Orders data in Marketing Mix Modeling (MMM)
How to provide order and subscription counts as the business outcome signal for Sellforte modeling.
Sellforte can use the count of orders or subscriptions as the business outcome signal for modeling. This makes sense when the number of completed transactions is what marketing is trying to influence and the monetary value of each transaction is either uniform, handled downstream, or not the KPI the business steers by. Typical fits include subscription services, memberships, telco sign-ups, insurance policies, app installs that convert, and commerce setups where the count of orders is a cleaner signal than revenue.
Order counts can be modelled alongside sales revenue or on their own. When pricing, mix, and discounting move week to week, an order count series often shows the true response to marketing more cleanly than a revenue series, because it removes the price and mix noise and leaves the acquisition signal.
Dataset definition
Counts of orders or subscriptions completed during the modeling period, structured either as one row per order or as daily aggregates per relevant dimension, used as business outcome signals for modeling.
Sellforte accepts order data in a single table or in several tables that can be linked with a unique identifier. Because the point of this dataset is the count, no monetary fields are required. If your source table happens to carry revenue or discount values you can include them or leave them out, but they are not needed for order-count modeling.
Grain & sample
Each row represents one completed order or subscription on a given date, with the dimensions your business tracks for attribution and segmentation. If your source system delivers pre-aggregated daily counts per dimension combination instead, that shape works equally well.
Example rows (illustrative, order-level):
| date | order id | web order id | sales channel | location id | product category 1 | customer type | campaign id |
|---|---|---|---|---|---|---|---|
| 2025-01-01 | O-10021 | W-88213 | online | Streaming plan | new | jan_launch | |
| 2025-01-01 | O-10022 | retail | 1042 | Mobile plan | existing | ||
| 2025-01-02 | O-10023 | W-88301 | online | Streaming plan | new | jan_launch |
Dimensions
- date (required for time-series modeling)
- web order id (links online orders to Google Analytics 4 sessions)
- order id
- item id
- product category 1 (the primary categorization used to group orders in modeling)
- sales channel (online, retail, call center, partner)
- location id (for orders tied to a physical location)
- loyalty card (for separating member and non-member orders)
- campaign id (if a campaign identifier is captured at order time)
- product categories 2 to 4, item name, item brand, EAN, customer type (new vs. existing), location name, campaign name
Metrics
- order count (the primary target KPI, expressed as the number of completed orders or subscriptions)
- volume (units, sessions, or plan seats attached to the order, when the count of items differs from the count of orders)
Order vs. subscription note
Because an order and a subscription can behave differently over time, please document which event a row represents. An order in a retail sense is usually a one-off completion. A subscription is a recurring relationship where the meaningful event may be the initial sign-up, a renewal, an upgrade, or a churn recovery. The interpretation of marketing effect changes depending on which event the count represents, so being explicit about this in the delivery avoids ambiguity in results.
Cancellations and returns
If orders can be cancelled or returned after the fact, decide how to handle them consistently. Either provide the gross count and a separate cancellation feed, or provide net completed orders and hold the delivery until the return window has closed. Mixing the two approaches within one series produces artificial dips and inflates the estimated effect of whatever ran on the days the corrections landed.