Marketing mix modelling: 3 case studies from 2020


28 Sep
28Sep

Introduction

My last blog was a review of my year with my business owner hat on, but what of the work that keeps that business ticking over?

The work that I have undertaken in 2020 has all turned out to be marketing mix modelling, by chance and circumstance rather than by design. But it has provided a useful opportunity to showcase the different ways that a consultant with my skills can help. It is also a good illustration of the variety of ways that the same approach is used for different end goals. Businesses remain anonymous for obvious reasons.

 

1. Tiny cog, big machine

The whole engagement is vast, and the ultimate output is a tool that the client uses to make a first pass at global marketing spend allocation by country, brand, and channel. The aim of these marketing mix models is to produce the response curves that fuel the allocation tool’s engine. I have written more about how that works here.

There are two main implications of this type of engagement. First, the number of models that needs to be produced runs to several hundred. Second, because countries, brands and channels are compared it is important that the models are structured similarly. Simplicity and consistency win out over complexity and uniqueness.

All the time in my book.

Thankfully, these days the heavy lifting in terms of data processing can be largely automated and initial models are built quickly to a template. But there are some countries where the data is not so good or complete. And others where the marketing has such a small relative impact that some judgement calls needed to be made.

It was those models that I worked on and other than to understand how it was put together I was not involved in the data processing nor the steps after the model was built. It freed up some senior time in the wider team to enable oversight rather than them stepping down into the nuts and bolts too often and losing that widescreen vision.


Global consumer packaged goods brand

My contribution: 13 countries, 25 brands

3-month attachment to team


2. Specialist subject

Another large project. In addition to marketing response curves the client delivery also involved presentations to marketing and consumer teams. With the purpose of explaining sales movements, diving into more detail about campaigns and providing recommendations on how to improve the efficiency of activities.

Like in the first case study, data had already been processed. But lockdown in the UK scattered the core team, so they sought out experienced hands who could take on chunks of the delivery and see it through right from exploratory analysis through to presenting to the client. I took a set of brands that sit in the same product category.

All of the models were ‘refreshes’ of existing models rather than new builds. One of the skills that experience gifts analysts is the ability to judge how much change should be expected versus previous results. With both the analytical and client teams all working remotely, clear communication was also needed.

The models were based on data pre-Covid and results were delivered while most countries were still in some sort of lockdown. Part of the discussion with clients was then naturally about whether recommendations still held. If not, what factors need to be considered for immediate decisions.


Global beverages brand

My contribution: 3 countries, 3 brands

3-month attachment to team

 

3. The Full Monty

A smaller business, but still multiple countries – a theme than runs through my career. Me operating independently to provide B2B consultancy. The client had never commissioned marketing mix modelling before, and the primary questions were about measuring the return on investment of the two main paid marketing channels.

Every stage of the engagement was undertaken by me with the cooperation of the client and data owners within the business. The process took 10 weeks following these steps:


  • Initial briefing à Proposed solution
  • Data request issued à Data collation and processing
  • Feasibility study including go/no-go recommendation
  • Exploratory data analysis à Initial model builds
  • Interim results à Feedback à Model refinement
  • Final model builds à Final results & recommendations
  • Delivery of assets: Models, Dataset, Forecasting tool


The model period included Covid-19 first waves and so the scale and speed of sales erosion and recovery due only to the pandemic was measured. The trough was driven largely by lockdowns with “uncertainty” (proxied by death rates) playing a smaller role. There was a sales bounce due to pent up demand, that settles at a slightly lower level.

There was no evidence that marketing worked differently during the lockdown and so it became a significant driver of the low sales during that period. The importance of connecting to customers was also highlighted with a peak in incremental sales driven by the website chat function during lockdown.

Both main paid marketing channels were working to drive sales but over different timeframes, which led to recommending that both should be continued. The channel where most money is spent experiences diminishing returns at higher spend levels, which led to recommending the most efficient spend ranges for each country.

The effect of paid marketing is not visible by looking at sales over time. Only by using a modelling approach to strip out other impacts was the marketing impact identified. These other impacts included: industry trends, underlying seasonality, changes to the businesses’ service offer and non-controllable factors like search algorithm changes.


Automotive

2 countries, 1 brand

2-month consultancy

 

Summary

About the only threads loosely tying these three engagements together is that the same analytical approach was used and that two ran headlong into Covid-19. The scale and the scope of each engagement were very different and show how versatile an approach marketing mix modelling can be.

‘Models’ were not the goal, just a means to get to business recommendations. To get those models and to translate the outputs needed as much attention and skill directed to data analysis (structured ‘looking’) and communication (effective ‘talking’) as to statistics.

And careful thought around the mechanics and implications of an unprecedented event.

If you would like my experienced eye on your marketing effectiveness, get in touch.

Please ask before reproducing my material partially or wholly for commercial use.
 © Jo Gordon Consulting Ltd 2020

Comments
* The email will not be published on the website.