At the ACA Annual Convention & Expo in Nashville this year, attendees and presenters were frequently heard discussing how artificial intelligence (AI) would apply to the ARM industry. Even though AI concepts have been around since just after World War II, the technology has historically been locked into the halls of learning or well-funded labs, and only in the last decade has it burst into practical application. The emergence of widespread enabling technology for massive amounts of data and the ability to manipulate that data in reasonable time frames fuels the practical use of AI to solve real world problems. Since the turn of the century, we have generated and then absorbed massive amounts of data through cheap, ubiquitous sensors (think “the Internet of Things”). We have also been able to store that data online, cheaply. The presence and continuous arrival of data invites us to act on it. For example, massive amounts of video data give rise to advances in facial recognition. Similarly, audio data provides the foundation for understanding conversations. Two particular concepts exploding today are machine learning and a related concept called deep learning, both considered subdomains of AI. While the applications have developed recently, their original concepts have existed since the 1950s, and advanced further in the 80s with the development of neural networks.
A hallmark of machine learning is the massive amounts of data curated when it is exercised. Managing this data requires significant responsibility to safeguard the consumer’s privacy and other rights, whether their data is simply stored or used to make decisions, automated or not. Any initiative to harness data for machine learning must include methods for identifying and protecting sensitive data within large datasets, both structured and unstructured. Our industry is already aware and conforming with the requirements of HIPAA and PCI, but it’s important to remember that when business decisions are made that affect consumers, meaningful information about those decisions must be provided, care must be taken to treat protected classes equally (ECOA), and EU citizens must be provided with a way to opt out and information into the decision process. The power and scale of AI’s capabilities should not be taken lightly as these factors are brought under consideration.
Machine learning, especially deep learning, runs well on hardware using GPU chips for massive parallel instruction processing and extremely fast memory. As a result, GPU chip demand has skyrocketed and the major cloud services suppliers have purchased a large percentage of available GPUs. And while AI platforms can be installed onsite, businesses who build this within their own data centers are generally creating AI technology in house. There’s good news though: Companies provide solutions that deliver intelligence or intelligent functions commercially. These companies draw the data they use to train and execute machine learning models into their data centers or to their virtual private cloud (VPC) in a commercial or FEDRAMP cloud platform. There are a few vendors who reluctantly allow the customers’ data to stay on-premise for an additional charge if the customer purchases their own hardware. Because latency matters with today’s expectations of the technology, AI at the Edge is the next frontier, focusing on processing AI functions on edge devices: smartphones, wearables, cars, traffic lights, factory machines. In other words, AI will be embedded in the Internet of Things (IoT).
Selecting Tools for the Purpose
Artificial intelligence replaces or augments human presence in business processes, allowing us to pull redundant and tedious tasks out of human hands. Machine learning and deep learning applications make those tasks done by the system intelligent, learning with more experience and data, without rules written by humans. Here’s where we depart from earlier automation of business processes in receivables. For our industry’s entire history, we have relied on those long-tenured or deeply-experienced managers to translate their smarts into dialer and correspondence strategies, what to say on the phone, how to respond to denials, and other processes. Machine learning promises to create good outcomes by reviewing all of what has gone before and charting the path to successful actions. Replacing a human with a computer requires the computer to be fast enough to make the process work. Some use case results can wait a day, or can be allowed to take minutes or seconds, while others must feel instantaneous.
Artificial intelligence is an exciting opportunity for our industry to improve results while keeping a sharp eye on margins. With all the hype, it’s important we keep our heads, understand the context and uses for artificial intelligence in its various forms, and continue to monitor the regulations that will govern behavior influenced by automation. As with several of our delivered solutions, Ontario Systems will seek good technology and partners to provide what they do best, embedded in our best-in-class workflow and integrated contact management. Each time, we will examine the business case to ensure the solution delivers value and enables you to take advantage of opportunities or solve problems in your real world.
For more information from Amy on this topic, listen to her ARM Perspectives podcast, What is AI? Artificial Intelligence in the Collections Industry.
Disclaimer: Ontario Systems is a technology company and provides this blog article solely for general informational and marketing purposes. You should not rely on the content of this material for any other purpose or as specific guidance for your company. Ontario Systems’ advice, services, tools and products described herein do not guarantee compliance with any law or industry standard. You are ultimately responsible for your own company’s actions and compliance efforts. Because everyone’s situation is different, you must consult your own attorneys, accountants, and/or other advisors to obtain specific advice on your company’s compliance, legal, tax, regulatory and/or other business needs. Despite Ontario Systems’ efforts to provide current and up-to-date information, you need to recognize that the information contained herein may become outdated quickly and may contain errors and/or other inaccuracies.
© 2018 Ontario Systems, LLC. All rights reserved. Information contained in this document is subject to change. Reproduction of this publication is not permitted without the express permission of Ontario Systems, LLC.