Modelling the omnichannel media plan to find the optimal investments.

A decision-making tool for a media plan optimised for performance.
La Redoute has been making the French art de vivre accessible, in fashion and in home, for a hundred and eighty years. From the wool mill to the legendary catalogue, the retailer is today one of Switzerland’s e-commerce leaders in both of its worlds.
In 2022 it asked us to model its media mix. The question was not to measure one channel, but to know what each really brings to transactions — online and offline, and separately on two language markets that do not answer the same way.
The context
For 180 years, La Redoute has made the French art of living, in fashion and for the home, accessible to as many people as possible. La Redoute is a timeless company, a story across generations founded on creation and constant innovation. From the wool mill to the legendary catalogue, La Redoute is today one of Switzerland’s leading e-commerce names in fashion and home.
01Challenge
Knowing
which lever
really pays
For a well-established retailer, the question is not only how to acquire new customers, but how to work its own again. We therefore had to identify the media levers that pay best, online and offline, and spot those getting too much or too little — to reallocate a budget rather than raise it.

Pillar 1
Beyond conventional attribution
Put in place a media mix model, which measures offline channels as well as digital ones.
Pillar 2
Valuing the customer over time
Calculate lifetime value by segment, and predict churn.
Pillar 3
Moving the investments
Shift budgets according to the real performance of each lever.


Close to one transaction in two comes from media
Transactions in French-speaking Switzerland, decomposed by the Marketing Mix Modeling model between baseline, CRM and media, from 15 December 2021 to 30 September 2022. Illustrated with fictitious data. Choose a lever and the unit.
Each bar starts where the previous one stops: together, the three levers make 100% of the period’s transactions.
Source: fictitious data for illustration, Bright Marketing Mix Modeling report · Chart: bright.swiss

02Strategy
Two markets,
two
models
A media mix that is optimal in French-speaking Switzerland is not optimal in the German-speaking part: the same francs do not produce the same transactions. Each market therefore got its own model, with its own variables.
The method
It matters to assess the impact of marketing channels separately by language region, because experience shows the optimal marketing mix can differ between them.
On that, the MMM (Marketing Mix Modeling) model we use is effective at identifying:
Pillar 1
The volume of conversions generated by each media channel.
Pillar 2
The real cost per acquisition (CPA) for each media channel, online and offline.
Pillar 3
The optimal budget per channel, to maximise the volume of transactions.
Pillar 4
The media investment that can be reduced if the annual budget falls.
The results of the model also let marketing intuitions be compared with what the data concludes, and set against each other. Where the divergences are large, it is essential to understand why. From there, certain areas for improvement can be identified, which lets the following year’s investment be adjusted accordingly.
For a company based in French-speaking Switzerland, the German-speaking market calls for a different strategy, specific to that market. Several models therefore have to be built, so as to take the cultural differences into account and bring in further, distinguishing variables for each market. That lets us produce a recommendation that is realistic and that takes each market’s particulars into account.
We fit our package to our clients’ business criteria. For La Redoute, we built our framework using, among other things, algorithms from Meta’s open-source package “Robyn”. The tool was developed by analysts and engineers at Meta and by the community — of which we are part. It offers, as things stand, mature algorithms for this kind of analysis.
Implementing the MMM project for La Redoute followed a process close to that of a technical project:
- Framing and design: examine the business model, the mechanisms that contribute to generating revenue, calculate the LTV (which we then broke down across cohorts), and so on. The aim being a design faithful to, and cut for, the client’s business.
- Setting the objectives and what is at stake, so as to deliver a result that is concrete and can be acted on.
- Collecting transactional (anonymous) and budget data (media spend).
- Interpreting and reviewing the data, so as to isolate and understand what can be used.
- Iterative modelling.
- Analysis, scenarios and recommendations
03Impact
A tool
for making
the decision
The recommendation delivered gives, lever by lever, the optimal investment — and a second scenario for the day the budget falls. It confirmed some of the retailer’s intuitions, corrected others, and brought out the role of CRM in the purchase journey.
What was delivered
Following the modelling, we delivered a recommendation setting out the optimal investment for every media lever in La Redoute’s marketing mix, online and offline.
To make sure our recommendations hold, we put the model to the test through a backtesting phase. Since the model is built to reproduce as precisely as possible the impact of spend across channels on sales, we generate data for the period analysed — following the model’s characteristics — and compare it with the real data.
The result is an index called R2, which we can read as the model’s precision, or dependability. Above 75%, we can consider the modelling relatively dependable. For La Redoute, every model built had an R2 above 80%.
We also proposed an optimal investment scenario on the assumption that La Redoute had to, or chose to, reduce its media investment — the aim being to limit any fall in revenue.
La Redoute commissioned us to analyse its media mix so as to map where the transactions come from, revealing which channels matter most and allowing budget to be reallocated towards those that are under-invested. What we delivered — a genuine decision-making tool — confirmed La Redoute’s intuitions on parts of its acquisition strategy, validated the soundness of its media strategy, and also gave useful business insight, notably on the role of CRM in the customer’s path to purchase.
La Redoute puts the MMM results to work by adjusting its media investment, so as to hold a media plan that is coherent and as effective as it can be.
- > 80%
- R², the measured reliability of the models delivered
- 2
- markets modelled separately, French- and German-speaking
- 6
- steps, from framing to recommendation

During the Marketing Mix Modeling study, Bright listened closely, developing a fine understanding of our business model and so of what is strategically at stake for us.
We spoke with the team regularly, so that we could interpret the study properly and turn its results into operational objectives. Working together was most pleasant, and instructive. Thank you.
Caroline Descamps
CMO – La Redoute Suisse
Our R&D
In a landscape where the channels multiply and tracking capability erodes, measuring what campaigns really achieve has become both critical and complex. Marketing mix modelling brings a rational answer to that complexity, drawing on the data to inform decisions — never replacing intuition or human expertise, but completing it intelligently.
For more than five years we have been running active R&D on the most advanced MMM frameworks on the market: Robyn, Lightweight, Simba and others. That groundwork lets us offer bespoke models today, fitted to each client’s business and to the reality of their sector. We do not sell a single solution; we build decision-making tools that are relevant, nimble and effective. Because understanding better is deciding better.
The expertise involved
Retraites PopulairesRetraites Populaires, brand firstHitting the lead targets while establishing the recognition of the brand and its products, all within a fixed dedicated budget.
SkySky Show, acquisition under controlGenerating growth in German-speaking Switzerland while keeping acquisition costs in hand.
VeepeeVeepee, data that makes it growDescriptive and statistical analysis of anonymised customer data, to build a set of key indicators for tracking performance.

