Optimize your resource allocation using a new decision science software – Epilogi® – our Budget allocation tool

Guillaume Roger

Guillaume Roger

Product officer

In global companies, markets currently divide up the Advertising and Promotions investments at their discretion or based on their business sense.
For Advertising investments to be effective, companies must spend on the right brands in the right countries and then convey the most compelling messages through the right channels to reach the most valuable consumers/shoppers.

The traditional budget breakdown suffers from several shortcomings:

  1. It gives limited consideration to the significant differences in ad/promo intensity and sensitivity that often exist given market context and brand growth situation
  2. It fails to recognize the fundamental choices needing to be made to fuel growth, maintenance, and harvest strategies for each brand, segment, and country.
  3. Market share and the impact of support investments on profits are not measured, and as a result, market share objectives are not linked to Ad budgets.

Based on our work with various FMCG leaders, we have developed Epilogi®  a new approach for a budget breakdown based on AI and Machine learning that   

  • Combines a multitude of databases to understand the historical market performance, and 100+ industry, consumer and macroeconomic drivers, to define parameters of growth
  • Applies machine learning and time series algorithms testing of more options than traditional analytics, and make the process much more fluid and transparent
  • Replaces simplistic forecasting with powerful AI algorithms that accurately and quickly predict market shifts on the country and product level
  • Create a common base of comparison between markets to rationalize the investment choices.

Business leaders face many challenges today, including high growth expectations, cutthroat competition, and the digital and social-media revolution. Embracing A&P optimisation as a discipline can help build strong brands that generate improved returns.

Optimization is one of the key elements of our economy: its objective is to produce more value (be it material, social, or financial for instance) out of a process, while consuming exactly the right amount of resources needed (be it money, time, or materials for instance).

The resource allocation challenge can be found everywhere: from managing a portfolio of investments to allocating a budget to manage your own personal agenda. The complexity of deciding what is right requires to perfectly understand and control what are the key drivers behind the choices and what could influence those drivers in our environment.

With digital revolution, it is now easier than before to collect very precise data on surrounding phenomenon’s. Literature has shown over the years that the best decisions are made by human guided by deep dive analysis, but not dictated by them.

One of our FMCG client wanted to understand and develop a rationale way to divide up budget across countries (more than 20+) and brands(more than 10+). At Pivot & Co, we’ve developed a combination of machine learning, software technologies and human behavioural sciences to build Epilogi®,

Epilogi® is an AI-based platform that

  • Combines a multitude of databases to understand the historical market performance, and 100+ industry, consumer and macroeconomic drivers, to define parameters of growth
  • Applies machine learning and scoring techniques testing of more options than traditional analytics, and make the process much more fluid and transparent
  • Replaces simplistic forecasting with powerful AI-algorithms that accurately and quickly predict market shifts on country and product level
  • Create a common base of comparison in between markets to rationalize the investment choices.

We disagree. We think in spite of tiny variations, consumer demand are perfectly predictable if we use the  right model and we take into account the economical environment.

The secret is to have the right toolbox and to be clear on the “what for” question. The usage and outcome of forecasting models as well as techniques used depend on the actions that business leaders want to undertake. Short term resilience of preparing the long term turn around.

Too often, companies are confused. They spend a lot of money with traditional providers (IRI, Nielsen, etc…) and don’t leverage the value of Machine Learning and big data capable to design complex combinations and analyze what they should do to capture growth opportunities.

One consequence of that is that  Executives spend 40% of their time making decisions, about the future and they tend to assess that most of it is poorly used.

Screenshot from Epilogi Dashboard, presenting Region Level results per brand (anonymized data sample)

It is not just about being right, it is about changing how we look things and leverage the data culture debt to design the right decision model

The data culture debt

Every transformation process starts by wondering about the starting point. When it comes to decision science, one must keep in mind that some people are culturally not really used to managing their operations using data & analytics insights. It doesn’t mean they can’t use it, it means the way to present and to use data & analytics insight has to be designed  in such way that it encompasses inherits habits whilst deriving and conducting a new way of working.

Design a computation model that shift the user perspective

Computation models are tricky: people always knows its business better than ourselves. However, we better know how to turn his experience and business insights into data science models that serve its predefined purpose

Epilogi®, A&P guidance softwate use case  


Recreating how humans think

For a global company with several brands (10+) in the cheese industry, produced and distributed in more than 20 countries across the globe. Every year, the global marketing team must define what A&P budget should every country receive to support maximum global performance in its yearly activities.

This maximum global performance aims at maximizing profit and supporting mid-term of long-term brand and/or market performance. The allocation process goes as following :

  • Every country is assigned a role to play to support the global strategy.
  • A global A&P budget is defined at the group level…
  • … and then the global team must split it between different countries.

This process used to be supported by some KPIs (net sales margin, market shares…) but was mostly done using experience or beliefs from the people around the table.

Obviously, in those conditions, it was difficult to get closer to optimization : every country is a different market, and is influenced by many drivers that are predicable, unpredictable, or unknown.

Challenges to face

One can propose the most relevant data analysis and push recommendations, but it will not be enough. Here are some of the reason why:

  • The statistician curse: the interlocutor doesn’t really understand what’s behind the data science, and is left alone.
  • The factory and the foreman: the results do not support the strategy that the interlocutor had in mind, and he/she rejects it.
  • The convenient truth: one might focus more on the on the design of the results presentation, without really displaying the important drivers behind the truth.

Creating decision algorithms with LSTM and gradient boosting

Our objective was to optimize this decision process using the most as possible relevant data analysis. Pivot & Co worked closely with its client to design based on machine learning models what could be the best way to optimise the A&P budget allocation decision of the top management.

The question that lies behind every A&P budget allocation process is how does every country performs when it comes to A&P, and what is the right amount to allocate without jeopardizing the rest of the portfolio?

Consequently, the computation model should be designed to:  

  • Score every country, and every brand per country, on its A&P performance.
  • Based on those scores and on the strategic roles those countries have, generate what could be the best trade-off.

The A&P performance of a country is driven by many different drivers: some are known, some are unknown, some are controllable, and some aren’t.

The scoring model is consequently designed to use a list of internal features (like Net Sales, Sales Margin, A&P budget, innovation rate…) and external features (unemployment rate, retail sales growth…), and to compute what is the impact of each feature on the A&P performance, while accepting some uncertainty (for uncontrollable and unknown drivers).

Using a LSTM machine learning approach and gradient boosting techniques, we were able to train a model that will be able to compute coefficients for every countries and every brands : the correlation between the features and the coefficients will generate final scores

The decision software

The project objective was to allow one user to generate those scores and to receive A&P budget allocation recommendations. To do so, in line with the previous approaches explained, an interactive dashboard has been designed to present the recommendations and allow the user to explore those results.

Screenshot from Epilogi Dashboard, presenting country Level results per brand (anonymized data sample)

The impact  

Using Epilogi, global managers rely effortlessly on relevant KPIs that could drive their decisions : uncertainty, sensitivity, risks… and what required long and complex analysis now is lying some clicks ahead.

Eventually, the ability to drive decisions with those KPIs also feeds the discussions between global team and different markets, or between different functions within the company.

This new “way of business management” eventually spread within the different layers of the company : every resource allocation process will require a more data driven approach, as people are getting accustomed to this new way of doping.

It opens new dimensions for similar approaches:

  • How can a country allocates its budget between its different brands ?
  • How can one country choose the best touchpoints for advertising campaigns ?