Enabling a Large FMCG company with demand forecasting models

DataSteroid

Enabling a Large FMCG company with demand forecasting models

Overview

Initiating a game-changing project with a prominent Fast-Moving Consumer Goods (FMCG) company, we aimed to devise a platform that could forecast demand for their e-commerce operations across platforms such as Amazon, Target, and Walmart, specifically for the US market. Demand forecasting is vital in retail, enabling category managers and merchandisers to discern customer demand patterns, streamline inventory, and make informed marketing decisions. The client was dissatisfied with Amazon’s existing demand forecasting and sought a more accurate and robust solution to harness their data and domain expertise fully.

Challenges

Inadequate Existing Platforms: The client desired a comprehensive platform to forecast demand for their e-commerce operations across multiple platforms for the US market. The existing demand forecasting systems, mainly Amazon's, fell short of their requirements.
Subpar Model Accuracy: Basic models offered only 60% accuracy, not meeting the client's standards for reliable demand prediction.
High Variability in Product Sales: High sales volatility across products added to the challenge, demanding a highly adaptive and precise forecasting model.

Solutions

Efficient Data Ingestion: We incorporated our solution, DataSteroid, into the custom-made platform deployed within the client’s infrastructure. DataSteroid effectively parsed and structured data into time series, partitioned by category, brand, and SKU. The data, sourced from the client's SQL server data lake, was preprocessed and cleaned for any inconsistencies and outliers.
Robust ML Models: The platform trained an array of univariate and multivariate forecasting models such as ARIMA, S-ARIMA, Prophet, LSTM, Random Forest, VAR, and Deep Learning. These models were fine-tuned to provide the required accuracy and correlations between demand, market spending, and allocation percentages.
Interactive Dashboarding: A lightweight application provided the final interface for the category managers to interact with the model data. It facilitated searches for categories, brands, or SKUs, defining forecast periods, viewing demand forecasts, and managing 'what-if' scenario components.

Result

Our custom-developed platform, incorporated with the immense capabilities of DataSteroid, delivered a significant boost in demand forecasting accuracy. The univariate models achieved 80% accuracy, while the multivariate models reached an impressive 90% accuracy. This precision improvement empowered the client to make superior planning decisions, reduce inventory costs, and execute effective marketing strategies. The client team received the lightweight application positively, utilizing it to guide their marketing decisions with newfound insight and precision.

DataSteroid

Product Boosters

DataSteroid