ML and Anaplan – forecasting and the S&OP process in Greenfields

Greenfields is a leader in distributing exotic fruits and berries in Russia, selling the produce in top national and regional retailers as well as HORECA. Considering the nature of perishable products, growing forecast accuracy is a top priority for the business. Chiefly, this is crucial for a short-term sales forecast, which used to be generated by an outdated IT tool and demand planners.
Developing a forecasting model based on Machine learning algorithms was part of a bigger project meant to automate the Sales and Operation process (S&OP) with Anaplan, a renowned connected planning platform.

The Augmento team developed and launched a ML-enabled sales forecasting model for different forecasting horizons and an interface to manage the forecasts by the demand planning team:

• A multivariate ML model using various input data (historical sales, prices, calendar events, COVID-effect) and a combination of ML algorithms such as gradient boosting, regression, neural networks (multilayer perceptron), KNN.
• The Greenfields’ team now have two ML forecasts: an operational one for the next 12 weeks (by days) and rolling strategic one for the next 12 months.
• An interface in Anaplan to visualize, review, correct and approve the forecast in real time.
• Fully automated integration between all input data, the ML model and Anaplan to regularly refresh the forecast without involving the external data scientists.

• The forecast accuracy increased by 13 p.p. compared to the existing statistical forecast.
• The operational forecast accuracy grew compared to the final forecast, corrected by demand planning managers.
• The S&OP process got faster thanks to the ML forecasting tool and the overall process automation in Anaplan.