Preparing for the storm: predicting share risk for Coca-Cola
For questions about this case study contact Andrew Glor.
What does AI have to do with it?
Whether you’re in favour or against it, AI is a hot topic. It has permeated most industries and its application has encouraged innovative approaches.
There are clearly opportunities for AI in marketing, but can it be applied to brand strategy?
Every good strategy starts with some objectives and ends with assessing whether you’ve achieved them. But understanding how to track your performance isn’t straightforward. As many other brands do, Coca-Cola faces some difficult questions: which KPIs matter, for which brands? And are the metrics tracked today going to propel the brand in the future?
“One of the responsibilities of the global team at Coca-Cola is to identify opportunities and gaps to the overall company’s growth model and provide guidance and learnings across our system to grow faster and better” says Brian Boever Director of Human Insights at Coca-Cola.
At Foresight Strategy we know that KPIs should be fact-based, dynamic, tailored, and forward-looking. So, this is a story about how we collaborated with Coca-Cola to make sense of the hundreds of metrics they tracked, predict what will matter in the future, and understand whether AI could help them identify the right KPIs.
A unique model for brands
In this pilot we used a variety of data: point of sale, consumer panel data, usage and attitude surveys, brand equity studies, category trends and macro-factors.
We started by defining the objectives and by cleaning the data. The modelling included a few steps such as brand classification, clustering, and Neural Networks which helped predict whether a brand share would increase or decrease in the future.
This resulted in a unique model for each existing brand with the best forward looking predictors of share gain based on tailored KPIs. Our model showed that different metrics really do matter for different brands. For example, the most relevant KPI for one brand was Price index to category and Yearly+ consumers, while for another was number of displays, or total distribution for yet another. Our system correctly predicted 86% of share increase or decrease in the test data set.
The goal of this pilot was to develop a generalized pattern that refined itself every time it ran, with a constantly changing dataset, and minimal human intervention. The AI helped us code complex models, get results with speed and efficiency, and build a platform that worked with various data types.
You’d take an umbrella if you knew it was going to rain
“While we look at data across our top 40 markets, we need to be able to identify issues and talk to the relevant team faster to help them adjust” says Brian “What are the key metrics that are going to help us understand the change in share? What are the key indicators that will impact the share performance in the next three months?”
One of the first uses of the dashboard is a loss or gain share indicator and how significant the loss or gain is.
“We thought to operationalize this as early warning system. Like the weather forecast can predict storms to help people prepare, we can predict loss of share to help brands fix it” explains Andrew Glor, Partner at Foresight Strategy.
“The global team can look at one brand across all markets and understand the share perspective for the next three months. Thanks to this, we can identify the markets that we need to pay attention to and talk to the leads in those markets to understand what to do to help alleviate those losses” says Brian.
How would this pilot come to life? When turned from vision to reality “Anybody could have access to the dashboard and be notified with updates on share losses or gains”. It could also be possible to use the tool to understand what are the key things that are going to impact share in the near future to build a plan against these metrics. The marketing teams would also have access to these data to make more strategic decisions in the market in the short term.
What’s the future holding?
The model could continue to become even more valuable, expanding the data set, incorporating regional factors, and non-numeric types of data that could turn into KPIs.
The AI hype is real, and it can be applied in brand strategy to improve the way KPIs are set and used. “This type of modelling is a lot more customizable by brand and market” – says Brian – “it better predicts the future and helps us be quicker to react, supporting a more proactive decision making. What is going to happen? We can change the future…to an extent!”