The client is a UK-based leading consumer goods company delivering health, hygiene and nutrition products across hundreds of product lines to global markets. In an increasingly competitive and volatile market, the company manages a complex supply chain network of 40,000+ individual retailers across 200 countries.
The company was using a demand forecasting model to predict weekly demand, plan raw material procurement and schedule production. To feed in historical data, they managed multiple data pools in SAP and JDE systems. Time series data models like ARIMA, STLARIMA were used to predict demand for each SKU. By considering parameters like sales, trends and seasonality, their system generated reports for forecast quantity vs. actual sales in a production line.
However, these reports were erroneous, incomplete and conflicting across functions due to lack of visibility into system metrics like logistics center and stock level.
The company needed to strengthen its forecast model and make it more robust to ensure enhanced forecast accuracy and timely responses to critical operational concerns like inventory optimization, procurement of raw materials, and stock visibility.
A root-cause analysis into the problem statement by Hitech data scientists and a review of the existing forecast model threw up the following challenges:
- Limited visibility of data across product lines to make accurate forecasts for existing and new product lines.
- Inaccurate connect between marketing insights, customer databases, supply systems and product lines i.e. no mapping between the data.
- Ineffective information system to match varying demand patterns in different markets like US and UK.
- Inability to make dynamic changes to the data models due to lack of information such as seasonal parameters, volume sold on deals (VSOD) and depth of discount (DOD).
- Inaccurate or inefficient inventory data, analytical tools and inability to decipher data to maintain product standards.
Delivered customized data solution using data mapping of sales, brand and promotional data for seamless integration of all supply chain operations including product availability, management and operations-right from suppliers to customers. Also provided analytical inputs to augment and enhance their existing data models through data pre-processing (data cleansing and standardization) and forecasting solutions.
The solution helped to:
- Collect legacy demand data to map different source systems to get better understanding of trend, seasonal, cyclical and random components of the time series data that contribute to better future demand forecast.
- Execute various time series forecasting algorithms and choose best algorithm using ranking approach from measure of error parameter i.e. MAD, MAPE and RMSE.
- Harmonize code to easily scale up for different countries with miniscule changes.
- Accurate monitoring of all products at stock levels from weekly reports of the forecast results.
- Hitech data analysts and scientists assessed the existing algorithms and interacted with the brand managers to gather and understand data, its sources and business needs.
- Checked deficiency in algorithmic data and identified course-correction for it i.e. data cross validation and feature selection which adds value to the forecast.
- Our data scientists helped to:
- Pace up existing datasets using data pre-processing and then apply classical time series algorithms for better forecast accuracy.
- Establish automated smart forecasting support to interact and engage with existing applications and manage forecast at store level.
- After consistent accurate forecasts in the USA region for six months, the model was replicated for markets like UK, France, Brazil, Mexico and Australia.
- Software and Technology used: Python R, SQL, Time series algorithms and Ensemble modeling etc.
Increased stock visibility in a regular and volatile environment
Better forecasting accuracy
Improved inventory management for all products
Enhanced product lifecycle