L’Oreal on the keys to success for marketing mix modelling

Vasileios Kourakis, Global Director for Marketing effectiveness (ROI) and Media in the consumer products division at L'Oréal, spoke to WARC about keys to success, building a measurement toolkit and making decisions with effectiveness in mind.

What role does marketing mix modelling play in your measurement and what challenges does it help you overcome?

MMM is one of the tools we utilise (among many in our complementary toolkit) to answer both strategic as well as operational questions for the short and the longer term. It helps us understand what the return on investment of our Advertising & Promotion across categories is and where we need to strategically allocate resources or protect value for our key beauty categories in a given period.

At a more operational level, it helps our local teams understand campaign effectiveness, and media best practices (reach, frequency, targeting, saturation points etc) and ultimately optimize budget allocation across all touchpoints for online vs offline channels. So ultimately it helps us make more data driven informed decisions and provides accountability in the short and the longer term.

What has your journey to supercharge your MMM program looked like? From inputs, to insights, to final impact, what have been the keys that drove success as you build out a reliable model?

Building reliable and timely models consists of four key stages in my opinion. Scoping is the first and most important stage. Clarity on the key questions from the beginning, helps structure the models to answer those questions effectively: ‘‘To ask the right question is already half the solution to a problem’’ said Carl Jung. Holistic Data collection and data automation is the second most important element, as more granular and higher frequency data delivers better insights and more robust outputs.

Nowadays the complexity of the touchpoint landscape requires more data collection as new touchpoints have emerged such as virtual try ons, ratings and reviews, influencer activations etc and as the majority of time is spent on data collection, the more automation the faster the delivery. Modeling online and offline sales by Promoted Product Group / Stock Keeping Unit is the third most important stage of the process where advanced analytical techniques such as Bayesian can be critical in an ever-evolving landscape.

We have developed a compliance guideline to ensure best practice modeling and model robustness for in house or outsourced modeling work. Lastly, having a harmonized taxonomy and reporting is the last part of the equation to keep things simple and easily digestible to a wide non-technical audience, while ensuring consistent benchmarking globally.

How does L’Oréal account for e-commerce sales within MMMs?

E-commerce sales can be a key channel depending on the market. In the case of China, e-commerce takes a significant share, while in some European markets it is growing but remains a smaller channel compared to offline sales. Nevertheless, it is important and we include it in our MMMs. In markets where we have brick & mortar retailer e-commerce sales data by week, we include it in the MMM either as part of the brick & mortar model or preferably as a separate e-commerce model (depending on the size and data availability from the respective retailer). In cases where we also have pure player presence and sales data available, we model it separately, such as Amazon. 

What challenges did you face when initiating e-commerce models or accessing accurate data, if any?

Accessing granular e-commerce sales data by SKU, by week at the geo-level, is one of the main challenges for e-commerce models. This requires working with several retailers or data vendors across markets such as Profitero. We also need e-commerce Advertising & Promotion as an input in the model which adds an additional challenge as this data is not always well tracked or readily available. Modeling e-commerce sales data is also more challenging than offline sales data as they can be more volatile. Therefore initiating models can take longer for developing markets and requires more validation on top of an extra cost (modeling fee).

What's your advice for marketers who are looking to use MMM? What does it take to get MMM right? How can marketers ensure they're using it effectively?

The main advice I have is that they need to understand that MMM can be a great tool for strategic direction and should be used like that. But it is not the holy grail to all marketing measurement questions. It should be part of a broader complementary measurement toolkit.

In addition, more value can be achieved if it's run by experts in the team who focus on best practices and influence. Granular data inputs are key for robust modeling and significant resource needs to be allocated to data collection and governance as we explained earlier. The more granularity, the better, such as store level sales models. Having a common framework and taxonomy to enable consistency and benchmarking is also paramount.

I recommend having a strong measurement partner relationship that understands your business model and needs. Publisher support is also important to drive data granularity, deeper insights, plus innovation and last but not least very good integration with the Media agency through the whole process.

How has your MMM strategy evolved since partnering with Amazon Ads?

A few years ago, Amazon sales models were at national level but as we have increased our collaboration with Amazon Ads we have moved towards geo-level data inputs and stronger Amazon models like we do with our offline retailers where data was always good. That way our model robustness has increased and so has our Amazon Media ROI accuracy. In addition, our measurement providers have direct access to the Amazon sales, media and promotional data by ASIN and by region directly from Amazon. This considerably speeds up the data collection process and lightens resources from our side.

Have you been able to dive deeper into your Amazon Ads performance? Are there any learnings you’ve achieved across specific tactics, ad products, etc?

Historical analysis was predominantly focused on Amazon media impact and ROI on Online and Offline sales channels. However, as we start getting more granular data on media channels, we started exploring a framework for Amazon deep dives. These deep dives focus on Amazon media campaign level best practices such as reach, frequency, Hero non Hero, 1p data, etc. These are all part of a global learning agenda that together we try to answer during a year.

Has including your Amazon retail data in your models helped improve your models overall?

Historically we only had national Amazon media and Amazon sales data. More recently we have moved to ASIN and geo-level data. Plus we can now access Amazon pricing and promotion data which really improves the model statistics and robustness. Furthermore, accessing Amazon media data at campaign level (Reach, Frequency, Sponsored Brands vs Products, Branded vs Generic etc) enables us to derive deeper insights for further campaign executional optimization.

Are there any improvements in ROI/performance you could share since initiating your collaboration with Amazon?

We constantly strive to increase our ROIs, either by allocating resources to more effective and efficient channels or by planning campaigns with best practice guidelines in mind. Our models have become more robust, our insights richer and our mix is working harder.