November 28, 2020

Automating business processes with machine learning in the COVID-19 pandemic

5 min read
COVID-19 has changed our world significantly. All of this change has been almost instantaneous, forcing...

COVID-19 has changed our world significantly. All of this change has been almost instantaneous, forcing companies to pivot quickly and find new ways to operate. Automation is playing an increasingly important role to help companies adjust. The ability to automate business processes with machine learning (ML) is unlocking new efficiencies and allowing companies to move faster where they might have otherwise been stuck using antiquated systems. What might have previously taken an organization years is now happening in weeks. In this post, we discuss how AWS customers are applying ML in areas such as document processing and forecasting to quickly respond to the challenges at hand.

Automating document processing

The ability to automate document processing remotely has proven essential as companies face new challenges in this pandemic. Demand for services like loan processing and grocery delivery has spiked in areas that no one could have predicted and the ability to quickly respond to those demands remains vital.

In April 2020, the US federal government announced the Paycheck Protection Program (PPP) to provide small businesses with funds to cover up to 8 weeks of payroll, mortgage, rent, and utility expenses. With phenomenal demand and over $349 billion allocated in just the first 13 days of the program, small business owners were scrambling to qualify.

BlueVine, a fintech company that provides small business banking, used their technology and engineering expertise to help process billions in loans. They chose Amazon Textract, a fully managed ML service that automatically extracts text and data from documents, to help automate the loan application process. In just a few days, they were up and running, analyzing tens of thousands of pages with high accuracy. In just 4 months, they were able to serve more than 155,000 small businesses with over $4.5 billion in loans. They delivered services to those who needed it most, with 68% of loans going to customers with fewer than 10 employees and 90% of loans under $60,000—serving small businesses struggling to remain afloat. BlueVine worked closely with DoorDash as their strategic partner to serve many stressed small independent restaurants, and simplify and accelerate the loan process. BlueVine used ML to automate loan application processing and scale quickly to meet the unprecedented demand. The company estimates they helped save 470,000 jobs as a result of their efforts.

Other areas of the economy were also experiencing unprecedented demand and needed to staff up quickly. However, it was a challenge to process new hire employment paperwork at the rate required. A typical PDF form has about 50 form fields; to recreate it as a digital form, the customer had to drag and drop data to the right location on each form—a particularly time-consuming task. Enter HelloSign, a Dropbox company that automates the signature process.  HelloWorks is a HelloSign product that turns PDFs into mobile friendly forms. It uses Amazon Textract to automate document processing and save customers hundreds of hours. A popular on-demand grocery delivery service was able to onboard millions of shoppers using HelloWorks in a few weeks. HelloWorks helped the company scale their onboarding paperwork quickly by automating document processing with ML. An NY-based urgent care started to use HelloSign to register new patients. An ambulance service started using HelloWorks to send out COVID-19 test applications. What’s more, this was all happening online. As organizations continue to limit in-person interactions, demand surged for HelloWorks with users creating 3x more forms than they used to. With Textract, HelloWorks was able to automate the process and automatically create all of the fields and components, saving customers time and keeping them safe.

Forecasting the pandemic

Forecasting is a growing challenge as supply chains and demand have been disrupted on a global scale. Amazon Forecast, a fully managed service that uses ML to deliver highly accurate forecasts, is helping customers deliver forecasts for everything from product demand to financial performance. Many forecasting tools only look at a historical series of data to predict the future, with the assumption being that the future is determined by the past. When faced with irregular trends this approach falters – as demonstrated by the challenges faced by companies to develop models that accurately capture the complexities of the real world since the beginning of the COVID-19 pandemic. With Amazon Forecast, you can integrate irregular trends and a multitude of other variables—on top of your historical series of data—to deliver the most accurate forecast possible with no ML experience required.

One of the largest challenges when it comes to forecasting has been understanding the projection of the disease itself. How quickly will it spread? When will it spike next? How many hospital beds will be needed to accommodate that spike? Forecasting models have the potential to assess disease trends and the course of the COVID-19 pandemic. However, the nature of the COVID-19 time-series makes forecasting extremely challenging, given the variations we’ve observed in disease spread across multiple communities and populations. COVID-19 remains a relatively unknown disease with no historic data to predict trends, such as seasonality and vulnerable sections of the population.

To better understand and forecast the disease, Rackspace Technology, University of California Irvine, Scientific Systems, and Plan4Co have come together to introduce a new COVID-19 forecasting model to deliver greater accuracy using Amazon Forecast. The team of medical, academic, data science modeling, and forecasting experts worked together to use Amazon Forecast DeepAR+ to incorporate related time-series to build more powerful forecasting models. Their model used deep learning to learn patterns between related time-series, like mobility data, and the target time-series. As a result, the model outperformed other approaches, such as those provided by the well-known IHME model.

With Amazon Forecast, the team was able to preprocess the time-series, train dozens of models quickly, compare model performance, and quantify the best forecasts. These forecasts can be developed on a daily and weekly basis, now available for countries, states, counties, and zip-codes. This information can, for example, help forecast what new cases will be in the short-term and long-term by learning from real-world data, like time to peak and rate of transmission. This information is critical because government agencies frequently use the occurrence of new cases in a population over a specified period of time to help determine when it’s safe to re-open sectors of the economy.

Conclusion

As the pandemic continues, new challenges will inevitably arise. When time was of the essence, these organizations looked to ML technology and automation to serve their customers’ needs and find new ways to operate. This use of new technology will not only help them respond to the pandemic today, but also set them up to thrive in the future.

To learn about other ways AWS is working toward solutions in the COVID-19 pandemic check out Introducing the COVID-19 Simulator and Machine Learning Toolkit for Predicting COVID-19 Spread and Intelligently connect to customers using machine learning in the COVID-19 pandemic.

 


About the Author

Taha A. Kass-Hout, MD, MS, is director of machine learning and chief medical officer at Amazon Web Services (AWS). Taha received his medical training at Beth Israel Deaconess Medical Center, Harvard Medical School, and during his time there, was part of the BOAT clinical trial. He holds a doctor of medicine and master’s of science (bioinformatics) from the University of Texas Health Science Center at Houston.

Source link

Copyright © All rights reserved. | Newsphere by AF themes.