Guides · Updated 29 October 2026
Marketing mix modeling (MMM) explains weekly sales with everything that moved them: media, price, promotions, distribution, season and the competition. For consumer packaged goods most sales happen in stores you don't own, so click-based attribution misses most of them, and a mix model built on aggregate store sales is the one view that covers every channel. The model is only as good as what goes in. Here is the data a CPG brand needs, and where each piece usually lives.
How much history
Weekly data, ideally two to three years. Google's open-source Meridian guidance is at least two years of weekly data for a model split by region and three for a national one. Use the finest regions your sales data has: TV markets or store clusters beat states or retailer regions.
1. Sales: the thing the model explains
| Item | Usual source | Grain |
|---|---|---|
| Retail sell-out, dollars and units | Syndicated store data, or retailer portal / POS feeds | week × market × retailer |
| Shipments (sell-in) | Your ERP | week × customer, for reconciling only |
| Ecommerce, own site and marketplaces | Commerce platform, marketplace seller reports | week × region if available |
Check your syndicated data licence lets you share the extract with a vendor.
2. Trade, price and shelf
In CPG a promotion week or new distribution can lift volume as much as a media flight. Leave them out and their credit goes to media.
- Average shelf price and base price
- Temporary price reduction depth, and the share of stores running it
- Feature ad, display, and feature plus display
- Distribution: store reach, items on shelf, out-of-stocks
- Promotion calendar, one row per event (needed to measure the dip after it: see trade promotion ROI)
- Trade spend per event, and gross margin per product line
- Launches, pack or size changes, price increases
3. Media, by market and week
- Linear TV: GRPs by market and spend, from agency buy and post-buy files
- Connected TV and online video, paid social, search and shopping: impressions and spend
- Retail media, on-site and off-site, per retailer network (keep each network's own attributed sales separately, for comparison only)
- Out-of-home, audio, print, sponsorships, influencer: spend plus delivery where it exists
- Coupons: distributed, redeemed, cost
4. Things outside your control
Seasonality and holidays, competitor media and promotions if you track them, category trend, weather, and one-off shocks such as recalls or supply problems. Add branded search volume by region and week: it is how the model can show TV and video driving people to search for your brand. Record gaps as gaps; never fill them with zeros.
5. Lift tests
Any geo-lift, matched-market or platform lift test from the last three years: dates, markets, spend, measured lift and its error. Two tests matter most: one on your biggest upper-funnel channel, and a brand-search holdout (paid brand ads paused in some regions). On our made-up test data, without them the model gave TV about a quarter of its true return and paid brand search about twice its own; with them every channel's true value fell inside the model's range.
Test size matters more than test count. Small tests spread across many channels can make a plan worse. A budget should move only when that channel's own test range backs the move.
What not to send
No personal data: no customer lists, loyalty-card or order-level files. A mix model works from totals by week and region.
What you get back
Sales per $1 and profit per $1 for each channel and promotion type, with ranges; which channels have enough evidence to move budget; and which test to run next for the rest. See a sample readout on made-up data.
Questions
What is marketing mix modeling for CPG?
A statistical model of weekly sales by region that measures how much media, price, promotions and distribution each added, using aggregate data rather than tracking individual shoppers.
How much data do I need for marketing mix modeling?
Weekly data for about two to three years; Google's Meridian guidance is at least two years for a regional model and three for a national one.
Do I need to share customer data?
No. A mix model uses totals by week and region. Customer lists and order-level files are not needed.
Why include trade promotions in a marketing mix model?
Promotions and distribution move CPG volume as much as media. If they are left out, the model hands their sales to whatever media ran in the same weeks.
Want this for your brand?
MMMflow is planning its first pilots with CPG and retail brands: 12 weeks, your store-level sales, media and promotion calendar in one open model, and a plain-words plan at the end. Ask about a pilot