We can no longer watch as almost every technology vendor announces that they are “using AI” or liberally uses the terms intelligence, AI or machine thinking in their product description. Since 1991, our company has been marketing and supporting a full-featured AI-driven collection platform. There are many years and tens or thousands of hours of design, programming, testing and implementation time that went into that effort. We have worked through many major revisions of the database and the software, as the collection industry and technology have evolved. Artificial intelligence is not a programming change you make in a few months. Is requires a different thinking, a radically different architecture, and can not be be created by a few months of work on an aging software platform. That is pure hype. It can be argued that the collection industry has to play “catch-up” in when it comes to modern technology. We can win but must do it quickly with actions, not talk. As we face the a new era in collections, it is important that we are honest about where we are and how we can get to our destination. As technology experts, we have a sacred obligation to be be honest and accurate with our motives and what we sell. It is impossible for a company that did not have AI anywhere in its sights a year ago, to suddenly add “AI” to its products. It is deceptive and irresponsible. Concerned that many collection technology users do not understand what AI means, this paper tries to introduce the reader to AI, and what we could expect from machine thinking in the collection industry. . . .

By Ranjan Dharmaraja, CEO, Quantrax Corporation


This is is an African grey parrot named Griffin. A new study shows the African grey can perform some cognitive tasks at levels beyond that of 5-year-old humans. The tests were conducted by researchers in Harvard’s Psychology Department and were able to show that “Griffin was working at the level of a 5-year-old, on a task at which even apes would not likely succeed.” Most of us would refer to that as some form of intelligence.

Humans have “natural intelligence”, as opposed to “artificial intelligence”. What is Artificial Intelligence (AI)? Wikipedia says it is a term used to describe machines (or computers) that mimic “cognitive” functions that humans associate with the human mind, such as “learning” and “problem solving”. In collections, computers have taken control of everything we do, but it would be grossly misleading to say that producing a client remittance statement via print, a web site or e-mail, or displaying a client or balance-based script is AI. When does traditional computation, automation or analytics become artificial intelligence? To answer that, let us look at the different types of AI. In June 2019, Forbes Magazine (whose article we have borrowed from) said there are two ways in which AI is generally classified. One type is based on classifying AI and AI-enabled machines based on their likeness to the human mind, while the other is based their ability to “think” and perhaps even “feel” like humans. According to this system of classification, there are four types of AI or AI-based systems: reactive machines, limited memory machines, theory of mind, and self-aware AI.

This has led to the defining following types of AI systems and machines.

  • Reactive machines – The simplest and oldest for of AI machines, automatically responding to a limited number of circumstances. IBM’s Deep Blue, the machine that beat chess Grandmaster Garry Kasparov in 1997 is an example.
  • Limited memory – In addition to the capabilities of reactive machines, these system use and learn from past experience and historical data to make decisions. Most of the applications we use today, fall into this classification. Fingerprint and face-recognition software, a dynamic scoring module in collections or being able to look at any account and make an expert decision on next steps are excellent examples.
  • Theory or mind – This is “the next level of AI systems that researchers are currently engaged in innovating. A theory of mind level AI will be able to better understand the entities it is interacting with by discerning their needs, emotions, beliefs, and thought processes.” Selecting the accounts we present to agents based on their personalities or moods would be an interesting example of “Theory of mind” in collections
  • Self aware – This does not currently exist and is the dream of AI. These machines will think like the human brain. These machines will “also have emotions, needs, beliefs, and potentially desires of its own. ” In collections, an example would be a machine that decides to move a batch of collectable accounts to Mary to temporarily help her increase her commissions because she recently lost her husband in an accident.
  • Artificial Narrow Intelligence (ANI) – All existing AI falls under this classification. These machines only do what they are programmed to do. According to the system of classification, these systems correspond to all the reactive and limited memory AI that exists today.
  • Artificial General Intelligence (AGI) – This refers to systems that have the ability to learn, perceive, understand, and function completely like a human being. They can “replicate a human being’s multi-functional capabilities.”
  • Artificial Superintelligence (ASI) – These machines would be smarter and better than humans because of overwhelmingly greater memory, faster data processing, analysis, and decision-making capabilities. They would also threaten our existence or at the very least, our way of life.

Associating an application with AI can be a moving target because prior benchmarks can become outdated with the progress of technology. Optical character recognition (OCR) uses machine learning and computer vision to recognize text. Once considered AI, OCR is today considered a standard function of a computer and as a result, may not be considered by many to be AI-driven. You will also find examples of “weak AI” that handles a single task or simpler problems, while “strong AI” takes on jobs that are more complex and human-like.



If we go back to Griffin the African grey parrot, it would be misleading to suggest that we could teach it to identify the accounts that were most likely to pay. As we start to call our products “intelligent” and advertise machine learning, we must put that in perspective. How will AI help us? How much of what AI can really do are we utilizing? Let us look at some examples.

  • Payment portals are a great example of “machines at work”. A consumer calls a phone number at 11 PM on a Sunday and is greeted by a human voice (text to speech). You are asked to enter your account number and last 4 of your social or month and year of birth. The machine accesses the account information `and authenticates the consumer. It states their balance and asks how much they can pay. You enter the amount on the key pad. The machine repeats the amount and if it is correct, it checks that it does not exceed the balance. We can go on to ask the consumer to key in a credit card number, an expiration date and CVV. We have been doing this for ever, and many years ago, this information would have been recorded and printed the next day for an admin person to process the payment. Today, we can immediately authorize the payment and give the consumer a confirmation number. We can also offer the a consumer a choice of monthly payment plans and at the end of the call, ask for a cell number and text them a receipt. Is this AI? We deserve much more than that and should “No”. We have been doing that for years and that solution was simply a matter of integrating traditional computer and telephony technology. If you did not look at it in that way, you would probably agree with someone who said they had an “intelligent payment portal”. What would make a portal more intelligent? What if we could offer a settlement option on that phone call? What if the machine could ask the consumer for their best settlement offer and could then negotiate a settlement in the same manner that an agent would? In order to do that, the machine would need to “have a conversation” with the consumer! That means understanding natural language. What if the machine could do that, making multiple offers, each time asking the consumer if they would accept? That would be close to a consumer’s interaction with a human agent, and much easier to qualify as “intelligent behavior”.
  • Our smartphones are millions of times more powerful than Apollo 11’s guidance computers of the 1960’s. Today’s computers can analyze big data in record time, offering feedback and predictions that were not economically feasible several years ago. We must be careful to understand the difference between AI and the results of great computational power.
  • In this digital age, digital communication channels are clearly better options than traditional mail and phone calls. If a system offered these communication channels, is that AI? Of course not! Just as the predictive dialer was not described as AI, we should not call text messaging or e-mails AI. If you found a way to select the best communication channel, based on consumer behavior and historical data, would that be machine learning and AI? Is Griffin the parrot intelligent? Does the fact that someone jut figured this out make it intelligent, when we have been doing it for 25 years?

Have we made our point? The messaging from technology vendors is changing very quickly. Other than for Quantrax, no other company talked about AI until very recently. We strategically stayed under the radar with our “invention” (we actually applied for initial copyrights in 1991). You may be aware of companies that heavily advertise machine thinking and AI. Some of these companies are well-funded by venture capital and seed investments. We encourage and welcome honest and healthy competition, but misrepresenting the potential or use of AI, and claiming to be the “first to use machine thinking” is deceptive when there is a collection software company that has enjoyed major success with AI for over 25 years. The next few months will see more companies advertise products that use AI. If you have a smaller product, for example, skip tracing, scoring, a chatbot or an IVR, this is not as complex as calling a collection platform “intelligent”. Take the example of the IVR. We showed you how that machine could work with basic data processing or be AI-driven. The question then is, how do we set a standard for what we should call an AI-powered application or feature? That is an extremely difficult question to answer, and we will try to respond by sharing some of the things we have done, that we consider intelligent.

If anyone is confused about the basics, the picture on the right illustrates a simple “AI” environment for collection software. In traditional “data- based” models, programs and data are closely related. To change the behavior of the system, you need programming changes. With an artificially intelligent system, you have your users and a “knowledge base”. User actions, different events and circumstances will access your data and look up the knowledge base, which you can compare to the brain of a collection expert. A powerful computer program called an “inference engine” will then think, make decisions and produce intelligent behavior. All this takes place in real time. True AI platforms can be trained to think and perform at the level of a human expert such as an experienced collection manager. The efficiencies and results cannot be matched with conventional systems.



Building an AI-driven collection platform is very different from building an intelligent payment portal, though they can both be advertised as AI. The difference is the depth and characteristics of the AI that exist within the underlying products. In 1991, we deployed our first version of RMEx (It was called Intelec), a complete collection system, built ground-up on an AI platform.

The system was significantly different from traditional data-based systems. These are some of the areas that we consider “intelligent”.

  • The software could be trained to make the same management-level decisions as an experienced collection manager. We did this by having a “knowledge base” that could store all of a collection manager’s logic, thinking and decision-making processes. We built a collection platform that talked to that knowledge base in real-time, at every decision-point in the collection process (E.g. when accounts were loaded, mail was returned, payments were posted, there was an NSF, a new address was obtained from skip-tracing, the consumer disputed an account etc.). The system would look up complex thinking logic and make the same decision that a manager would. You were able to set up this powerful thinking with no programming or technical knowledge. This analysis and decision-making could be initiated when an account was worked or by looking at all the active accounts during the day or at night. The latter offers machine thinking that could ensure that due diligence requirements were met.
  • Intelligent scripting was included. For inexperienced agents, a standard script based on the client type or balance was not very smart or flexible. Having the system read all the linked accounts for a consumer and create a script on the fly, was much more exciting and closer to human thinking. And why not allow scripting at the key decision points involved in working an account? For example when the consumer said they had an attorney or insurance, or when a payment arrangement was discussed? And why not translate that script into Spanish at the click of an icon?
  • A mid-sized agency could have 300,000 active accounts being worked at any time, with only 15 agents. How do you make sure that you present the right accounts to the right agents at the right time, with the best phone number that should be attempted? How do you identify and target only the accounts you need to work the next day and classify and sort them so they are worked efficiently to produce the best results? This is a job for powerful computers and intelligent software.
  • What about attempting multiple phone numbers to get to an RPC as quickly as possible? The fastest path to an RPC is to make sure each number is attempted evenly as far as the number of attempts. Each attempt must be made at a different time in the day (morning, afternoon and evening). Easy to accomplish? No, it is extremely difficult and why most companies can not do something which is so basic even after 30 years of automation! We did it. Did it make a difference? During the month of placement, one of our clients saw their largest client go from 0.10% to 4.70% in recoveries, topping 10.82% in month 2.
  • At our 2017 annual user conference, we made an ambitious promise to build a collection robot that would perform at the level of a mid-level agent within 12 months. Today, that robot talks to consumers in natural language and negotiates payment arrangements, settlements, “I can’t pay”, disputes, attorneys, bankruptcies, deceased, insurance and wrong parties. Regardless of how or when the consumer contacted Alex (our chatbot), everything is documented and updated as an agent would have done. Alex will even do real-time speech analytics. Today, Alex can communicate in 10 languages and can be contacted through the internet or a phone call. On the internet, a consumer can “talk” to Alex, use the text interface or use a more traditional web form.

There is a test for AI. It is called the Turing Test. The Turing test, developed by Alan Turing in 1950, is a test of a machine’s ability to exhibit intelligent behavior. equivalent to, or indistinguishable from, that of a human. The logic is simple. Put a machine and a human expert in different rooms, and ask them the same questions. If the machine gives you the same answers as the human, it passes the test and has “artificial intelligence”. Can the products that claim to have AI pass that test? Before every technology vendor announces and “AI product”, it is important to understand and set sone minimum qualification standards for marketing this important certification.


Quantrax has been in business for 30 years. Most of that time was spent designing, building and deploying AI-based software for the collection industry. As we watch the industry evolve, we are genuinely concerned about misinformation in this exciting new era of collections. Those of us who have been in the industry for a while know that some things have not changed. You need to find new clients, some who will expect you to work all their accounts in ways that they ask you to, not just a small percentage that you predict will pay. In addition to trying to contact consumers, you have to send mail, handle disputes, credit report accounts, close accounts and report to your clients based on their many requests. Core functionality like account posting, linking, letters, payment posting, productivity reports and month-end processing are critical and can take many years to build and perfect. It may explain why the software platforms launched in the last 15 years have failed or had very limited success in terms of acceptance and market share.

This is a time for unity and honest consulting. The industry depends on its technology vendors to consult with and provide the best advice. At the risk of being controversial, we believe that it is important for us to speak up and share our experience and our concerns in a critical period for this industry.

  • We were recently critical of the technology companies that did not offer the visionary thinking that the industry needed.
  • We have faced compliance challenges for many years, without receiving proven systemic solutions from the technology industry. Why?
  • We were given plenty of warnings, but are not ready for effectively and compliantly using today’s modern digital communication channels.

AI offers us the much-needed motivation and technology partner we need to transform negative perceptions and boost collection results. It requires honest consulting, not hype. All AI implementations are not the same. There are basic and more sophisticated applications of AI. There are relatively simple features that could be classified as machine thinking, and more complex and powerful end-to-end uses of AI. Those who have no proven track record of AI will say that “We can do this and that with AI”. After many years of having extremely powerful computational power available to us, why are some companies only talking about AI today? The collection industry has many things working against its, and this is the time to put aside individual ambitions and work together in the interests of our industry. Technology vendors are trusted influencers and we must pledge to understand that responsibility and offer the most objective advice at all times, as we look forward to the great times ahead of us. . . .

NOTES – Some information was quoted from a June 2019 Forbes article “7 Types Of Artificial Intelligence” by Naveen Joshi, a columnist and Founder and CEO of Allerin, which develops engineering and technology solutions focused on optimal customer experiences.


Quantrax Corporation is a technology company that created an intelligent collection platform over 25 years ago. They believe that the ARM industry has been poorly served by collection technology that has not evolved or kept up with the great potential of computing power, or challenging industry changes. Self-funded, Quantrax has continued to successfully develop and deploy technology that offers modern solutions to old problems. – (301) 657-2084

Leave a comment