Glossary

The vocabulary, defined honestly.

Including where each term misleads, because most of the confusion in advertising measurement is definitional rather than technical.

Attribution: The rule that distributes credit for an outcome across the touchpoints that preceded it. It is bookkeeping rather than causality: change the model and every channel's number moves while nothing has changed in the market. Useful for comparing channels consistently over time, misleading the moment it is read as what each channel caused.
Attribution window: The period after a click or impression in which a conversion may still be credited to it. A short window undercounts long sales cycles; a long one credits channels for outcomes already in motion before the ad appeared. Two ROAS figures computed on different windows are not comparable, so every figure here carries its window.
CAC: Customer acquisition cost: ad spend divided by the number of new customers acquired in a period. It misleads when spend and customers fall in different windows, since customers closing in March on January's spend make March look efficient. The business metrics service computes it against a stated window and cohort; it is never a model estimate.
Campaign Constitution: The account-level limits the AI may never exceed: spend ceilings, maximum daily movement, maximum CAC, minimum margin, protected campaigns, approval requirements and minimum evidence thresholds. The action service enforces it before any change reaches an ad account. You own it, the model cannot edit it, and anything outside it becomes a proposal instead of an action.
Campaign intent: The structured statement of what a campaign is for: the business outcome, the market, the audience, the budget ceiling and the constraints it must respect. It is what the Growth Brain plans against and what makes a campaign reviewable months later. Without it, a campaign can only be judged against whichever metric its own network happens to report.
Channel adapter: The service that translates between this platform and one ad network's API — reading entities and reported figures, writing actions, and handling that network's rate limits, quirks and schema changes. Each network gets its own, because their objects and constraints do not map onto each other cleanly. The adapter is also where a network's reported numbers are captured as network truth, kept separate from your own data.
Contribution margin: Revenue minus the variable costs of delivering it: cost of goods, payment fees, shipping, discounts and refunds. It is what advertising actually has to cover, and it is the basis for Margin ROAS. The definition comes from your finance or accounting system; Realvertising reads it and never writes it.
Conversion value: The monetary amount attached to a conversion event and sent to a network's bidding model. Static values — every lead worth the same — teach the network to maximise the wrong thing. Real values drawn from your CRM and finance system change what the auction optimises toward, which is the whole reason for sending them back.
Creative concept: The idea behind a set of ads: the angle, the promise, the audience and the journey stage it addresses. Variants are executions of a concept, and when performance drops the useful question is whether the concept is exhausted or only this execution of it. Tracking concepts separately from variants is what keeps that question answerable.
Creative fatigue: The decline in performance that follows repeated exposure of the same creative to the same audience, visible as falling click-through with rising frequency and cost per outcome. It is easily confused with auction pressure or seasonality, which look almost identical on a single trend line. Telling them apart needs frequency and cohort data, not a headline metric.
Daily movement limit: The most budget that may shift between channels in one day. It exists because ad networks re-enter a learning phase after a large budget change, so a move that is correct in direction can still cost more than it gains if it is made all at once.
Decision object: The stored record of a single action: what changed, the previous and new values, the evidence that justified it, the rule that permitted it, who approved it and when. It is what makes a change reviewable long afterwards and reversible at any point. Every action taken in your ad accounts produces one.
Deterministic services: The engines that own the arithmetic: the attribution service, the budget engine, the business metrics service, the experiment service and the action service. The model chooses which of them to call and explains what comes back; it never computes a figure itself, so the same question asked twice returns the same number.
Enhanced conversion: A conversion returned to a network with hashed first-party identifiers attached, so it can be matched to a click that cookies no longer cover. It recovers measurable conversions otherwise lost to browser restrictions and consent choices. What may be sent is governed by the state configured in your consent platform, which this product enforces.
Exploration and exploitation: The trade-off between spending budget on what is already known to work and spending it to learn something new. Pure exploitation decays as audiences saturate and creative fatigues, and the account narrows without anyone deciding that it should. A fixed exploration share, set in your Constitution, is what keeps the next winner findable.
Four truth layers: The four separate readings of the same campaign: what the ad network reports, what independent attribution credits, what the CRM records about lead quality, and what the finance system records as collected revenue. They are never averaged into one score, because the distance between them is usually the finding.
Growth Brain: The planning engine that decides how channels work together — which deserve budget, which outcome each should optimise for, which audiences to expand or suppress, which creative to test. It plans the analysis and chooses which deterministic services to call; it does not compute business metrics itself. It acts only inside the Campaign Constitution.
Holdout: A group deliberately kept from seeing a campaign, so outcomes in the exposed and unexposed groups can be compared. It is the cleanest estimate of incremental effect available to an advertiser. It also costs real revenue while it runs, which is why its size and duration are set deliberately and reviewed rather than left open.
Idempotency key: A unique identifier attached to an action so that repeating the request applies the change exactly once. Ad network APIs time out and retry, and without a key a retried budget change can double a budget or leave a second campaign behind. Every action this platform sends carries one.
Incrementality: The share of an outcome that would not have occurred without the advertising. Only an experiment — a holdout or a geo test — estimates it; no attribution model, however detailed, can. Branded search and retargeting are the classic cases where credited revenue looks high and incremental revenue is far lower.
Learning phase: The period after a change during which a network's bidding model is recalibrating and its reported performance is unstable. Judging a campaign inside it, or editing again mid-phase, resets the clock and produces noise that looks like a result. Structural changes here respect each network's learning phase instead of overriding it.
Margin ROAS: Contribution margin credited to advertising divided by the spend. It answers whether a campaign pays for itself after cost of goods, discounts and refunds, which revenue ROAS cannot do. It depends entirely on the margin definition your finance system supplies; Realvertising reads that definition and does not author one.
Minimum evidence threshold: The spend, conversions and elapsed days a comparison needs before any action may be taken on it. Below the threshold no ranking is produced and the system states which condition is short, rather than acting on a difference that has not separated from noise.
Offline conversion: An outcome that happens away from the website — a qualified opportunity, a signed contract, a paid invoice — uploaded back to the ad network and matched to the original click. It lets network bidding optimise for outcomes it could never observe on its own. Match rates are never complete, so a network's offline conversion count is a floor rather than a census.
Pipeline: The value of open opportunities attributed to a source, before anything is won or collected. In a long sales cycle it is the earliest credible signal of ad quality and the easiest to overread, because pipeline that never closes is not revenue. Pipeline and collected revenue are therefore reported in separate columns, never summed.
Protected campaign: A campaign the Campaign Constitution places out of reach of automation entirely. Brand search is the usual example: it looks inefficient on incremental measures and cutting it is almost always a mistake, so the rule removes the decision rather than trusting a threshold.
Qualified lead: A lead your CRM has judged to meet its own criteria for fit, intent or stage. It matters because networks optimise toward whatever you send them, and sending raw form fills teaches them to find people who fill in forms. The definition stays in your CRM; this platform consumes it and returns qualified outcomes as offline conversions.
ROAS: Return on ad spend: revenue credited to advertising divided by the spend, usually as the ad network reports it. It says nothing about cost of goods, refunds, or whether the revenue would have arrived without the ad. Rising ROAS alongside falling profit is common and entirely possible.
Suppression list: An audience deliberately excluded from targeting: existing customers, serial refunders, unqualified segments, people who already converted. It keeps budget off people who cannot become new customers and keeps frequency off people who should not see the ad again. It needs a live CRM connection, because a static list is stale the week after it is built.
True ROAS: Revenue collected in your finance system set against spend, after reconciling network truth, attributed truth, CRM truth and financial truth rather than averaging them. It differs from platform ROAS because it drops double-counted conversions and refunded orders and includes revenue no network claimed. It remains credit assigned under an attribution rule, not evidence that the spend caused the revenue.
Value signal: The outcome data sent back to a network's bidding model: which conversions were real, what they were worth, which were later refunded. It closes the loop that otherwise leaves the network optimising against its own count of form fills. A network can only optimise toward what it receives, so the quality of this signal sets the ceiling on its bidding.

One goal. Every channel. Real business truth.

Connect your ad accounts, say what a good customer is worth, and let the Growth Brain work the objective inside limits you set.

Start free See how it works

No ad spend commitment · Connect one channel to start · [email protected]