The Paid Media Measurement Glossary

Measurement vocabulary is mostly platform-friendly language designed to obscure. Words like attribution, lift, contribution, and ROAS get bent in vendor decks until they all sound like proof that the channel worked. The terms below are the plain definitions QRY uses to evaluate vendor claims, platform case studies, and internal measurement plans. None of them require a statistics background to apply; they require a willingness to ask what the number is actually measuring before quoting it. The broader taxonomy of how these methods relate sits in MMM vs MTA vs incrementality.

Five categories cover the working vocabulary: attribution and credit assignment (who gets the conversion), incrementality and lift testing (what would have happened without the spend), marketing mix modeling (how channels contribute over time), statistical foundations (what makes a result a result), and the economics lens (what the spend means for the business). Each term gets a definition, the reason it matters, and the practical caveat that decides whether it belongs in a budget conversation.

Attribution and credit assignment

Last-click attribution. The model that assigns 100 percent of credit for a conversion to the last trackable click before purchase. Useful for in-platform optimization because it ties directly to the click the platform can see. Catastrophic as a budget input because it systematically overcredits lower-funnel channels that intercept demand the upper funnel created. Brand search wins last-click; the campaign that drove the brand search loses it.

View-through credit. Credit assigned to an impression that was served but not clicked, when a conversion happens within a defined window afterward. The window is usually 24 hours to 30 days and is set by the platform. The problem is the counterfactual: a view-through conversion looks the same in the data whether the impression caused it or merely accompanied it. Treat view-through as platform marketing material, not measurement.

Multi-touch attribution (MTA). A model that distributes credit for a conversion across multiple touchpoints in a user's path. Algorithmic versions weight each touch based on observed conversion patterns. MTA is correlation on observed paths and cannot see users it failed to track or counterfactual journeys that never happened. Useful for diagnosing within-funnel handoffs. Not a substitute for a causal read.

Data-driven attribution (DDA). Google's algorithmic version of MTA, applied to conversions inside Google's ecosystem. It uses machine learning on path data to assign fractional credit across touchpoints. The output is more defensible than last-click for within-Google optimization, but it shares MTA's fundamental ceiling: the platform is grading touches it can see, on conversions it can attribute, using a model only the platform can audit.

Platform-reported conversions. The conversion count a platform shows in its own dashboard, using its own attribution rules and lookback windows. Meta, Google, TikTok, and the retail media networks all run their own bookkeeping on themselves. Sum the dashboards and you will overcount real conversions by 30 to 100 percent because every platform is claiming the same conversions. Trust the direction, not the magnitude.

Incrementality and lift testing

Incrementality. The share of an observed outcome that would not have happened without the spend. The counterfactual is the comparison point: revenue with the campaign, minus the revenue the business would have earned without it. Everything else in measurement is a story about credit assignment. Incrementality is the only frame that answers the question a CFO actually asks about a marketing budget.

Lift (and lift percentage). The measured gap between a treatment group exposed to a campaign and a matched control group that was not. Absolute lift is the dollar or unit difference; lift percentage normalizes that gap against the control's baseline. A 10 percent lift on a $1M control means the campaign added $100k on top of the baseline the business would have earned regardless. The full math is in the incrementality formulas reference.

Incremental ROAS (iROAS). Return on ad spend calculated using only the revenue the campaign actually caused, not the revenue the platform attributed. Formula: incremental revenue divided by treatment spend. The honest version of ROAS for budget decisions. Platform ROAS routinely runs two to three times higher than iROAS on the same window because platform ROAS counts conversions that would have happened anyway.

Geo holdout. An experiment that turns a channel on in treatment markets and off in matched control markets, then measures the revenue gap. The geographic split builds the counterfactual into the design rather than modeling it afterward. The deeper walkthrough on test design, matched markets, and how to read the result lives in geo-lift testing explained.

Conversion lift (platform-run). A lift test executed inside a platform, where the platform holds out a randomized control group from its own audience and then grades the result itself. Meta Conversion Lift, Google's equivalent, and the major DSP versions all work this way. Faster and cheaper than a geo holdout, less independent because the platform designs the test and reports the answer on its own performance. Use to read fast creative cycles, not to settle budget arguments.

Marketing mix modeling

Marketing mix modeling (MMM). A statistical model that estimates each channel's contribution to revenue from aggregate weekly or daily spend, revenue, and external variables. MMM models the counterfactual rather than constructing it; the answer is an estimate, not a measurement, and its quality depends on spend variation, model specification, and calibration against experiments. The deeper QRY position lives in the MMM glossary and math reference.

Adstock. The MMM concept that ad exposure has a decaying effect over time rather than only on the day it ran. A flight last week still influences this week's revenue at a diminishing rate. Adstock parameters control how fast that decay is and matter enormously for long-cycle categories. Get the decay wrong and the model will misallocate credit between channels that pulse and channels that run continuously.

Saturation curve. The function inside an MMM that describes how each incremental dollar in a channel returns diminishing revenue. Real channels do not return linearly; the tenth dollar in Meta is worth less than the first. The shape of the saturation curve is what tells you whether a channel is undersaturated and worth scaling, or oversaturated and worth pulling. Most of the practical budget guidance MMM produces lives in this curve.

Calibration. The practice of using a real experimental result, usually a geo holdout, to constrain an MMM's estimate for the tested channel. The MMM is refit so its estimate for that channel matches the experimental read, and the rest of the model must adjust around it. Uncalibrated MMM is a coherent story with no anchor in causal data; calibrated MMM is the only version a finance team should be allocating against.

Statistical foundations

Confidence interval. The range of values consistent with the data at a stated level of certainty, usually 90 or 95 percent. A lift estimate of 8 percent with a 95 percent interval of 2 to 14 percent is a real result; the same point estimate with an interval of -3 to 19 percent is a coin flip. If the interval crosses zero, the test could not distinguish the channel from no effect. Always read the interval, not just the point estimate.

Minimum detectable effect (MDE). The smallest lift a given test design can reliably distinguish from zero. MDE is set by sample size, baseline variance, and the chosen significance and power levels, not by the lift you wish you could detect. A test sized for a 20 percent MDE cannot diagnose a real 5 percent lift; it will return a null and the team will conclude the channel is dead. The math for sizing creative and channel tests lives in the creative testing statistics reference.

Power. The probability that a test will detect a real effect of a given size if that effect exists. Standard practice is 80 percent power; the test should catch a true effect four times out of five at the targeted MDE. Underpowered tests are the silent killer of measurement programs because they produce nulls that look like the channel does not work, when the real story is that the test was never going to read out.

Economics and business lens

Contribution margin. Revenue from a sale minus the variable costs of producing and delivering that sale: cost of goods, shipping, payment processing, returns. Contribution margin is what a marketing dollar actually competes against, not top-line revenue. A campaign with a 3.0 iROAS on a product with 20 percent contribution margin is funding itself; the same iROAS on a product with 60 percent margin is funding the whole company. Always read paid media results through the margin lens.

Marketing efficiency ratio (MER). Total revenue divided by total marketing spend across all channels in a period. MER does not assign credit to any particular channel; it just measures whether the whole marketing function is producing revenue at the efficiency the business needs. Useful as a sanity check on the sum of platform ROAS claims; if platform ROAS averages 4.0 but MER is 1.8, the platforms are double-counting and the real efficiency lives closer to MER.

Lifetime value (LTV). The contribution margin a customer generates over their full relationship with the business, discounted back to today. LTV is what determines how much a business can afford to spend acquiring a customer, and it is the denominator in every serious LTV to CAC ratio. The trap is using gross revenue rather than contribution margin, or projecting LTV against cohort data that is too thin to support the curve. The formula derivations live in the CAC and unit economics formulas reference.

Vocabulary is not a side concern. Every budget conversation that goes sideways starts with two people using the same word for different things: ROAS that means platform-attributed to one side and incremental to the other, lift that means a platform conversion-lift readout to one side and a geo holdout to the other, contribution that means revenue to one side and margin to the other. The vocabulary is the difference between vendor-friendly storytelling and a CFO-defensible budget call. The companion piece on the language platforms use to dress weak measurement up as strong lives in the paid media anti-pattern glossary.

The working test for any term in this glossary: can it be defined without referencing the platform's own dashboard? If the answer is no, the term is platform bookkeeping in measurement clothing. If the answer is yes, it earns a seat at the budget table. Use the definitions above to keep the conversation honest, then run the experiments that produce numbers worth defending.

Platform attribution is not measurement. It is bookkeeping done by the entity being measured.

— QRY Measurement Glossary