Retail marketing budgets, with the store sales in the model.

The sections below cover how the store effect is measured, how offline media gets into the same structure as digital, how to read a baseline that looks large, and how to handle TV when you know it does more than drive this month's sales.

Click attribution sees the webshop. For a chain where most of the turnover happens in stores, that means digital media is judged on a fraction of what it does. People search, compare and look at a product online, then buy it in the store, and the ad platforms never hear about it. Budgets set on those reports cut digital too hard.

Marketing mix modeling works on aggregated sales over time, so it needs no tracked click. When store sales are part of the outcome, the model sees them move with digital spend, whether or not anyone clicked. The same applies the other way: a TV campaign that lifts webshop orders is credited with them.

Total sales, stores and online together, give the cleanest answer to how much to spend. If you also need to know how much of a channel's effect lands in stores and how much online, store and online sales can be modeled as separate outcomes. Each split needs more data per model, so we usually start with the total and split when the question calls for it.

For most retailers, getting offline data right is the largest part of the setup. TV, radio, print and out-of-home live in media plans, spot reports and invoices from the media agency, in formats that differ by agency and by year.

We set up a pipeline for media plans and spot reports, from you or your media agency, so offline lands in the same structure as digital. Everything sits in one detailed hierarchy: market, product, then the media down to each campaign and spot, online and offline together. TV is broken down to the individual spot, print to the insertion date, out-of-home to the flight. With a good media agency, the history is in place within a day. See how we collect offline media.

Retailers often spend heavily on print, and print covers very different things. A weekly flyer full of prices works like a promotion in paper form: it drives store visits that week and little after. A brand campaign in a magazine works more slowly. Lumped together as one channel called print, they average into something that describes neither.

Where the spend allows, offer-led print and brand print go in as separate channels, so each gets its own response curve and carry-over. The same thinking applies to TV and out-of-home: what a campaign is for matters as much as where it runs.

Retailers are sometimes surprised by how much of their sales the model places in the baseline, the sales that would have happened without paid marketing. In grocery, most purchases happen anyway, so the baseline is most of sales. For businesses that sell mainly online, marketing drives a much larger share. A retailer with stores usually sits somewhere in between.

The share is a property of the business, not a grade on the marketing. A chain with a large baseline can be spending very well or very badly. The question the model answers is what the next part of the budget earns in each channel, and that does not depend on how big the share is.

Every variable added to a model takes some explanatory power away from the media channels, and anything the model uses to explain the past also has to be known when it forecasts. So we use few controls, and only the obvious ones with a direct effect.

• Distribution. New stores, closed stores and new sales channels change what marketing can sell, so the store count goes in.
• Special days and large campaigns. Black Friday, sales periods and large price campaigns are marked, so their lift is not credited to whatever media ran alongside.
• Stock. Periods of unusually high or low stock are recorded on the timeline. Stock would explain the past better, but future stock is unknown, so in the model it would weaken the forecasts.

Smaller, slower changes such as a competitor growing or organic search drifting are left to the trend, which keeps the model stable and its forecasts usable.

Many retailers know their TV does more than drive this month's sales. It keeps the brand in people's heads for the next season and the one after. When a model focused on the coming weeks suggests cutting TV, a common and sensible response is to lock TV at the planned level and let the model optimize everything else.

The scenario planner shows what that choice costs: the forecast with TV locked next to the forecast with TV free, so the trade-off is a number the leadership group can take a position on. TV's near-term pull on search and social over the following weeks already shows up as TV's own effect. For the slower brand effect, the model can be extended with a brand component that captures sales that arrive months later. See how we model brand effects.

Chains and countries usually behave differently enough to need their own model. We build one market or chain first, the largest. What that model learns about each channel, such as how fast it saturates and how long its effect lasts, becomes the starting point for the next one, scaled to its size, and that market's own data then corrects it. Products and categories stay dimensions in the data; one model per market on total sales is usually the right level for the budget decision.

Odins is a managed marketing mix modeling platform. We connect the ad platforms, set up the pipeline for offline media with you or your media agency, structure your store and online sales, 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 Megaflis, Høie and Vakre Vene run their marketing budgets on Odins. See the managed service.