E-commerce marketing budgets, set by what the next order earns.

The sections below cover the break-even return your margin sets, why platform numbers can't size the budget, and the three things that trip up most e-commerce models: returns, pay-per-sale affiliates and campaign weeks.

Revenue does not pay for advertising. Gross margin does. With a 40% gross margin, an order has to bring in 2.5 times its marketing cost in revenue to break even. At 25% margin it takes 4 times. The ratio is simply one divided by the gross margin, and it is your break-even return on ad spend.

Two things follow. A store with thin margins needs much higher returns than one with healthy margins, so a ROAS target borrowed from a benchmark or another business can be badly wrong for you. And the target belongs on the margin: the last part of the budget has to clear break-even, the average is not enough.

Add up what Google, Meta and the affiliate networks report, and the total is often larger than the sales you actually made. Each platform counts the orders it can see and claims them. Branded search and retargeting are especially good at claiming orders from people who had already decided to buy.

Even a perfectly counted platform ROAS would be an average. It tells you what the channel returned on all the spend so far, and nothing about what the next part of the spend returns. As a channel saturates, the two drift apart: the average still looks fine while the last part of the budget has stopped paying.

Platform numbers remain the right tool inside a channel, for choosing campaigns, audiences and products. The budget across channels is a different question: what each channel adds on top of what would have happened anyway, all channels at once. That is what marketing mix modeling estimates. Why Google Analytics and Meta report different numbers.

In fashion and other categories with many returns, the outcome you model matters. A webshop's sales feed often books each return as a negative line on the day it is processed, weeks after the order. Fed straight into a model, those returns pull revenue out of the wrong weeks. A week with heavy Meta spend and strong orders gets hit by returns from earlier orders, and Meta looks weaker than it is. The error is systematic, which makes it worse than noise.

The fix is to date each return back to the order it belongs to. Where the shop platform links a return to its original order, the model runs on net revenue after returns. Until that link exists, the model runs on sales before returns, with the assumption that the return rate is roughly the same whichever channel brought the order. In practice returns tend to vary far more by product category than by channel, which keeps the assumption workable. Either way, the choice is agreed with you before the first model is built.

Affiliate paid per sale is a special case. Because the cost follows the sales, affiliate spend rises and falls with sales whatever the affiliate actually did. A model that treats it like any other channel reads that correlation as effect and hands affiliate a large share of the budget.

We keep pay-per-sale affiliate in the model at its historical spend, so the other channels are estimated correctly, and we keep it locked in the recommendations rather than letting the optimizer scale it. Its own estimated effect is not read as the truth, whether it comes out high or low. Affiliate placements bought at a fixed cost, such as a banner on a partner site, behave like media and can be modeled as an ordinary channel. The same reasoning applies to price comparison sites, depending on how you pay them. See how we collect affiliate and price comparison data.

Black Friday, Christmas and sales periods pull the budget in two directions at once. Demand is high, so marketing converts more easily and the same spend brings more orders. But discounts cut the margin, so each order is worth less and the break-even return goes up.

A simple example: a product with a 50% margin sold at 30% off ends up at about 29% margin, and its break-even ROAS rises from 2 to about 3.5. The right spend for campaign week depends on both effects together: how well each channel converts in that week, and what an order is worth at campaign prices.

In the model, Black Friday and other special days are among the few events we mark explicitly, so their lift is not credited to whichever channels ran that week. With that in place, the model shows which weeks the next part of the budget works hardest, and the plan can follow it instead of spreading spend evenly. Read how MMM separates promotions from advertising.

Most stores model total sales, because paid marketing also brings back existing customers, and total sales is the number the budget has to answer to. Where the business case rests on growing the customer base, a second model on new customers' orders shows which channels bring people who had never bought before.

The two models often agree. When they don't, the difference is useful: a channel that looks strong on total sales but weak on new customers is mostly reaching people who would have come back anyway. Some customers run both side by side, one for the overall budget and one for growth.

Odins is a managed marketing mix modeling platform. We connect the ad platforms, affiliate networks and your shop's sales data, build and maintain the model, and review every recommendation before it reaches you. Data flows in continuously and the model is retrained once a month, while scenarios and forecasts are yours to run whenever you need them, in the platform or from your own AI assistant.

Moving budget toward channels with headroom typically gives 5 to 15% more effect from the budget you already have. Companies like CDON, Kin and Vakre Vene run their marketing budgets on Odins. See the managed service.