MMM Predictions Table (BETA)

mmm_predictions_table

The MMM Predictions table tracks predicted and actual KPI values for each model run over time. One row per model run per date per set type. Train and held-out test rows are separated by set type.

📋
Table Guide
What You Can Answer With This Table

How closely did a model run's predictions match actual results over time? — use date, actual, prediction

What did the model predict for the held-out test period versus what actually happened? — filter set_type = test, use actual, prediction

What is the confidence range around a prediction? — use upper_ci, lower_ci

Before You Query

This table uses date, not event_date, for time-based queries.

For multiplatform models, the data is broken out by sales platform: each date and set type has one row per sales platform plus an additional _all row that aggregates across all of them.

For multiplatform models, only the _all rows carry a real confidence interval. On per-platform rows the interval collapses (upper_ci = lower_ci = prediction).

Key Relationships

Table Join Key Use This Join To
MMM Models model_id Get the configuration and accuracy scores for the model run that produced these predictions

When to Use a Different Table

Use MMM Attributions when you need channel-level KPI contribution. MMM Predictions stores total predicted and actual KPI per date.

View the full Triple Whale Data Ontology →

Table	Join Key	Use This Join To
MMM Models	`model_id`	Get the configuration and accuracy scores for the model run that produced these predictions

Dimensions

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

Title	ID	Type	Description
Prediction Created Date	`created_at`	timestamp	The time at which the prediction row was written to the database. Formatted according to the ISO 8601 international standard. Example value: `2024-01-22 06:15:00`
Date	`date`	date	The date the prediction applies to, at the model's `time_unit` granularity (daily, weekly, or monthly). Example value: `2024-01-15`
Model ID	`model_id`	string	The unique identifier of the model run that produced the prediction. Joins to MMM Models. Example value: `9f2c1a7e-3b44-4c9a-8e21-7d5f0b2a1c33`
Platform	`platform`	string	The sales platform associated with the prediction. `_all` represents the aggregate prediction across all sales platforms, while platform-specific values represent predictions broken down by individual platform. Example values: `_all`, `shopify`, `amazon`, `tiktok-shops`
Set Type	`set_type`	string	The dataset split the row belongs to: training data (in-sample) or test data (out-of-sample, held-out). Possible values: `test`, `train`
Settings ID	`settings_id`	string	The unique identifier of the model configuration. All runs of the same model share one `settings_id`. Example value: `3a7b2c9d-1e44-4f8a-9c12-6b0d8e2f4a51`

Measures

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

Title	ID	Type	Description
Actual KPI	`actual`	numeric	The actual observed KPI value for the date.
Prediction Lower Bound	`lower_ci`	numeric	The lower bound of the prediction's confidence interval.
Predicted KPI	`prediction`	numeric	The model's predicted KPI value for the date.
Prediction Upper Bound	`upper_ci`	numeric	The upper bound of the prediction's confidence interval.