Forecasting in the New Normal – Why Accuracy Matters
by Emil Capino, Founder and President of Info Alchemy Corporation Info Alchemy AWS Microsite Blog - August 2020 Amidst the impact of the coronavirus pandemic, the global economy is expected to bounce back from the crisis slowly but surely into unprecedented business conditions. Organizations are urgently looking to adapt new ways of maximizing limited operating hours, manpower resources, and evolving customer buying behavior in order to survive in the new normal.
Marchine Learning Use Case Frequency
Based on the global coronavirus heatmap infographics by Moody’s Investors Service http://moodys.com/coronavirus industries are affected by the pandemic in varying degrees. This shows that businesses need to react differently based on their exposure to the economic impact of the virus. Companies which belong to highly exposed industries must resort to drastic cost cutting measures and optimize their operations to stay afloat. On the other hand, businesses that are in low exposure industries need to take advantage of new business conditions by accelerating their adoption of digital technologies to address new trends in online shopping and consumption. In the new normal, organizations must open up to explore and adopt new technologies to optimize their costs and business processes. Now more than before, companies must leverage their data to generate new insights on their internal operations, to optimize and reduce costs while maintaining positive impact on customer experiences.
Machine Learning Adoption
In a survey by Algorithmia (http://info.algorithmia.com), on the ways companies are using machine learning, and published in a whitepaper entitled “The state of enterprise machine learning”, results show that majority of the respondents or about 38% leverage machine learning for cost reduction. Another 37% of respondents are using AI/ML technology to generate customer intelligence, while 34% of surveyed companies use machine learning to improve customer experience. The survey results clearly show that organizations are finding more benefits in adopting AI/ML as an integral part of their new normal operations. Let’s explore how businesses can benefit from machine learning to help reduce costs.
The business case for forecasting entails having an accurate projection into the future that is based on evidence of past results and other relevant information, and more importantly, determining the direct financial impact of the forecast to a business. With the inherent uncertainty in current business conditions, the impact of under or over forecasting has never been more critical. Forecasting accuracy in the new normal takes center stage as it can significantly affect the survival of organizations particularly in highly exposed industries. In a forecast error benchmark survey across supply chain and demand planning professionals from various industries, Forecasting Blog (http://forecastingblog.com) shares the average forecast error by industry:
The survey reveals that there is a huge discrepancy between forecast and actual values, most notably in the Consumer Packaged Goods (CPG) and Manufacturing-Industrial industries. The average forecast error is a clear indication of the huge opportunity for companies to improve their forecasts and reduce costs. For example, in Inventory Planning, the impact of over-forecasting results into excess inventory which leads to unnecessary costs. On the other hand, under-forecasting may lead to lost revenue from non-sale of out-of-stock items. By having an accurate forecast, these scenarios can be avoided which would translate to cost savings and increased sales. In Workforce Management, the accurate forecast in manpower resources planning can lead to either cost savings or additional expenditures: under-forecasting results in overtime costs; over-forecasting leads to under-utilized resources. Another example is in Capacity Planning, where most industries that produces or manufactures products struggle in generating accurate forecasts, under-forecasting brings uncapitalized infrastructure while over-forecasting turns into unmet demand and opportunity lost. Lastly, in Financial Planning, where forecasting plays a critical role in determining the additional financing a company may require, over-forecasting can lead to depleted reserves while under-forecasting may result into under-cutting due. The following diagram summarizes the business impact of forecast accuracy:
Recent advances in Artificial Intelligent and Machine Learning (AI/ML) are enabling the use of advanced neural networks in forecasting. Deep Learning, a part of a broader set of machine learning methods based on artificial intelligence, uses a set of algorithms that are modeled loosely on the human brain, called neural networks, that are designed to recognize patterns. Using neural networks, deep learning transforms and extracts features to establish relationships between an input stimuli and associated output responses. One example of advanced machine learning which are used in forecasting is DeepAR, a methodology for generating accurate probabilistic forecasts based on training an Auto Regressive Neural Network (RNNs) model on a large number of related time series data. With probabilistic forecasting, i.e. estimating the probability distribution of a time series future values based on historical data, DeepAR can be used to apply deep learning techniques to forecasting. In retail businesses, for example, forecasting demand is crucial for having the right inventory available at the right time and at the right place. With DeepAR, retailers can overcome many of the challenges faced when using traditional forecasting methods. Through extensive testing and empirical evaluation using real-world timeseries data sets, DeepAR can help retailers achieve more than 15% improvement in forecast accuracy compared to widely used statistical methodologies.
Amazon Forecast Service
Amazon Web Services (AWS), the leading cloud platform service provider has recently launched a cloud-based AI service that uses deep learning to generate highly accurate forecasts. Amazon Forecast, a fully managed time series data forecasting service on the AWS platform, uses deep learning and other forecasting algorithms, to make predictions or forecasts that are applicable in the areas of product demand, inventory, supply chain, workforce, capacity and financial planning. While the service