Proving the Business Case for the Internet of Things

PSE turns to AWS for machine learning

Steve Rogerson
September 3, 2019

Washington utility Puget Sound Energy (PSE) has become one of the first users of Amazon Forecast, a managed service from Amazon Web Services (AWS) that uses machine learning to deliver accurate forecasts based on the same technology that powers
The state’s largest utility supports 1.1 million electric customers and 825,000 natural gas customers in communities in ten Washington counties.
“At PSE, we’ve used Amazon Forecast to forecast electric and gas consumption at a typical residence,” said Paul Johnson, senior cloud architect at PSE. “We found that even with a very limited set of historical consumption and weather data, Amazon Forecast performed very well at forecasting 30 days out with virtually no manual effort. With the increased emphasis on environmentally-friendly energy, the ability to produce more accurate energy usage projections at each of our customers’ homes and businesses will be essential for energy service providers like PSE. With these enhanced analytical capabilities, PSE will be able to identify custom energy saving programmes and services, ultimately reducing customer bills.”
Amazon uses forecasting to make sure that the right product is in the right place at the right time by predicting demand for hundreds of millions of products every day. Amazon Forecast uses this technology to build precise forecasts for virtually any business condition, including product demand and sales, infrastructure requirements, energy needs, and staffing levels, with predictions that are said to be up to 50% more accurate than traditional methods.
Amazon Forecast requires no machine learning experience. The service automatically provisions the necessary infrastructure, processes data and builds custom, private machine learning models that are hosted on AWS and ready to make predictions.
Forecasting is the science of predicting the future. By examining historical trends, organisations can make a call on what might happen and when, and build that into their future plans for everything from product demand to inventory to staffing. Given the consequences of forecasting, accuracy matters. If a forecast is too high, customers will over-invest in products and staff, which ends up as wasted investment, and if the forecast is too low, they will under-invest, which leads to a shortfall in raw materials and inventory, creating a poor customer experience.
Today, companies try to use everything from simple spreadsheets to complex financial planning software to generate forecasts, but high accuracy remains elusive for two reasons. First, traditional forecasts struggle to incorporate very large volumes of historical data, missing out on important signals from the past that are lost in the noise. Secondly, traditional forecasts rarely incorporate related but independent data, which can offer important context, such as sales, holidays, locations, marketing promotions and so on. Without the full history and the broader context, most forecasts fail to predict the future accurately.
Amazon has a wealth of knowledge in building accurate forecasts using machine learning from over 20 years of experience operating its ecommerce business. Delivering billions of packages per year, with a multitude of delivery options in more than ten thousand postal codes, Amazon has developed forecasting capabilities that incorporate the full product history and overlay context from related business activities, such as promotions and pricing changes. Due to this diverse and large-scale forecasting experience at Amazon, businesses have asked AWS to share this knowledge with them to help make their own forecasts more accurate.
The general availability of Amazon Forecast provides a step towards putting the power of Amazon’s deep experience in forecasting into the hands of everyday developers in virtually every industry. Amazon Forecast produces private, custom models that can help developers make predictions that are up to 50% more accurate than traditional methods.
Using machine learning, Amazon Forecast automatically discovers how variables such as product features, seasonality and store locations affect each other. These complex relationships can be difficult to spot using traditional forecasting methods, but Amazon Forecast uses the machine learning developed at Amazon to recognise complex patterns to improve forecast accuracy.
Amazon Forecast automatically sets up a data pipeline, ingests data, trains a model, provides accuracy metrics and performs forecasts. Developers do not need to have any expertise in machine learning to start using Amazon Forecast, and can use the Amazon Forecast API or console to build custom machine learning models in less than five API calls or clicks. They can achieve accuracy levels that used to take months of engineering in as little as a few hours.
“Amazon Forecast now offers the forecasting expertise from Amazon’s first 25 years of building the world’s largest ecommerce business in a managed service for any company to leverage,” said Swami Sivasubramanian, vice president for Amazon Machine Learning. “We’ve built sophisticated, machine learning forecasting algorithms over many years that our customers can now use in Amazon Forecast without having to know anything about machine learning themselves. We can’t wait to see how our customers use the service to reduce operating expenses and inefficiencies, ensure higher resource and product availability, deliver products faster, and lower costs to delight their customers.”
Amazon Forecast is available in Ohio, north Virginia, Oregon, Tokyo, Singapore and Ireland with more availability zones coming soon.