-1.png?width=1132&height=292&name=Asset%201@2x%20(12)-1.png "Odins.ai Logo")
New to marketing mix modeling?
Marketing mix modeling (MMM) is a statistical method that estimates how much each marketing channel adds to sales, using spend and sales over time rather than tracking individual people. It works across digital and offline channels and needs no cookies. If you want the basics first, start with our complete guide to marketing mix modeling. This page assumes them and goes into what is specific to marketplaces.
The short answer
Decide what pays: completed transactions, or the revenue you keep from them. Measure the buyer side and the seller side as separate outcomes when both are marketed, and set the budget for each by its own marginal return. Brand and direct traffic carry more of a marketplace's sales than in most businesses, so the model has to separate the baseline from the marketing with care.
The sections below go through the outcome choice, the gap between a request and a deal, the two sides of the market, brand in the baseline, and how to structure models across countries and categories.
GMV, revenue or completed deals?
Marketplaces have more candidate outcomes than most businesses, and the choice changes the answer. Gross merchandise value is the biggest number and the one most dashboards show, but the marketplace keeps only its take. Two categories with the same GMV can earn very different revenue, and a model on GMV will push budget toward the category with the lower take rate.
For budget decisions, the outcome should be what the marketplace earns: commission and fee revenue, or completed transactions when the take is roughly the same across them. Where margins differ a lot between categories or sides of the business, some marketplaces run two models side by side, one on volume and one on margin, and look at where they disagree.
| Outcome | Good for | Watch out for |
|---|---|---|
| GMV | Reading overall demand | Overweights categories where you keep little |
| Commission and fee revenue | Budget decisions | Needs revenue dated to the transaction |
| Completed transactions | When the take is similar across deals | Ignores deal size |
| Requests or listings | Early read, short delay | Many never become a deal |
A request is not a deal
On service and vehicle marketplaces, the marketing produces a request, a listing or a lead, and the money comes when a job is agreed or a car is sold. That can take days or weeks, and the share of requests that turn into deals varies by channel. A channel that brings many requests that never close looks excellent on request volume.
Where both steps are in the data, we take both in and build the first model on both, then decide which to carry forward, sometimes market by market. The delay between marketing and the deal is part of the setup. When a lead closes weeks later, the model credits the sale to whatever brought the lead in: if search brought the lead, search gets the sale.
Two sides, two outcomes
A marketplace spends to win buyers and to win sellers, and the two campaigns often run at the same time. Measured on total transactions alone, the model can't tell which side's spend did the work.
When both sides are marketed, each side gets its own outcome: new sellers or listings on the supply side, transactions on the demand side, each measured on the spend aimed at it. The budget for each side is then set by its own marginal return. Where supply limits growth, the seller side needs particular care, because buyers can only buy what is listed.
Brand and direct traffic
A well-known marketplace gets much of its traffic direct: people type the name, open the app or search for the brand. In a model, that traffic sits in the baseline, and the effect of the brand spend that built it hides there too. A model looking only at the coming weeks will tend to undervalue brand for exactly the businesses where brand matters most.
Two things help. TV and other brand channels drive extra searches and visits in the weeks after a burst, and that pull shows up in the model as their own effect. For the slower effect, the model can be extended with a brand component that captures sales arriving months later. Organic search and slow shifts in direct traffic are left to the trend. See how we model brand effects.
Countries and categories
Marketplaces that operate in several countries need one model per market, with the same setup in each so the numbers can be laid side by side. We build the largest market first. What it learns about each channel becomes the starting point for the next market, scaled to its size, and the local data then corrects it.
Categories are usually dimensions in the data rather than separate models. A marketplace with dozens of categories can't run a model for each, and most of its marketing serves the whole site. Large campaigns aimed at one category go in as their own channels where the spend allows.
Affiliates and price comparison
Product marketplaces often buy traffic from affiliates and price comparison sites. When those are paid per sale, the cost follows the sales, and a model that treats them as ordinary channels credits them with sales they did not cause. We keep pay-per-sale partners in the model at their historical spend and lock them in the recommendations. Placements bought as media are modeled like any other channel.
How Odins works for marketplaces
Odins is a managed marketing mix modeling platform. We connect the ad platforms and your transaction data, build and maintain one model per market, 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, Nettbil and Mittanbud run their marketing budgets on Odins. See the managed service.