Marketing mix modelling for different business models

Marketing mix modelling works the same way in every industry: it relates marketing spend over time to an outcome, after separating out what would have happened anyway. What changes from one business model to the next is which outcome you model, how long it takes marketing to show up in it, what else moves it and how the business is structured. Getting those four choices right matters more than the choice of algorithm. This guide goes through retail, financial services and B2B, subscriptions, and promotion-driven businesses such as grocery and travel.

The short answer

Any business with a marketing budget above about EUR 1 million a year and a sales signal the model can learn from can use MMM. The work is in setting it up for your business model:

• The outcome: the result that matters to the business, such as total sales, paid-out loans or new subscriptions, not the easiest number to count.
• The delay: how many days or weeks pass between the marketing and the outcome.
• The controls: prices, promotions, launches and shocks that move the outcome without marketing.
• The structure: markets, brands and sales channels, and whether they need separate models.

Retail and e-commerce: stores, webshop, marketplaces and wholesale

Retailers usually ask one question first: does online advertising drive sales in our stores? MMM can answer it, because it works on aggregated sales rather than tracked clicks. If store sales are part of the outcome, the model sees when they move with digital spend, whether or not anyone clicked an ad.

The setup choices that matter:

• Which sales to model. A model on total sales (stores plus online) gives the cleanest answer to "how much should we spend?". The honest limit is that it does not tell you how much of each channel's effect landed in stores and how much online. If that split matters, store and online sales can be modelled as separate outcomes, at the cost of more data per model.
• Marketplaces and wholesale. If you sell through Amazon, other marketplaces or retailers, include those sales in the outcome when you can get them, at least weekly. Marketing that drives demand there is otherwise counted as zero. Retail media on those platforms is a channel like any other.
• Several brands or countries. Brands and markets usually behave differently enough to need their own model. A practical route is to build the largest market or brand first and carry what it teaches across to the rest.
• Controls. Price changes, campaigns and sales periods, store openings and stock-outs all move sales. Include the large ones, or the model will credit them to whatever channel happened to run at the same time.

Expect the share of sales that marketing explains to vary a lot. For grocery and other businesses where most purchases happen anyway, the baseline is most of sales. For businesses that sell mainly online, marketing drives a much larger share. Neither is a verdict on how good the marketing is.

Banking, insurance, lending and B2B: long sales with several steps

Financial services and B2B companies often doubt that MMM fits them, because the sale happens weeks after the marketing and passes through several steps: an application, a credit check, an offer, an acceptance, a payout. It fits well, with three adjustments.

• Model the outcome that pays. Pick the step closest to revenue that still has enough volume to learn from. For a lender that might be paid-out loans, approved loans or completed applications, not clicks on the form. For a B2B company, qualified pipeline or won deals, not raw leads.
• Set the delay. The number of days between marketing and outcome is part of the model setup, so cause and effect line up. When a later step closes, the model credits whichever driver started it.
• Include what is not marketing. Changes to the application process, rate changes, adviser capacity and sales team activity all move results. They belong in the model as controls, or the model will attribute them to media.

Sales through advisers or branches belong in the outcome if marketing influences them. And because these businesses often have fewer, larger conversions, priors matter more: what you already know about how your channels work helps the model reach stable estimates on less data.

The approach works. Aprila Bank, a Norwegian challenger bank, got 37 percent more loan applications from the same budget after reallocating spend based on its Odins model. Read the Aprila Bank case study.

For B2B SaaS with very long cycles and few deals a month, be realistic. MMM needs enough outcomes to learn from. Modelling an earlier stage, such as qualified opportunities, with a longer delay is often the workable choice. Fintech apps can model new active users in the same way.

Subscription businesses: telecom, streaming, energy

For subscription businesses, model what marketing actually drives: new subscriptions or gross additions. Churn and the revenue of existing customers are mostly driven by price, product and service, not by this month's media. Keeping them out of the outcome keeps the model honest.

• Offers as controls. Introductory prices, bundles and campaign periods change conversion rates. Mark them, so the model does not mistake a discount for a channel effect.
• Seasonality that drifts. Peak seasons shift as a market matures and competitors act. A model whose seasonality and channel response can change over time handles this far better than one that assumes every year looks the same. We explain why in time-varying saturation.
• Brand building. Subscription brands often spend heavily on brand. Its effect builds slowly and carries over for weeks. See how we approach brand effects in the model.

Promotion-driven and high-frequency businesses: grocery and travel

Grocery retailers run new promotions every week, and travel companies sell months before the trip. Both can use MMM, with care.

• Weekly promotions belong in the model as control variables, with their depth and reach. Otherwise the channels that happened to run alongside a big promotion get its credit.
• Loyalty data is useful as an outcome or a control when it is consistent over time, for example active members or member sales per week.
• Bookings or travel dates. Travel companies need to decide whether to model bookings, which react to marketing, or travel dates, which drive revenue. It is often worth taking both in and choosing per market.
• Shocks. Weather, strikes, outages and other events that marketing did not cause are flagged in the history so the model does not learn from them, and they are not forecast forward.

Questions to ask a vendor about your business model

• Can you model our real outcome, with the delay it has, rather than clicks or leads?
• Can store, marketplace, wholesale or adviser sales be part of the outcome?
• How do you handle several brands or countries?
• Which controls will you include, and how do you treat promotions and one-off shocks?
• Can seasonality and channel response change over time?

Odins builds and runs these models for retailers, e-commerce companies, marketplaces and banks. Companies like CDON, Nettbil, Hyre and Megaflis run their marketing budgets on Odins. For the data side of the setup, read what data marketing mix modelling needs.

Frequently asked questions

Which MMM platforms work for a bank or insurance company with long application processes?

Look for a vendor that can model the final outcome, such as paid-out loans or signed policies, with a configurable delay, and that includes non-marketing drivers like rate changes and process changes. Odins does this for banks and lenders such as Aprila Bank.

Does marketing mix modelling work for B2B with long sales cycles?

Yes, if there are enough outcomes to learn from. Model qualified pipeline or opportunities with a delay rather than closed deals when deals are few, and include sales activity as a control.

How do we measure how online advertising drives sales in physical stores?

Include store sales in the outcome the model explains. Because MMM works on aggregated sales over time, it captures store effects from online media without any tracking of individual customers.

Which MMM tools suit a grocery retailer with weekly promotions?

Tools that let you add promotions as control variables, flag one-off shocks and model baseline sales conservatively. In grocery, most sales are baseline, so the marketing effect is a smaller share and needs careful separation.

Start with the full guide to marketing mix modelling, or book a demo to see a model for a business like yours.