A Retail Tale - Data is talking about the future of retail demand. Are you listening?
It’s tough - forecasting retail demand accurately for inventory and supply chain managers who count on it. They need to be able to predict demand, days, weeks, and at times, months in advance to be able to sync stock fill rates with changing demand, make sure sales promos don’t result in product cannibalization, and unexpected demand spikes don’t increase inventory costs. Several retailers have started to look deeper into data from their demand history to see if it can give them a clue into how the future might look.
It is easy to look at historic data and see the patterns that point to the reality of present demand. But it is significantly more difficult to arrive at a model that uses this same data to correctly identify the patterns that will also repeat in the future. The real problem is to find a way for retailers to automate demand forecast accurately based on both the history of demand and the evolving realities of the market, channel and sales ecosystem.
Solving the Problem
Design, test and rollout – for resellers, brands and retailers - an automated demand forecasting model (as part of a suite of pre-integrated analytical solutions) based on diverse parameters such as channel, category, and SKUs along with data from the macro-ecosystem. It entails:
- Harvesting holistic demand intelligence
Demand is shaped by both market sentiment and business-driven variability caused by marketing promotions. Forecasting methods can gauge demand with a higher degree of accuracy when all variability factors are integrated.
Extract historical sales data (preferably from at least 2-3 years) to analyze causal relationships between sales and variables - including season, week of the month, day of the week, time of day, holidays, festivities and end of season promotions. Analyze demand at the master SKU as well as the aggregated periodic levels to identify the most suitable statistical method and time horizon for forecasting demand trends with minimal forecast error. Exploratory Data Analysis (EDA) presents a great way to break down datasets and construct this predictive model.
- Transforming insights into forecasts
Adopt multiple time series models to ensure all relevant parameters, including complex permutations such as channel and category, are considered at every level of forecasting. Create plots in R and Tableau to distill insights from EDA and enhance the accuracy of modeling. Use Auto.Arima, STLF and ETS functions in R to generate point forecasts, actual forecasts, and the mean of predicted values. Use the TBATS time series decomposition method to measure diverse components and decipher patterns. Select the most appropriate time series technique to forecast multi-dimensional data. Include Ensemble machine learning algorithms to ingest new data points and improve the credibility of forecasting.
- Building informed business plans
Build a plug-and-play demand modeling tool to deliver business intelligence for operations planning. The dashboard of current sales and future demand, enables merchandise planners to view channel-wise traffic, along with the spikes and troughs in demand, anytime-anywhere.
The Outcomes
- Predict sales by channel, product category, and time period with more than 85% accuracy
- Gain better visibility into short-term order volumes; which then streamlines budgeting, sales planning and warehousing operations
- Spot bestsellers easily; this helps drive revenue growth, improve merchandise and seasonal planning, as well as streamline fulfillment and improve in-stock situations to potential revenue growth of 1 to 2%
