is the share of video impressions that counted as a view, but the metric hinges on each platform's definition of 'a view'. A skippable in-stream (TrueView) view counts after 30 seconds or completion (or an interaction); auto-play, muted, and click-to-play videos each trigger views under different rules. Because the threshold varies, view rates are only comparable within one platform's definition.

Video completion rate is the percentage of video plays that reached the end. It is usually built from quartile progress events — playback milestones at 25%, 50%, 75%, and 100% — so completion rate is the 100% milestone divided by starts. It signals whether content holds attention, but auto-play, muting, and background tabs can inflate completions that no one actually watched.

Video play rate is the share of opportunities that resulted in a video play — typically plays divided by the number of times the video loaded or the page was viewed. It measures how often people start a video, but the metric is dominated by the denominator choice and by whether playback is auto or user-initiated, so play rate is meaningful only when those are held fixed.

Sessions per user is total sessions divided by the number of users — the average number of visits each distinct user made in the period. It reads as a return-frequency signal, but it inherits every weakness of the user count: when identifiers reset, returning visits split across several 'users', dragging sessions per user toward one and understating real loyalty.

Conversions per user is the total number of conversions (key events) divided by the number of users. It measures how many converting actions an average user took, which differs from conversion rate (conversions per session or per user as a percentage). Its value depends on which events are marked as conversions and on the same identifier limits as any user count, so the definition must be fixed to read it.

Average revenue per user (ARPU) is total revenue in a period divided by the number of users in that period. It is a standard unit-economics metric for subscription and consumer products, summarizing how much revenue each user generates. ARPU depends heavily on which users are in the denominator (all users vs active vs paying) and the length of the period, and it differs from ARPPU and lifetime value.

Average revenue per paying user (ARPPU) is total revenue divided by the number of paying users — it excludes everyone who did not spend. By isolating the paying base, ARPPU separates how much paying customers spend from how many people convert to paying. It is always at least as large as ARPU, and reading the two together reveals whether revenue is driven by spend depth or by the share who pay.

Monthly recurring revenue (MRR) is the normalized, predictable subscription revenue a business expects each month. Annual and multi-month plans are divided down to a monthly figure so the run rate is comparable. MRR is decomposed into new, expansion, contraction, and churned components, and it deliberately excludes one-off and usage-based charges — so it is a run-rate concept, not booked or recognized revenue.

Annual recurring revenue (ARR) is the annualized value of a business's recurring subscription revenue — the run rate it would earn over a year if nothing changed. It is closely tied to MRR (often MRR × 12) and is used for longer-horizon planning and contracts. ARR is a forward run-rate snapshot, not historical annual revenue, and like MRR it excludes one-off and usage charges.

Net Promoter Score (NPS) is a survey metric derived from one question — how likely you are to recommend, on a 0–10 scale. Respondents are bucketed into promoters (9–10), passives (7–8), and detractors (0–6), and NPS is the percentage of promoters minus the percentage of detractors, yielding a number from −100 to +100. It is simple and widely used, but the bucketing discards detail and ignores who answered.

Customer satisfaction score (CSAT) measures how satisfied respondents are with a specific interaction, product, or experience, usually from a short rating scale. It is commonly the percentage of responses at or above a 'satisfied' threshold (for example the top two boxes of a five-point scale). CSAT is moment-specific and threshold-dependent, so the same data can yield different CSAT values under different scoring rules.

Customer effort score (CES) measures how much effort a customer had to expend to complete a task — resolving an issue, making a purchase, finding an answer. It is captured by an agree/disagree statement about ease, scored on a scale, and lower effort is treated as better. CES targets friction specifically, which makes it different from satisfaction (CSAT) or recommendation likelihood (NPS).

Page value estimates the average monetary value of a page by crediting it with revenue from transactions (and goal values) that occurred in sessions where the page was viewed before the conversion. It is a way to surface which content contributes to revenue, not just which page closes the sale. Page value is an attribution-style estimate, so it shares the assumptions and limits of crediting upstream pages.

Event value is a numeric value attached to an event via a value parameter, letting analytics sum the worth of actions that are not direct purchases — a lead, a sign-up, a key interaction. It turns counted events into an aggregable monetary or proxy figure. The catch is that event values are assigned by the implementer, so inconsistent or arbitrary values quietly distort every total and comparison built on them.

Organic vs paid traffic share is the proportion of sessions classified into organic channels (unpaid search, referral, social) versus paid channels (search/display/social ads). It comes from channel-grouping rules that read the referrer and campaign parameters. The split is only as accurate as that classification: untagged paid links can land in organic, and stripped referrers fall into direct, so the share reflects tagging as much as reality.

Branded vs non-branded search share is the proportion of search clicks or impressions from queries that contain your brand name versus those that do not. It separates demand you already earned (people searching your name) from discovery (people finding you for a topic). The split is usually built by filtering Search Console queries, and it is limited by query redaction and by the fuzzy boundary of what counts as 'branded'.

Gross merchandise value (GMV) is the total monetary value of merchandise sold through a platform over a period, typically measured before subtracting platform fees, refunds, returns, cancellations, or discounts. It is a marketplace and e-commerce headline figure, but its meaning depends entirely on the inclusion rules a company chooses, so two GMV numbers are rarely comparable without reading the definition.

Add-to-cart rate measures how often shopping activity leads to an item being added to the cart. Depending on the denominator it can be add-to-carts per session, per user, or per product-detail view (cart-to-detail rate). GA4 exposes related ratios in its ecommerce reports. The metric is an early funnel signal that sits well before purchase, so it must be read alongside checkout and purchase steps.

Cart-to-detail rate divides the number of add-to-cart events by the number of product-detail-page views for the same items. By anchoring the denominator to product views rather than sessions, it measures how effectively a product page converts an interested viewer into a cart add, independent of how much general traffic the store receives. GA4's ecommerce engagement reporting exposes this ratio.

Checkout completion rate measures the share of started checkouts that end in a purchase. It is computed as purchase events divided by begin_checkout events over the same window. As the inverse of checkout abandonment, it isolates the final stage of the e-commerce funnel — payment, shipping, account, and form friction — from discovery and cart behavior earlier in the journey.

Purchase rate measures how often shopping activity ends in a purchase, computed as purchase events divided by a base such as sessions or active users. Unlike checkout completion rate, which is scoped to started checkouts, purchase rate spans the whole journey from arrival to order. Its meaning depends on the denominator, so the base must be stated for the number to be comparable.

Refund rate measures how much of what was sold is given back to buyers. It can be computed by count (refunded orders ÷ orders) or by value (refunded amount ÷ revenue), and partial refunds make these two diverge. GA4 has a dedicated refund event so refunds can be tracked rather than guessed. The metric is a quality and margin signal that erodes GMV and recognized revenue.

Repeat purchase rate is the proportion of customers who place more than one order within a defined window. It is a loyalty and retention signal distinct from session-level conversion: it counts people who came back to buy again. Because it depends on the time window and on identifying the same customer across orders, the cohort definition and identity rules govern what the number actually means.

Items per order (average basket size) is the total quantity of units sold divided by the number of orders over a period. It describes how many items a typical order contains, independent of price. Together with average order value it decomposes revenue per order into quantity and price effects, which is why merchandising and bundling work is often judged on basket size rather than value alone.

Conversion value is the monetary worth attached to a conversion or key event. In GA4 it comes from the value parameter (with currency) sent on events such as purchase or generate_lead, and it feeds revenue, ROAS, and page-value calculations. Because it is whatever you assign — a real order total or an estimated lead worth — its reliability depends entirely on consistent, correctly scoped tagging.

The DAU/MAU stickiness ratio divides daily active users by monthly active users. It approximates how many days in a month a typical active user shows up, making it a habit and engagement signal for apps and products. Its value hinges entirely on how 'active' is defined and on the DAU/MAU averaging method, so the underlying definitions must travel with the number.

Crash-free users rate is the percentage of active users who used an app over a period without hitting a crash. It is a mobile stability metric, distinct from crash-free sessions, which counts at the session level. Because one user can have many sessions, the two rates differ: a single crashing user can drag the user-based rate while barely moving the session-based one.

App retention rate measures how much of an install or first-use cohort comes back after a number of days. Definitions vary: day-N retention counts users active exactly on day N, while rolling or range retention counts users active on or after day N. Because these methods produce different curves from the same data, the retention definition must be stated for the number to mean anything.

Uninstall rate is the proportion of app installs that are subsequently removed from devices. It is a direct churn signal for mobile apps, but it is notoriously hard to observe precisely: mobile platforms restrict how and when removals are reported, so uninstall data is often delayed, aggregated, or modeled rather than exact. It is best read as a directional trend alongside retention.

Real user monitoring (RUM) measures web performance from actual visitors' browsers in the field, as opposed to synthetic lab testing in a controlled environment. Its headline metrics are the Core Web Vitals — Largest Contentful Paint, Interaction to Next Paint, and Cumulative Layout Shift — collected via the browser's performance APIs. Field data reflects real devices and networks, so it varies far more than lab numbers.

Apdex (Application Performance Index) is an open standard that condenses response-time measurements into a single 0–1 satisfaction score. Each sample is classified as satisfied (at or under a target T), tolerating (between T and 4T), or frustrated (over 4T). The score is satisfied plus half of tolerating, divided by total samples. It makes performance legible as one user-centric number, but the chosen T defines it.

Error rate is the proportion of requests, page loads, or interactions that fail over a period. It comes in several flavors — server-side HTTP error rate (5xx/4xx share), client-side JavaScript error rate, and failed-interaction rate — each with its own numerator and denominator. As a RUM and reliability metric it signals broken experiences, but only when the failure definition and base are stated clearly.

Return frequency measures how often a given user returns within a period, expressed as visits, sessions, or purchases per user. It is an engagement and loyalty signal that captures habit rather than reach, and it is the 'F' in RFM analysis. Because it averages repeat behavior across a base, the window and the unit of return (visit versus purchase) determine what the number describes.

Recency measures how long it has been since a user last did something meaningful — visited, engaged, or purchased. Lower recency (a more recent action) is generally associated with higher likelihood of returning, which is why recency is the leading dimension of RFM analysis. It is a per-user time measure, so it is summarized across a base via distributions or segments rather than a single average.

RFM is a customer-segmentation framework that scores each customer on three dimensions — recency (how recently they acted), frequency (how often), and monetary value (how much they spent) — typically by ranking customers into quantiles per dimension. The combined score sorts customers into segments such as best customers, lapsing, or new. It is a concept built from three underlying metrics, not a single measured quantity.

Blended customer acquisition cost (CAC) divides total acquisition spend over a period by the total number of new customers acquired, mixing paid and organic together. It differs from paid CAC, which divides only paid spend by only paid-acquired customers. Blended CAC answers 'what did each new customer cost on average overall,' while paid CAC isolates channel efficiency — both are valid, for different questions.

Contribution margin is revenue minus the variable costs of producing it — the money each unit or order contributes toward fixed costs and profit. It can be expressed per unit, in total, or as a ratio of revenue. Because it isolates variable costs, it differs from gross margin (which uses cost of goods sold) and is the figure used to reason about scaling, pricing, and break-even.

The SaaS quick ratio divides recurring revenue gained — new plus expansion MRR — by recurring revenue lost — churned plus contraction MRR — over a period. It summarizes how efficiently a subscription business grows: a value above 1 means more revenue is being added than lost. It is a momentum and efficiency signal that sits on top of MRR movements rather than a standalone measured quantity.

Net revenue retention (NRR), also called net dollar retention, measures how much recurring revenue a fixed cohort of customers produces at the end of a period versus the start, counting upgrades (expansion) and subtracting downgrades (contraction) and churn — but excluding revenue from brand-new customers. Above 100% means the cohort grew on its own. It is a subscription-economics convention, and definitions vary by vendor.

Gross revenue retention (GRR) measures how much of a cohort's recurring revenue survives churn and downgrades over a period, with expansion excluded. Because upgrades cannot count, GRR is capped at 100% — it can only stay flat or fall. It isolates raw stickiness, separate from a company's ability to upsell. GRR is a subscription convention and the exact construction varies by vendor.

Expansion revenue is the additional recurring revenue earned from existing customers within a period — through plan upgrades, added seats, usage growth, or cross-sell — without acquiring anyone new. It is the positive component that lifts net revenue retention above gross. Isolating it cleanly from new-customer and reactivation revenue is the main measurement challenge, and the categorization is a vendor convention.

Logo churn rate is the percentage of customers — 'logos', meaning whole accounts — that cancelled during a period, counted by number of accounts rather than by revenue. It differs from revenue churn because each account counts equally regardless of size. A business can have low revenue churn but high logo churn if it loses many small accounts, or the reverse. It is a subscription convention; the window varies.

The Rule of 40 is a heuristic for software businesses: add the revenue growth rate (percent) to a profitability margin (percent), and the sum is the score. The convention holds that a healthy company should reach roughly 40, trading growth against profit — fast growth can justify thin margins and vice versa. It is an industry rule of thumb, not an accounting standard, and the choice of margin varies.

The burn multiple divides net cash burned in a period by the net new annual recurring revenue (ARR) added in the same period. It answers: how many dollars of cash did the company spend to add one dollar of new recurring revenue? A lower multiple means more efficient growth. It is a startup-finance convention popularized as a capital-efficiency gauge, not an accounting standard.

The SaaS magic number relates new recurring revenue to the sales-and-marketing spend that produced it. A common form divides the annualized increase in recurring revenue in a quarter by the prior quarter's sales-and-marketing cost. It estimates how much new annual recurring revenue each dollar of go-to-market spend generates. It is a venture-finance convention with several formula variants.

Activation-to-paid conversion rate is the percentage of users who reached an activation milestone and then became paying customers within a window. It is narrower than signup-to-paid because it conditions on activation — users who experienced the product's core value first. The metric depends entirely on how 'activated' is defined, which is a per-product choice, so it is a convention rather than a standard.

Lead-to-MQL conversion rate is the percentage of captured leads that meet a marketing-qualified-lead (MQL) bar — typically a scoring or fit threshold marketing applies before passing a lead toward sales. It measures top-of-funnel quality. Because the MQL definition is set internally (fit criteria, scoring rules), the rate is an organization-specific convention, not a standardized metric.

MQL-to-SQL conversion rate is the percentage of marketing-qualified leads (MQLs) that sales accepts and promotes to sales-qualified leads (SQLs). It measures alignment at the marketing-to-sales handoff: how often marketing's 'qualified' leads meet sales' bar. Because both MQL and SQL are defined internally, the rate is an organization-specific convention rather than a standardized figure.

Cost per lead (CPL) is marketing spend divided by the number of leads generated in a period. It measures the cost of capturing a contact — a form fill, a download, an inquiry — before any qualification or sale. It sits earlier in the funnel than cost per acquisition (CPA), which counts paying customers, and it says nothing about lead quality, so it must be read with downstream conversion rates.

Lead velocity rate (LVR) is the percentage growth in qualified leads from one month to the next. It is a forward-looking pipeline indicator: because today's qualified leads become tomorrow's revenue, a rising LVR signals future growth ahead of bookings. It is a go-to-market convention that depends on a consistent definition of 'qualified lead' to be meaningful month over month.

Take rate is the percentage of gross merchandise value (GMV) that a marketplace retains as its own revenue — fees, commissions, and charges — rather than passing to sellers. It is the core monetization ratio for marketplaces: revenue divided by GMV. The headline fee schedule and the effective take rate often differ once discounts, subsidies, and mixed fee types are netted out. It is an industry convention.

GMV per buyer divides total gross merchandise value by the number of active buyers in a period. It measures how much the average buyer transacts on a marketplace, a core demand-side health signal. As an average it is sensitive to skew — a few high-spend buyers can pull it up — so it is best read with the buyer distribution and the definition of 'active buyer', which is a per-platform convention.

Marketplace liquidity measures how reliably a two-sided marketplace matches supply and demand. Common operational definitions include the share of listings that sell within a period, or the share of buyer requests that get fulfilled. High liquidity means participants reliably find a match; low liquidity drives them away. There is no single formula — liquidity is defined per marketplace, so it is an industry convention.

Page RPM (revenue per mille) is estimated earnings per thousand pageviews: total revenue divided by pageviews, times 1,000. It is the publisher-side companion to CPM — where CPM is what an advertiser pays per thousand impressions, RPM is what a page earns per thousand views, blending fill, viewability, and multiple ad units. Google AdSense documents the calculation; it is a derived metric, not a guaranteed rate.

Session RPM is estimated revenue per thousand sessions: total revenue divided by sessions, times 1,000. It normalizes earnings to visits rather than pageviews, so it rewards monetizing an entire session — across every page a visitor sees — rather than a single page. It became prominent as ad programs shifted toward session-based reporting. The session definition matters, and it is a publisher convention layered on standard RPM.

Ad fill rate is the percentage of ad requests that were answered with an ad — ads served divided by ad requests. A request that returns no ad ('unfilled') earns nothing, so fill rate directly gates a publisher's revenue: low fill means inventory went to waste. Google Ad Manager reports fill rate from match rate and coverage. It is a documented programmatic metric, though exact request/impression accounting varies by platform.

Subscription conversion rate, in a publisher or media context, is the percentage of a chosen audience — visitors, registered users, or paywall encounters — who convert to paid subscriptions in a period. It measures how effectively a reader relationship turns into recurring revenue. The metric hinges on which denominator is used, and because publishers choose that base differently, it is a convention rather than a standardized rate.

ARPDAU (average revenue per daily active user) is total revenue on a day divided by that day's daily active users. It is a high-frequency monetization signal common in mobile apps and games, where revenue from ads and in-app purchases is averaged across the active base each day. Because it is daily, it reacts fast to changes — but it depends entirely on how a 'daily active user' is defined, which is a per-product convention.

K-factor, or viral coefficient, measures how many new users each existing user brings in: the average number of invitations a user sends multiplied by the rate at which those invitations convert to new users. A K of 1 means each user replaces themselves through referral; above 1 implies self-sustaining viral growth. It is a growth convention adapted from epidemiology, with the invite and conversion definitions set per product.

Day-N retention measures the percentage of a user cohort that returns on a specific day after first use — D1, D7, and D30 being the common checkpoints. It is a core mobile and product retention curve. The subtlety is that 'returned on day N' has three competing definitions — classic (exactly day N), range (by day N), and rolling — which produce different numbers from the same data, so the definition must always be stated.

Time to first byte (TTFB) measures the interval between the browser starting a navigation request and receiving the first byte of the server's response. The Performance Timeline derives it from responseStart minus the request's start time, so it folds in redirect, DNS, connection, TLS, and server processing time. Because nothing can render before bytes arrive, a slow TTFB delays every downstream metric, which is why web.dev treats it as a diagnostic for First Contentful Paint and Largest Contentful Paint.

First Contentful Paint (FCP) measures the time from navigation start to when the browser first renders any DOM content — text, an image, a non-white canvas, or SVG. The Paint Timing API exposes it as the first-contentful-paint entry, and web.dev treats it as the moment a visitor first sees that something is happening. It precedes Largest Contentful Paint, which marks the largest element, so the two answer different questions about perceived load.

Largest Contentful Paint (LCP) reports the render time of the largest image or text block visible in the viewport, measured from when the page starts loading. It is one of Google's Core Web Vitals, exposed through the Largest Contentful Paint API, and the candidate element can change as larger content paints — the final value is taken at the last candidate before user interaction. web.dev breaks LCP into TTFB, resource load delay, load duration, and render delay to localise the bottleneck.

Cumulative Layout Shift (CLS) measures the largest burst of unexpected layout shifts during a page's lifetime. Each shift contributes a layout-shift score equal to the impact fraction times the distance fraction, and the Layout Instability API reports those entries. To avoid penalising long-lived pages, CLS is the maximum sum within a session window of shifts rather than a running total, which is why a stable page that occasionally moves can still score low.

Interaction to Next Paint (INP) measures a page's responsiveness by observing the latency of every click, tap, and key press during a visit and reporting a representative high value — close to the worst. Latency spans from the input to the next frame the browser paints. INP became a Core Web Vital in March 2024, replacing First Input Delay, because it captures the full processing-plus-render cost across all interactions, not just the delay of the first one.

Total Blocking Time (TBT) measures how long the main thread was blocked between First Contentful Paint and Time to Interactive. For each task longer than 50 milliseconds, the portion above 50ms counts as blocking time, and TBT is the sum of those portions. It is a lab metric — Lighthouse reports it — and web.dev treats it as a proxy for field responsiveness because high TBT usually predicts a poor Interaction to Next Paint.

Time to Interactive (TTI) measures how long it takes a page to become reliably interactive — visually rendered, with event handlers registered, and responding to input quickly. The definition looks back from a five-second quiet window (no long tasks, limited network requests) to the last long task before it. TTI is a lab metric from Lighthouse; web.dev now steers teams toward TBT and the field metric INP, since TTI is sensitive to single long tasks.

Speed Index measures how quickly the contents of a page are visibly populated during load. Instead of marking one moment, it records the visual completeness of the viewport frame by frame and integrates the un-painted area over time, so a page that fills in steadily scores better than one that snaps in late. It is a lab metric computed by Lighthouse from a video capture of the load, expressed in milliseconds where lower is faster.

Page weight is the total number of bytes transferred to load a page — the sum of the compressed transfer sizes of the document and every sub-resource it pulls in. The Resource Timing API exposes transferSize per resource, which differs from the uncompressed decodedBodySize. Page weight correlates with load cost on slow or metered connections, and breaking it down by resource type shows whether images, scripts, fonts, or media dominate.

Request count is the number of network requests a page issues to load — every HTML document, stylesheet, script, image, font, and API call. The Resource Timing API lists each as a PerformanceResourceTiming entry. The raw count matters less than which requests sit on the critical rendering path and how they contend for connections, since modern protocols multiplex but third-party and render-blocking requests still gate the experience.

Marketing efficiency ratio (MER) is total business revenue divided by total marketing spend over a period, across every channel at once. Unlike per-channel return on ad spend, it claims no attribution: it asks how much revenue the whole marketing budget produced, including organic and brand effects. As an industry convention it is read as a trend over time, and pairs with channel-level ROAS rather than replacing it.

Advertising cost of sales (ACoS) is ad spend divided by the sales attributed to those ads, expressed as a percentage — the inverse of return on ad spend. It is the standard efficiency metric in retail-media platforms such as Amazon Ads, where it measures what fraction of attributed revenue was spent on advertising. A lower ACoS means a smaller cut of sales went to ad cost, but break-even depends on a product's own margin.

Total advertising cost of sales (TACoS) divides ad spend by total revenue — organic plus ad-attributed — rather than by attributed sales alone. By using all revenue as the denominator, it reveals how ad spend relates to the whole business, capturing the organic halo that advertising can build over time. A TACoS that falls while sales rise suggests advertising is increasingly leveraging organic demand rather than carrying every sale itself.

Incremental return on ad spend (iROAS) divides the incremental revenue advertising caused — the lift over a control group — by ad spend. Unlike attributed ROAS, which credits every conversion an ad touched, iROAS isolates causation using experiments such as geo holdouts or ghost-ad tests. It answers a different question: not how much revenue was attributed to ads, but how much would not have happened without them.

Reach and frequency are paired media metrics: reach is the number of unique people who saw an ad at least once, and frequency is the average number of times each reached person saw it. Together they decompose total impressions, since impressions equal reach times frequency. Platforms expose both in reach-and-frequency reporting and let advertisers set frequency caps to limit how often one person is shown the same ad.

Share of search is the volume of searches for one brand divided by the total search volume for all brands in its category, as a percentage. Computed from search-volume tools or Google Trends-style indices, it is used as a leading, attribution-free indicator of relative brand demand. It measures interest expressed as queries, not sales, so it complements rather than substitutes for share-of-market figures.

Email open rate is the number of opens divided by the number of emails delivered, as a percentage. It is measured by a tiny tracking pixel that loads when the message is viewed. Since Apple's Mail Privacy Protection began pre-fetching images regardless of whether a person opened the email, pixel-based opens are inflated and unreliable, so open rate is now read as a soft signal rather than a precise engagement measure.

Click-to-open rate (CTOR) is the number of unique clicks divided by the number of unique opens, as a percentage. By using opens rather than deliveries as the denominator, it isolates how compelling the email's content and calls to action were among people who actually opened it. Because the opens denominator is now inflated by privacy-driven image pre-fetching, CTOR has become harder to trust and is read alongside raw click rates.

Email unsubscribe rate is the number of recipients who opted out divided by the number of emails delivered for a send, as a percentage. It signals when content, frequency, or relevance is pushing people to leave the list. Bulk-sender requirements now mandate a working one-click unsubscribe, so a clear opt-out path is expected — and a very low rate can hide people who instead mark mail as spam.

Email deliverability rate is the share of sent emails that were accepted by receiving servers — delivered divided by sent, the inverse of the bounce rate. But 'delivered' only means not bounced; it does not say whether mail reached the inbox or the spam folder. True inbox placement depends on authentication (SPF, DKIM, DMARC), sender reputation, and engagement, which is why deliverability is read with placement and complaint signals.

Spam complaint rate is the number of recipients who marked a message as spam divided by emails delivered, as a percentage. Mailbox providers report it through feedback loops, and it is one of the most damaging signals a sender can accumulate. Major providers' bulk-sender requirements set a complaint-rate threshold senders must stay under, making it a compliance metric, not just an engagement one.

Email list growth rate measures how a subscriber list changes over a period: new subscribers minus unsubscribes and spam-complaint-or-bounce removals, divided by the total list size, as a percentage. It is a net figure — gross signups alone hide churn — and its value depends on consent quality, since a list that grows through unconsented or purchased contacts inflates the number while harming deliverability.

Feature adoption rate is the share of eligible users who used a specific feature in a period — users who used it divided by users who had access to it. It tells a product team whether a capability is reaching its audience. The number hinges on two choices: who counts as eligible (the denominator) and what counts as 'used' (one click, or a meaningful completion), so the same feature can show very different adoption depending on definitions.

Time to value (TTV) measures how long it takes a new user to reach their first meaningful outcome with a product — the moment the product demonstrably helped them. It is usually measured from signup to a defined value milestone (the 'aha' or activation event). The metric's usefulness depends entirely on choosing a milestone that genuinely represents value, since a shorter TTV to a trivial event tells you nothing.

Daily active accounts (DAA) counts the number of distinct accounts — organisations, teams, or workspaces — that took a qualifying action on a given day. It is the account-level analogue of daily active users, and matters for B2B and multi-seat products where the customer is an account, not a person. DAA depends on defining 'active' (any activity, or a meaningful action) and on correctly grouping users under their account.

Email list churn rate is the share of subscribers a list loses over a period — removals (unsubscribes, hard bounces, complaint-driven purges) divided by the list size. It splits into transparent churn (visible opt-outs and bounces) and opaque churn (subscribers who silently stop engaging without leaving). A low transparent churn rate can mask a large opaque segment of dead addresses that quietly erodes deliverability.

Net new MRR is the change in monthly recurring revenue over a period, built from four movements: new MRR from new customers, expansion MRR from upgrades, contraction MRR from downgrades, and churned MRR from cancellations. Net new MRR = new + expansion − contraction − churn. It distils a month of recurring-revenue movement into one figure while keeping the components visible so the source of growth or decline is clear.

Expansion MRR rate is the expansion revenue earned from existing customers in a period — upgrades, add-ons, and seat increases — divided by the MRR at the start of the period, as a percentage. It isolates the growth that comes from deepening relationships with current customers, separate from new-customer acquisition. A strong expansion rate indicates a product whose value grows with usage, often the engine behind net revenue retention above one hundred percent.

Gross margin is revenue minus the cost of goods sold (COGS), divided by revenue, as a percentage. It shows how much of each revenue dollar remains after the direct cost of delivering the product, before operating expenses like sales and R&D. For software, what belongs in COGS — hosting, third-party APIs, support, payment fees — is a judgement call that materially changes the margin, so the definition must travel with the number.

Gross profit margin in retail is gross profit — net revenue minus the cost of goods sold — divided by net revenue, as a percentage. It measures how much of each sales dollar is left after the cost of the merchandise itself, before operating expenses. It is distinct from markup (profit over cost) and is reduced by discounts and returns, which is why it is computed on net rather than gross sales.

Cart abandonment rate is the share of shopping carts that never resulted in a purchase: one minus (completed purchases ÷ carts created), as a percentage. It measures drop-off after a customer adds an item but before they buy. It is broader than checkout abandonment, which starts at the checkout step, so the two should not be conflated — and bot-created or test carts inflate the denominator if not filtered.

Discount rate in retail (markdown rate) is the total value of discounts and markdowns divided by gross sales, as a percentage. It shows how much of potential revenue was given up to promotions, coupons, and clearance. It is distinct from the finance term 'discount rate' used in present-value calculations. A persistently high markdown rate erodes gross margin and can train customers to wait for sales rather than pay full price.

Micro-conversion rate is the rate at which visitors complete a smaller, intermediate action on the path to a primary goal — newsletter signups, add-to-cart, video views, or account creation — rather than the macro-conversion (purchase, qualified lead). Defined as micro-conversions divided by a relevant audience, it surfaces engagement and funnel drop-off earlier than waiting on the final outcome, making it a leading diagnostic signal.