Marketing Mix Modelling for Ecommerce Brands

Ecommerce brands generate more data than almost any other sector, yet most still rely on last-click attribution. Marketing mix modelling (MMM) cuts through the noise and shows which channels genuinely move revenue.

Every ecommerce team has access to click data, conversion data, and platform dashboards. The problem is that each platform takes credit for the sale, and the numbers never add up. A customer sees a YouTube ad, clicks a paid search ad a week later, and buys through an affiliate link. Google claims the sale. Meta claims the sale. The affiliate claims the sale. MMM cuts through that overlap by looking at the relationship between total spend and total revenue over time, without relying on any individual click trail.

The ecommerce measurement problem

Ecommerce sits in an uncomfortable middle ground. You have more trackable data than a TV advertiser, but far less than you think once you account for iOS privacy changes, browser tracking restrictions, and cross-device behaviour. Studies consistently show that platforms like Meta and Google over-report their contribution by 30 to 60 percent when compared against controlled experiments. If you are using platform data alone to make budget decisions, you are almost certainly over-investing in channels that look efficient but are picking up credit for sales that would have happened anyway.

MMM works by analysing historical data at an aggregate level, meaning it does not track individuals. It measures how changes in media spend, price, promotions, and external factors (like bank holidays or competitor activity) correlate with changes in revenue. The model separates out the baseline, which is the revenue you would get with no media spend at all, from the incremental revenue each channel contributes.

The typical ecommerce media mix

Most ecommerce brands spread budget across paid search (Google and Bing), paid social (Meta, TikTok, Pinterest), display and programmatic, affiliates and cashback sites, email marketing, and organic search. Each channel plays a different role in the purchase journey. Paid search tends to be highly efficient at the bottom of the funnel, capturing intent that was already created by other channels. Paid social builds that intent earlier. Affiliates often convert customers who were already on the path to buying.

Without MMM, brands typically over-invest in paid search because it looks cheap on a last-click basis. The model reveals that a significant portion of those clicks are from people who would have found the site anyway, either through organic search or direct traffic. Knowing that changes how you allocate budget.

Data sources and what you actually need

• Weekly revenue and orders by channel, pulled from your ecommerce platform (Shopify, Magento, etc.)
• Media spend by channel and week, ideally going back two to three years
• Promotional calendar: sale events, discount codes, free delivery periods
• Pricing data if average order value fluctuates
• External factors: bank holidays, competitor promotions, economic index data
• Seasonality indicators specific to your product category

You do not need perfect data to run a model. You need consistent data. A two-year weekly dataset is the minimum. Three years is better because it gives the model enough seasonal cycles to separate the effect of Christmas spend from the underlying trend. If you have large gaps in your spend data (periods where reporting was paused or channels changed), flag them early so the model can account for them properly.

Ecommerce seasonality and why it matters

Ecommerce is one of the most seasonal sectors in marketing. Black Friday, Cyber Monday, Christmas, Valentine's Day, Mother's Day, and summer sales all create sharp spikes in revenue. If your model does not account for this correctly, it will attribute the Black Friday revenue uplift to whatever media was running at the time, rather than to the event itself. That leads to wrong conclusions about which channels work.

Build your promotional and event calendar before you start modelling. Seasonal events that are not flagged in the data get absorbed by the media variables, which inflates their apparent effectiveness.

Common measurement mistakes in ecommerce

The most expensive mistake is treating platform-reported ROAS (return on ad spend) as ground truth. Platform numbers are self-reported by the platform selling you the media. They have no incentive to undercount. The second common mistake is ignoring the adstock effect. Adstock refers to the fact that advertising has a delayed and decaying impact. A TV ad or YouTube campaign running in week one will still be driving sales in weeks two, three, and four. If you measure only within the week of spend, you will undervalue upper-funnel channels every time.

Brands also frequently forget to model the impact of returns. High return rates, common in fashion ecommerce, mean that reported revenue and actual net revenue diverge. Model on net revenue wherever possible, or at least run a sensitivity check to understand how returns affect your channel effectiveness estimates.

What to do with MMM results

A good MMM output gives you three things: the contribution of each channel to revenue, the marginal return on each channel (what you get from the next pound spent), and a budget optimiser that shows you what mix would maximise revenue or profit at your current total budget. Use the optimiser to identify where you are overspending relative to the point of diminishing returns, and where you have room to invest more.