MMM data requirements and prerequisites

Marketing Mix Modeling uses historical revenue, media investment, and relevant business events to estimate how much each marketing channel contributed to your results.

You do not need perfect data to use MMM, but you should understand where gaps or limitations exist. Missing data, limited spend variation, and channels that consistently move together can affect the model’s ability to separate their individual contributions.

Your model needs historical business and marketing data from the sources you want it to evaluate.

Your sales platform must be connected so Triple Whale can access your historical order and revenue data.

For Shopify stores, this data is imported through the Triple Whale Shopify integration.

Connect each advertising platform you want included in the model, such as Meta, Google, TikTok, or another supported platform.

Triple Whale uses historical media data from these integrations, including daily spend by channel. Where available, the data may also be organized by campaign type or media subcategory.

Include events that may have significantly affected revenue or marketing performance, such as:

Providing this context helps the model distinguish the effects of advertising from other changes in the business.

Approximately 12 months of historical data is the minimum for creating an initial model.

For more reliable results, 18–24 months of history is recommended. A longer history gives the model more information about:

A model created with less history may produce less stable results, particularly for channels with limited spend variation.

Use custom media uploads for sources that do not have a direct Triple Whale integration. Examples include:

Provide the data at a weekly cadence. Each source should include one or both of the following:

Custom media sources are subject to the same history and data-quality considerations as connected channels.

If a platform was disconnected or a channel stopped reporting during part of the model period, the model may not have enough information to estimate that channel’s contribution accurately.

Review connected platforms for missing or incomplete historical data before creating the model.

MMM learns from changes in investment over time. If a channel’s spend remains nearly identical throughout the model period, the model may have difficulty determining how changes in that channel affect results.

If you consistently increase or decrease investment across multiple channels at the same time, the model may have difficulty separating their individual contributions.

For example, if Meta and Google spend always rise and fall together, there may not be enough independent variation to determine the effect of each channel.

Branded search activity can be closely related to existing customer demand. If it is not categorized appropriately, the model may assign too much contribution to branded search.

Some affiliate costs are recorded only when a conversion occurs. Because the cost is directly tied to a purchase, affiliate data may need to be excluded or handled separately.

MMM refreshes weekly. Each refresh incorporates newly available revenue and media data and updates the model outputs.

No. However, you should identify missing periods, disconnected integrations, tracking changes, or other known limitations before interpreting the results.

Data gaps do not always prevent a model from being created, but they may reduce confidence in specific channel estimates.

Yes. From the Models page, open the model’s actions menu and select Update Model to revise its configuration. A newly added channel still needs sufficient history and spend variation before the model can estimate its contribution reliably.

Yes. Sources such as TV, radio, podcasts, out-of-home advertising, and direct mail can be included by manually providing data for channels without a supported integration.

The model may have difficulty separating their individual effects. This does not necessarily invalidate the entire model, but the individual channel estimates should be interpreted carefully.

Review the model’s data coverage, selected period, channel classifications, significant business events, and spend variation.

For more information about model fit, see Reading Your MMM Results.