The MMM Models table stores Marketing Mix Modeling model runs, including configuration and train/test accuracy. One row per model run. A configuration can have multiple retrained runs over time.

📋
Table Guide
What You Can Answer With This Table

Which model run is current for a configuration? — use settings_id, window_end_date

How accurate was a model run on held-out data? — use test_smape, test_wmape, test_r_squared

Which KPI, efficiency ratio, and granularity define the run? — use kpi, efficiency_ratio_type, time_unit

Before You Query

This table is not time-series, so no date filter is needed.

Key Relationships

Table Join Key Use This Join To
MMM Predictions model_id Compare a model run's predicted vs actual KPI over time
MMM Attributions model_id Break a model run's KPI into per-group contributions
MMM Marginal Lifts model_id Get each modeled group's marginal return for a model run
MMM Campaign Groups model_id See which campaigns were grouped into each modeled group for a model run
MMM Reallocation Settings model_id Get the budget optimization and simulation scenarios run on a model

When to Use a Different Table

Use MMM Reallocations when you need scenario budget moves per modeled group. MMM Models stores model configuration and aggregate run accuracy.

Use MMM Predictions when you need predicted vs actual KPI values over time. MMM Models stores run-level accuracy metrics.

View the full Triple Whale Data Ontology →

Table	Join Key	Use This Join To
MMM Predictions	`model_id`	Compare a model run's predicted vs actual KPI over time
MMM Attributions	`model_id`	Break a model run's KPI into per-group contributions
MMM Marginal Lifts	`model_id`	Get each modeled group's marginal return for a model run
MMM Campaign Groups	`model_id`	See which campaigns were grouped into each modeled group for a model run
MMM Reallocation Settings	`model_id`	Get the budget optimization and simulation scenarios run on a model

Dimensions

Dimensions are immutable properties that can be used for grouping data.

Title	ID	Type	Description
Model Run Created Date	`created_at`	timestamp	The time at which the model run was written to the database. Formatted according to the ISO 8601 international standard. Example value: `2024-01-22 06:15:00`
Exclude Amazon	`exclude_amazon`	boolean	True if Amazon data was excluded from the model's training. Possible values: `true`, `false`
Exclude Black Friday	`exclude_black_friday`	boolean	True if the Black Friday / Cyber Monday period was excluded from the model's training. Possible values: `true`, `false`
Is Deleted	`is_deleted`	boolean	True if the model has been deleted. Deleted models are retained (soft delete) rather than removed. Possible values: `true`, `false`
Is Experimental	`is_exp`	boolean	True if the model run is experimental rather than a standard production run. Possible values: `true`, `false`
KPI	`kpi`	string	The business outcome the model is trained to explain and predict, including the customer population (total or new customer). Possible values: `NC Revenue`, `Revenue`, `NC Profit`, `Profit`, `NC Orders`, `Orders`, `NC Subscriptions`, `Subscriptions`
Efficiency Ratio Type	`efficiency_ratio_type`	string	The name of the marketing efficiency ratio paired with the model's KPI, used to express return on ad spend. This is a label such as `ROAS`, not a numeric value, and is derived from `kpi`. Possible values: `ROAS`, `NC ROAS`, `POAS`, `NC POAS`, `CAC`, `NC CAC`, `SCAC`, `NC SCAC`
Model ID	`model_id`	string	The unique identifier of the model run. Each scheduled retrain produces a new `model_id`. Example value: `9f2c1a7e-3b44-4c9a-8e21-7d5f0b2a1c33`
Settings ID	`settings_id`	string	The unique identifier of the model configuration. All runs of the same model share one `settings_id`; this is the key that ties a model's runs together across the MMM tables. Example value: `3a7b2c9d-1e44-4f8a-9c12-6b0d8e2f4a51`
Settings Name	`settings_name`	string	The human-readable name of the model configuration. Example value: `revenue_weekly_a3b2`
Training Start Date	`start_training_date`	string	The first date of historical data used to train the model. Example value: `2023-01-01`
Time Unit	`time_unit`	string	The granularity at which the model is trained and reports results. Possible values: `daily`, `weekly`, `monthly`
Use Clicks	`use_clicks`	boolean	True if ad clicks were included as a signal in the model. Possible values: `true`, `false`
Use Discount	`use_discount`	boolean	True if discount activity was included as a signal in the model. Possible values: `true`, `false`
Use Email/SMS	`use_email_sms`	boolean	True if email and SMS activity was included as a signal in the model. Possible values: `true`, `false`
Use Impressions	`use_impressions`	boolean	True if ad impressions were included as a signal in the model. Possible values: `true`, `false`
Window End Date	`window_end_date`	string	The end date of the model run's evaluation window. The most recent model run for a configuration has the latest `window_end_date`. Example value: `2024-01-22`
Window Start Date	`window_start_date`	string	The start date of the model run's evaluation window. Example value: `2024-01-15`

Measures

Measures are numeric fields that can be aggregated and/or combined to calculate new metrics.

Title	ID	Type	Description
Test CRPS	`test_crps`	numeric	Continuous Ranked Probability Score on the held-out test period, evaluating the full predicted distribution against actuals. Lower is better.
Test CRPS Percentage	`test_crps_percentage`	numeric	Test-period CRPS expressed as a percentage of the actual KPI. Lower is better.
Test MAPE	`test_mape`	numeric	Mean Absolute Percentage Error on the held-out test period. Lower is better.
Test Mean Error	`test_mean_error`	numeric	Average signed error between predicted and actual KPI on the test period, indicating directional bias.
Test Pearson Correlation	`test_pearson`	numeric	Pearson correlation between predicted and actual KPI on the test period. Closer to 1 is better.
Test R-Squared	`test_r_squared`	numeric	R-squared (coefficient of determination) on the test period. Closer to 1 means more variance explained.
Test RMSE	`test_rmse`	numeric	Root Mean Squared Error on the test period, in KPI units. Lower is better.
Test SMAPE	`test_smape`	numeric	Symmetric Mean Absolute Percentage Error on the test period. Lower is better.
Test SMAPE CI Lower	`test_smape_ci_lower`	numeric	Lower bound of the confidence interval for the test-period SMAPE.
Test SMAPE CI Upper	`test_smape_ci_upper`	numeric	Upper bound of the confidence interval for the test-period SMAPE.
Test WMAPE	`test_wmape`	numeric	Weighted Mean Absolute Percentage Error on the test period. Lower is better.
Train CRPS	`train_crps`	numeric	Continuous Ranked Probability Score on the training (in-sample) data. Lower is better.
Train CRPS Percentage	`train_crps_percentage`	numeric	Training-data CRPS expressed as a percentage of the actual KPI. Lower is better.
Train MAPE	`train_mape`	numeric	Mean Absolute Percentage Error on the training (in-sample) data. Lower is better.
Train Mean Error	`train_mean_error`	numeric	Average signed error between predicted and actual KPI on the training data, indicating directional bias.
Train Pearson Correlation	`train_pearson`	numeric	Pearson correlation between predicted and actual KPI on the training data. Closer to 1 is better.
Train R-Squared	`train_r_squared`	numeric	R-squared (coefficient of determination) on the training data. Closer to 1 means more variance explained.
Train RMSE	`train_rmse`	numeric	Root Mean Squared Error on the training data, in KPI units. Lower is better.
Train SMAPE	`train_smape`	numeric	Symmetric Mean Absolute Percentage Error on the training data. Lower is better.
Train SMAPE CI Lower	`train_smape_ci_lower`	numeric	Lower bound of the confidence interval for the training-data SMAPE.
Train SMAPE CI Upper	`train_smape_ci_upper`	numeric	Upper bound of the confidence interval for the training-data SMAPE.
Train WMAPE	`train_wmape`	numeric	Weighted Mean Absolute Percentage Error on the training data. Lower is better.