Marketing Measurement Dictionary

The decaying residual impact of past advertising on current sales. The simplest form is geometric:

$$A_t=M_t+\lambda A_{t-1},0\le \lambda <1$$

A higher λ means a longer-lasting flight (TV, brand campaigns typically λ ≈ 0.7); a lower λ means same-week response (paid search λ ≈ 0.2).

The classic media-buying unit: GRP = Reach × Frequency. One GRP = 1% of the target audience reached, on average, one time.

Used because it normalizes for price differences (a Super Bowl spot and a 3am spot may deliver wildly different costs but comparable GRPs), and because it measures exposure — the actual input to memory and behavior.

Apple’s per-device advertising identifier on iOS. Since iOS 14.5 (2021) apps must request user permission to read it (“App Tracking Transparency”), and most users decline.

The “post-IDFA era” is shorthand for the broader collapse of cross-app user tracking on mobile, which broke a lot of user-level attribution pipelines and pushed measurement back toward aggregate techniques like MMM and incrementality holdouts.

The conversions that would not have happened without the marketing — as opposed to conversions that were going to happen anyway and just happened to follow an ad exposure.

The gold-standard way to measure it is a holdout test: randomly withhold ads from a comparable group and compare. MMM estimates it econometrically; multi-touch attribution does not measure it at all.

Marketing Mix Modeling is the best default term for the broader discipline: estimating how marketing activities and non-marketing factors contribute to outcomes such as sales, revenue, leads, signups, or visits.

In practice, many MMM projects focus heavily on paid media allocation across channels such as search, social, TV, CTV, display, audio, and OOH. That is why media mix modeling is often used interchangeably, especially in ad measurement discussions.

The distinction is useful:

• Marketing Mix Modeling is the broader term. It can include media, promotions, pricing, seasonality, distribution, macro factors, brand effects, and other business drivers.
• Media Mix Modeling is narrower. It usually emphasizes media-channel response and budget allocation.
• Market Mix Modeling appears occasionally, but it is less standard in U.S. analytics language. Use it only when quoting a source.

A user-level approach that distributes credit for a conversion across the ads that user saw — usually via rules (last-click, first-click, linear, U-shaped) or a fitted model (Markov, Shapley).

MTA needs a stitched user graph (cookies, MAIDs, logged-in IDs). Since the deprecation of third-party cookies and the IDFA opt-out era, that graph is increasingly broken, which is why MMM has come back into focus.

MTA also doesn’t measure incrementality — it splits observed credit, not causal credit.

The non-linear response that turns extra ad spend into less-than-proportional extra sales. The Hill function is the workhorse:

$$f(x)=\frac{\beta x^{\alpha }}{K^{\alpha }+x^{\alpha }}$$

• β — the channel’s asymptote (maximum incremental conversions).
• K — the half-saturation point.
• α — the steepness (α > 1 gives an S-curve).

The slope at your current spend is the marginal return — the real number that should drive reallocation decisions.