Client background

• An iconic beer company in Australia
• Possesses a diverse portfolio of over 400 brands
• A collective strength of over 200,000 employees across 50 countries

Challenges

With significant cost pressures comes opportunity for more optimized revenue management. The client identified the following business requirements to manage one of their key revenue spend lever trade promotions:

• A platform with multi-causal prediction capability built on machine learning framework and ability to also forecast baseline sales
• Ability to support extensive data transformation requirements (Ex-factory data, Retailer POS data, Master Data, P&L inputs) in varying formats, hierarchy levels and frequency
• Capable of Seamless Integration with existing Trade promotion systems
• Ability for Baseline forecasting with Ex-factory data or limited data availability

Solution

The client signed up for the Data modeling module of the Promax Advanced TPM Platform for Baseline forecasting and Inventory management to enable optimizing their trade spend. The data modeling module integrated with the TPM system is based on machine learning framework with patented automated causal selector which can ensure high accurate predictions.

Uplift/Baseline model understanding:

• The SKU volumes and their prices are harmonized with the category level information, thus providing the dataset to be used for modeling
• The SKU trend is forecasted using a Time Series algorithm for a period of ‘n’ weeks and is stored against the week numbers
• The coefficients of the model are used to generate the baselines using shelf price and setting other causals to zero except trend and seasonality
• Another aspect which is unique to our solution is to adjust the predicted volumes using a “brand trajectory”, which helps in keeping the prediction in line with how the SKU is expected to grow with respect to the price changes and the market as a whole.

Uplift/Baseline model understanding:

• One can use the uplift models to generate baselines and compute the financials against each promotion that has been run to judge whether the promotion has achieved the planned revenue or not
• User can manipulate causals and generate what-if scenarios and judge the impact of the change and also the cost

Driven by the convincing results using machine learning algorithms from the initial POC, the client is now ready to deploy the solution for one of their biggest retailers. The POC phase provided a template design which can be adapted quickly for the new markets while ensuring high user adoption.

Business impact

The client decided on a 6 months POC to deploy the modeling engine for their Australia operations, initially to support one of their largest retailers and distributors, and then sequentially add more retailers/distributors over time.

The initial business benefits include:

The customers sees machine learning with predictive modelling as a way to change the types for conversations their teams are having. With this type of technique, teams can now take a longer-term view of their forecast, expanding their horizon. Sales teams are less focused on short-term tactical outcome, more on longer-term strategic thinking.

Pat Zalewski, Global Delivery Director