Glossary
Key terms and definitions for marketing mix modeling, attribution, measurement, and optimization.
A
Adjusted R-squared
A version of R-squared that accounts for the number of variables in a model, penalizing unnecessary complexity; useful for spotting when a model has more variables than its data can support.
Adstock
The lingering effect of advertising spend after a campaign ends.
Attribution
The practice of assigning credit for conversions or revenue to marketing channels or campaigns.
B
Baseline leakage
When an MMM incorrectly attributes baseline sales to paid media, inflating channel performance.
Baseline revenue
The revenue a brand would generate without any paid marketing activity.
Behavioral data
Information about how customers act, such as purchase patterns, browsing history, and engagement depth, used to build more accurate predictive models.
Bottom-of-funnel
The final stage of the marketing funnel where customers are closest to converting.
Branded search
Search queries that include a brand's name, often driven by upper-funnel awareness activity.
Budget allocation
How marketing spend is distributed across channels and campaigns.
Budget optimization
The process of reallocating marketing spend across campaigns or channels to maximize revenue or efficiency based on model outputs.
C
Campaign-level granularity
The ability to measure marketing performance at the individual campaign level, rather than only at the channel level.
Channel-level attribution
Measuring performance aggregated at the channel level (e.g., all of Meta vs. individual Meta campaigns).
Churn
When an existing customer stops purchasing from or engaging with a brand; predictive models can flag churn risk early so retention efforts can kick in before customers disengage.
Confidence score
An indicator of how reliable a model's recommendation or output is.
Conversion
A desired customer action, such as a purchase or sign-up.
Cookie deprecation
The phasing out of third-party cookies by browsers and platforms, reducing the accuracy of pixel-based and user-level tracking tools like MTA.
Cross-channel interaction
The way campaigns on one channel influence performance on others, such as a Meta awareness campaign lifting branded search volume or Amazon conversion rates.
Customer acquisition cost (CAC)
The total cost of acquiring a new customer through marketing efforts.
Customer lifetime value (CLV or LTV)
The total revenue a brand can expect from a single customer over the course of their relationship with the brand.
Customer segmentation
The practice of grouping customers into meaningful categories based on demographic, behavioral, or predictive signals to personalize marketing efforts.
D
Data privacy
Regulations and practices that limit how user-level tracking data can be collected and used.
Demand forecasting
Using historical data and external signals to project future consumer demand, helping brands plan inventory, promotions, and channel spend ahead of peak periods.
Descriptive analytics
The practice of summarizing historical data to explain what has already happened, as distinct from predictive or prescriptive analytics.
Diagnostic analytics
Analysis focused on understanding why something happened, rather than just what happened or what is likely to happen next.
Diminishing returns
The point at which additional spend on a campaign produces progressively smaller gains in revenue or conversions.
Direct traffic
Website visits where users navigate directly to a brand's site, often influenced by awareness campaigns.
F
First-party data
Data collected directly by a brand from its own customers and platforms.
First-touch attribution
A single-touch model that gives all conversion credit to a customer's first interaction with a brand.
Forecasting
Using model outputs to project future revenue or performance under different spend scenarios.
G
H
I
Impressions
The number of times an ad is shown to users.
Incrementality testing
An experiment designed to measure the additional revenue or conversions directly attributable to a specific marketing activity.
Independent variable
In a marketing mix model, the inputs being tested for their effect on revenue, such as spend by channel, seasonality, or promotional events.
L
Last-touch attribution
A single-touch model that gives all conversion credit to the final interaction before a conversion.
Lift
The measurable increase in a metric (like sales or conversions) resulting from a marketing activity.
Linear attribution
An MTA model that distributes conversion credit equally across all touchpoints.
M
Machine learning
A branch of AI that uses algorithms to identify patterns in data and improve model accuracy over time.
Marketing mix
The combination of channels, campaigns, and tactics a brand uses to reach its customers.
Marketing mix modeling (MMM)
A statistical approach that uses historical data to measure how marketing and non-marketing factors contribute to revenue.
Mean absolute error (MAE)
A validation metric that measures the average size of errors between a model's predictions and actual outcomes, without regard to direction.
Media mix
The specific combination of paid advertising channels a brand invests in.
Model accuracy
How closely a model's outputs reflect real-world outcomes.
Model calibration
The process of adjusting a model's parameters using external data, such as incrementality test results.
Model validation
The process of assessing how well a model reflects reality, often by testing its outputs against known results.
Multi-retail connectors
Integrations that allow an MMM to incorporate sales data from retail partners like Target, Walmart, or Ulta.
Multi-touch attribution (MTA)
An attribution methodology that distributes conversion credit across multiple touchpoints in a customer's journey.
O
Omnichannel
A marketing approach that accounts for a customer's interactions across multiple platforms and retail environments.
Organic traffic
Website visitors who arrive through unpaid search results.
Out-of-sample accuracy
How well a model predicts data it was not trained on, a more meaningful measure of model reliability than in-sample fit alone.
Overfitting
When a model has essentially memorized historical data rather than learned the underlying patterns, producing high accuracy in retrospect but unreliable outputs when conditions change.
P
Paid media
Advertising placements a brand pays for, such as Meta ads, Google ads, or CTV.
Platform-reported ROAS
The return on ad spend as reported directly by an advertising platform, which may overclaim or underclaim due to tracking limitations.
Position-based attribution
An MTA model that gives weighted credit to the first and last touchpoints in a customer journey.
Predictive analytics
The use of historical data and statistical modeling to forecast future outcomes, such as campaign performance, customer behavior, or revenue.
Prescriptive analytics
Analysis that goes beyond predicting what is likely to happen and recommends specific actions to take, such as budget reallocations.
Prospecting campaigns
Ads targeted at new potential customers who haven't interacted with a brand before.
R
R-squared (R²)
A statistical measure of how well a model's predictions match observed data, expressed as a value between 0 and 1; represents the proportion of variance in the outcome variable that the model explains.
Residual analysis
The process of examining the errors between a model's predicted and actual values to identify patterns that suggest missing variables or structural problems.
Retargeting campaigns
Ads shown to users who have previously visited a brand's site or interacted with its content.
Return on ad spend (ROAS)
The revenue generated for every dollar spent on advertising.
Rolling backtest
A validation technique where a model is repeatedly re-trained on earlier data and tested against subsequent periods to measure how well it forecasts over time.
Root mean square error (RMSE)
A validation metric that measures the typical size of prediction errors, giving more weight to larger errors than MAE does.
S
Saturation
The point at which additional spend on a campaign no longer produces proportional returns.
Saturation curve
A visual representation of how a campaign's returns change as spend increases.
Seasonality
Predictable fluctuations in consumer demand tied to time of year, holidays, or events.
Single-touch attribution
An attribution model that assigns all conversion credit to one touchpoint, either first or last.
Spend efficiency
How effectively each dollar of marketing spend translates into revenue.
Structural misspecification
When a model's underlying assumptions don't match the actual dynamics of the system it's measuring, leading to persistent errors regardless of data quality.
Symmetric mean absolute percentage error (SMAPE)
A validation metric that measures forecast accuracy as a percentage, designed to be scale-stable and comparable across different data ranges.
T
Test-calibrated MMM
An approach that uses incrementality test results to adjust the parameters of a marketing mix model; Prescient's position is that this can't fix a structurally misspecified model.
Time-decay attribution
An MTA model that gives more credit to touchpoints closer in time to a conversion.
Top-of-funnel
The awareness stage of the marketing funnel, where potential customers are first introduced to a brand.
Touchpoint
Any interaction a customer has with a brand before converting.
V
Validation layer
Prescient's approach of running parallel model versions with and without test data to assess whether incrementality test results improve or degrade model accuracy.
Variance
The degree to which a set of values spread out from their average; in modeling, separating meaningful variance from noise is central to accurate attribution.
View-through attribution
Crediting a conversion to an ad a user saw but didn't click.