See how BluePi can help you leverage your complex data in entirely new ways.
The need for forecasting is probably as old as human history is. Ancient maritime navigation would be based on the knowledge of seasonality and weather forecasts. While we have advanced as a race, the need for forecasting has remained steady if not increased but the techniques have vastly evolved owing to our knowledge of mathematics.
In the field of retail, sales forecasting is helping business achieve a clear competitive advantage. In particular in consumer oriented markets like fashion and electronics market demand is often uncertain and product life cycles are short thereby making forecasting more challenging.
The impact sales forecasting on success and performance of companies is crucial. Inaccurate forecasts can lead to stock-outs or over stock and lead to spiralling losses for the business.
Specifically longer time-to-market means that production plans have to be submitted much before future demand is known. Factors as varied as weather, holidays, economic situation, public events all have an impact on demand. Moreover in case of fashion the SKUs are usually replaced end of the season thereby making for a very dynamic product portfolio.
The primary use cases in sales forecasting can be summed up in three different areas
Armed with sales forecasting data businesses can better plan when and what products to order. The next seasons purchases can be determined by the volume of sales of the current year. Buying too little or too much of inventory can be a disaster. Imagine the impact on planning if the forecast could anticipate the economic conditions and the changes in buying patterns of consumers.
Which items sell better in which stores can help in better distribution logistics.
The knowledge of which products are successful and which have not really taken off can help in the reordering decision. To mitigate stock out and over stock scenarios goods can be moved from one store to another. This is a complex process and needs to take in to take into account logistics costs and time of travel to ensure the business benefits accrue as expected.
The production planning can fare much better if sales forecasts are taken into consideration.
Heads up on revenue growth in coming months can help business conserve cash for the unfavorable times.
Based on sales forecast for specific items at different stores can help push inventory that is lying idle.
Forecasting can be utilised for better budgeting decisions by the purchasing department.
Traditional forecasting methods have been a mix of approaches that have utilised expert judgements and time series forecasting models like Moving Average smoothing of time series data. These methods are prone to provide to erroneous forecasts in conditions when the data is noisy or stock out data is missing for example.
Another challenge is that most of the forecasting tools are at a gross volume for each store instead of individual SKUs per store. This grain of result is too coarse to garner any business value. This is done primarily because the number of SKUs and Store leads to an extremely large dataset. Moreover there is great variance between the sales trends of these different SKUs.
Hence while there has been no dearth of appreciation of the business value an accurate forecast can bring to the business the tools and techniques have not yielded desirable results.
With the advent of artificial intelligence the use of artificial neural networks (ANNs) have been progressively deployed for achieving greater accuracy.
Specifically for Fashion retail forecasting a new set of algorithms called Evolutionary learning machines that provide better performance and faster learning compared to gradient-based learning algorithms.
Further studies have applied Evolutionary neural networks and have demonstrated very promising results especially in case of noisy data.
A new model was promoted to deal with idiosyncrasies of the apparel industry and lack of historical sales data that applies soft computing methods like fuzzy inference systems and neural networks.
However these techniques are only used to baseline by most of the experts in the apparel industry due to either inadequacies or large confidence intervals.
Apart from the above consideration BluePi has significant variance in the forecasting needs and approaches depending upon the industry and its peculiarities. For instance the solution deployed at two different multi-format retail stores tend to be different depending on the clientele. This is where a custom tailor made solution for each business come to the rescue.
BluePi has built several forecasting models that consist of the following approaches
Time Series Forecasting with external variables
AI driven ANN models using the LSTM architectures
ELM models specific to fashion industry
Fuzzy inference systems to determine historical stock-out scenarios if data is not ready available
XGBOOST - a variant of gradient boosting trees
In our experience a single model rarely suffices and the accuracy thus provided is not sufficient. Thus we deploy an ensemble model that uses a combination of one or more of the above techniques.
Our customised solution also extend beyond the forecasts to help you reap benefits from the above three use cases of ordering, distribution and stock replenishment. These require additional modeling and integration with the existing systems.
Our services extend end to end and encompass the following areas to help you meet your business goalsREQUEST A DEMO