I love robot movies even though they often pit machines vs. humans. The first Terminator was not good for humans (especially if your name was Sarah Conner). And I just recently saw the original Westworld for the first time, with Yul Brenner as the gun-slinging robot gone haywire in a western vacation village. That didn’t end well for the vacationers. No doubt, bad robots are compelling but scary.
But, there are good robots that help humankind, like the Autobots in Transformers and, well, I was glad to have Arnold on our side in Terminator 2 and 3.
In this blog, we will be focusing on good robots and Artificial Intelligence (AI), and the positive effects upon corporate credit, collections and receivables personnel and processes. The discussion will include several Artificial Intelligence (AI)/receivables subject matter experts from HighRadius Corporation: Sahil Hakim, Associate Vice President, Product Marketing; Jay Tchakarov, Vice President, Product Management and Marketing; and Elaine Nowak, Director of Product Marketing and Management.
Rob: Everyone likes their invoices to be paid, and to be paid on time, in full, with a seamless flow from credit and invoicing through cash application. While credit/collections/receivables professionals seek seamless circuitry (sticking with the robot metaphor), quite often things may spark and short circuit. What are the things that throw the credit-to-cash process off-course and add cost, labor and make things more unmanageable for the respective teams and customers?
Sahil: Starting with the credit process, the biggest challenge is that a chunk of credit decisions require a great deal of human intervention as analysts need to manually consume multiple pieces of information from credit agencies, public financials, open A/R, customer payment and credit history. The problem is this is not scalable, and adding more analysts is not an everlasting option. There is growing complexity in the data and credit scoring methodologies compared to 20 years ago, where an analyst could assess credit worthiness for each account on a first name basis. Now, an analyst might have 5,000 accounts, making it very difficult to sustain that knowledge and prioritize with deep insight on each account. One of the key questions this presents is ‘how could decision making be standardized,’ and ‘how could analysts be supported with rich insight for making better decisions.’
Elaine: For collections, the process of gathering remittance advice and payment information is very manual as well. The collections team has to keep tabs on payment status at an invoice level and also account for any outstanding payment commitments. The approach to prioritizing accounts is often simply based on invoice aging.
The deductions team also spends copious amounts of time downloading and reading through backup documents to determine the validity of a customer deduction. This slow process could lead to unnecessary holds on the credit limit and the blocking of future orders. And when a deduction is proven to be valid, all that work researching the deduction was a waste of time as no money comes back when the customer is proven right. What this means is that a lot of time is lost in the simple aggregation and collation of information across the A/R landscape, before an analyst is able to act on that information. Every day lost adds to DSO in one way or the other.
Rob: There’s a lot of buzz about how new technology such as how Artificial Intelligence and robotics could address manual and arbitrary processes you noted. Is AI really a solution to these problems?
Sahil: I believe there is definite potential here. Consider the cash application process. There is a promise of faster receipt and lower processing costs for electronic payments when compared to checks. But more electronic payments with decoupled remittance could easily increase manual work. Today, AI automates redundant processes and helps analysts do less work to post payments, irrespective of the remittance information format or payment type.
AI could also drive decision making, where machine learning takes in all the data sets such as open A/R and credit history, and automates low-value decisions or provides insight.
Rob: Understanding that AI is a solution, let’s level-set on some terms and definitions. There are a lot tech references that we’ve tossed around, and it would be good to get some basic understanding. So, for example, what is Artificial Intelligence/AI compared to machine learning, and robotics?
Elaine: It’s good to distinguish the terminology, but it’s difficult as there is no governing body or standardization process to define each. We should think of “AI” as a broader concept where technology is able to carry out a task with some smarts. “Machine learning” is the application of AI. It works by learning through accessing tremendous amounts of data and recognizing patterns to arrive at conclusions.
For example, for Accounts Receivable, the cash application solution learns how an analyst would handle exceptions for incoming remittances. After seeing a familiar pattern as carried out by the analyst, the machine would execute that same task when it next sees the same pattern.
Robotic process automaton is something that works best with well-defined processes and parameters that are static and repeatable.
Sahil: That’s right. Robotic Process Automation (RPA) gets from point A to point B after being programmed on how to do it. Cash application could be automated through RPA for reading 200 different remittance formats received every day. But what about the 201st remittance format? The machine won’t be able to handle something it has not encountered before, and needs additional programming. This is commonly observed with cash application automation systems based on OCR and reading templates which end up requiring excessive amounts of manual inputs for template management and training.
Rob: With that background on AI and how it works, what is a before and after AI implementation scenario that further illustrates the benefits of embracing the “good” robots?
Elaine: For cash application, the analysts have to gather remittance information from a lot of sources, and the information could be in many different formats. This payment detail then has to be manually matched to open A/R. With an AI-based solution, the remittance information is automatically captured from check scans, emails, attachments, customer websites and EDI files; there is no need for time-consuming template management, as remittance and payments are automatically linked to open invoices, and cash is directly posted to the ERP.
Sahil: Think about collections too and how pre-AI is very arbitrary. Collectors will look at a list of what invoices are due today, and what is past-due. But collectors are looking at thousands of customers with multiple invoices; how do they know which to prioritize? A McKinsey report stated that 70% of all collection actions are wasted because the customers would have paid anyway. This indicates a lot of wasted collection effort.
With AI, decisions could be automated and prioritized based on previous patterns. Payment history, along with other parameters, could be used to predict future payment dates. With that insight, collectors are able to prioritize much better. If the customer is going to pay in the next five days, collecting on that invoice does not have to be a priority. Machine learning guides decisions and improves outcomes.
Jay: AI also helps customer service and communications. Credit management, collections, billing and payment deductions all need to seamlessly flow the latest available information to the customer. Here’s an example: Suppose that a customer gets blocked due to a credit limit. The collections team is then reactive, and reaches out to the customer. Once payment is received the cash application reconciles the open A/R, which then updates the credit balance to release the order. All this information flow is now triggered manually, and the consequence is that the order release is delayed, time is lost, and the customer is frustrated.
Instead of being reactive, AI could predict the likelihood of an order being blocked based on history. With predictive insight, the credit team could then proactively trigger a series of automated actions to obtain a payment or a commitment from the customer and release credit. The insight from machine learning improves the process, changing it from reactive to predictive. The result is better decisions and customer service, while making the entire credit-to-cash process seamless.
Rob: Those are good examples of how AI improves cash application, collections and credit decisions. No doubt, AI, machine learning, and robotics offer lots of benefits. So, where are we on the adoption curve?
Sahil: How Accounts Receivable practitioners have taken to AI is an interesting story. Because AI in credit and A/R has been focused on reducing manual intervention or faster processing, it has helped the C-suite visualize actual impact and be more willing to bet on AI. This has driven adoption across company sizes, from the Fortune 1000 to small and medium businesses based on value. AI in credit and A/R is being delivered by the cloud, and it brings ROI democratically across all industries and sizes.
This interview was originally posted on The Elevation Blog, NACHA by Robert Unger