You are on page 1of 19

A SEMINAR REPORT

ON

“Future of Accounts Receivables in Age of


Artificial Intelligence”

in Partial Fulfillment of the Requirement for the Award of

BACHELOR’S DEGREE
IN
ELECTRONICS & ELECTRICAL ENGINEERING

SUBMITTED
BY

Pratik Ghosh (1603219)

SCHOOL OF ELECTRICAL ENGINEERING


KALINGA INSTITUTE OF INDUSTRIAL
TECHNOLOGYBHUBANESWAR, ODISHA - 751024

1 | P a g e School of Electrical Engineering, KIIT, BBSR

CERTIFICATE
This is to Certify that the SEMINAR REPORT entitled “Future of Accounts
Receivables in Age of Artificial Intelligence”, is a bonafideworkdone by Pratik Ghosh
(1603219), in partial fulfillment of the requirement for thedegreeof Bachelor of Technology in
Electrical Engineering at School of ElectricalEngineering, KIIT Deemed to be University,
Bhubaneswar. He has workedwiththeAccount Based Marketing team for HighRadius
Technologies PrivateLimited,Hyderabad during the Internship at the company and sincerely
completedtheassignedtasks.

Signature of Mentor

(Mr. Sarvesh Singh)


HighRadius Technologies

2 | P a g e School of Electrical Engineering, KIIT, BBSR


ABSTRACT
“Artificial Intelligence is the science of making machines do things that would require intelligenceif doneby men.” -
Marvin Minsky
Technology has been instrumental in electronically centralizing information and enabling
workflowcollaborationbetween users. However, when it comes to enabling key-decision making, the accounts
receivablesteamislefttodeal with mountains of data. This is now, a thing of the past.
AI or Artificial intelligence has long been a buzz word that has been thrown around. With Artificial
Intelligence,technology has now forayed into a space where it will drive resolution for low-value transactions,
andsimultaneously assist with decision making for the high-value ones. In the Age of fintech startups it'simperativefor
executives especially in the financial sectors to understand what AI can do for them
To begin with one should understand how AI is evolving and the silos in which it is evolving. Theobviousquestion
people have when talking about AI is whether it is some sort of black box that can take anydecisionashumans do .To
answer this, it is important for people to understand the most common use cases of AI indifferentsectors/industries.
This seminar report gives in-depth insights on the impact that AI has on Accounts Receivables; beit streamlining the
process, helping analysts make better decisions or ensuring productivity. The processesthathave been highlighted are:
• Credit Management
• Billing and Payments
• Cash Application
• Collections
• Deductions

Along with weighing the benefits of AI coupled with A/R, a sneak peak of what AI has in store for AccountsReceivable
is also elaborated.

3 | P a g e School of Electrical Engineering, KIIT, BBSR


Contents
1. INTRODUCTION ............................................................................................................................. ...............5 1.1. AI
in Everyday Life..............................................................................................................................................51.1.1.
Examples Of AI In Everyday Life.......................................................................................................................51.1.2. The
Goal Of AI - Improving The Quality Of Life.............................................................................................51.2. Clearing
the Air Around AI.................................................................................................................................61.3. What is RPA -
rules-based automation...............................................................................................................61.4. What is Machine
Learning - decision making....................................................................................................61.5. Combined
effect....................................................................................................................................................62. DEEP DIVING TO
EACH A/R PROCESS ......................................................................................................7
2.1. Credit Management..............................................................................................................................................72.2.
Cash Application.................................................................................................................................................112.3.
Deductions............................................................................................................................................................132.4.
Collections............................................................................................................................................................152.5.
Electronic Invoice Presentment and Payment..................................................................................................18
3. CONCLUSIONAND FUTURE SCOPE OF AIIN AR.................................................................................... 204.
REFERENCES ............................................................................................................................. .................21
4 | P a g e School of Electrical Engineering, KIIT, BBSR
1. INTRODUCTION

1.1. AI in Everyday Life


Artificial Intelligence (AI) generates a wide spectrum of emotions. On one end, some viewAI ascapableof improving the
quality of life, while on the other it gets viewed as a threat to human livelihoodandjobs.This article exemplifies how ‘real’ AI
is built with a focus to automate only manual and clerical tasksandto assist humans in making complex decisions, rather than
replacing them altogether. While Artificial Intelligence has recently been hyped; it has been deeply rooted in our
dailylivesfora long time now, here are some examples.

1.1.1. Examples Of AI In Everyday Life


Amazon - Targeted product suggestions using past search
Has it ever happened that after spending the last 30 minutes unsuccessfully searching for thoserunningshoes on
Amazon.com, you end up seeing ads for active wear and fitness bands? If yes, thenit’smorethan just coincidence. It’s
Artificial Intelligence.
The “Your Recommendations” link on Amazon.com directs users to products recommendedjust forthem. Amazon
recommends a range of products based on parameters such as categories theuserhasbeen browsing, items viewed,
demographic data and interests – with the aimof displayingproductsthat users are more likely to buy. 35 percent of what
consumers purchase on Amazon comefromproduct recommendations. As of 2018, Amazon.com is the world’s most valuable
brand. Source: “Amazon.com Recommendations, Item-to-Item Collaborative Filtering” by GregLinden, Brent Smith, and
Jeremy York, Amazon.com

Tesla – Self-driving cars


Imagine the year 2020. You enter your car at 7:30 am on a Monday morning. Avoice asks youtofasten your seat-belt and sets
off for your workplace along the fastest route. Don’t worry, you’renot being kidnapped. The voice is a digital assistant and
the car is an AI-enabled self-drivenTesla. What used to be science fiction is being brought to life by Elon Musk and his
engineers intheformof Tesla’s line of self-driving vehicles. The self-driving capacity of Tesla cars has a safetylevel
substantially greater than that of a human driver. And if you are not part of the Elon Muskfanclub,then there’s Google and
Apple who are working on their own autonomous car initiatives as well. Andif you are still against this entire concept, then
you could just wait till you’re not at the drivingwheelanymore – or till there’s no driving wheel.

1.1.2. The Goal Of AI - Improving The Quality Of Life


All business processes rely on human activity – primarily for abstract and complex decisionmaking,but unfortunately also
for a lot of repetitive, low-skill, clerical tasks – such as matchingtransactionsorclearing invoices. Nonetheless, with growing
business complexity, it is nearly impossible for analyststo account for all the various dimensions and patterns that influence
process outcomes, andhencedecision making is often based on individual judgment and experience.

Artificial Intelligence is already solving this problem for dozens of industries includinghealthcare, media and insurance. AI
could tap into company data and historical outcomes to either completely automate the decision or to support an analyst with
insight for decision makingtodrive better outcomes.

The ultimate goal of AI is to improve the quality of human life. Its purpose is not to replacehumans,but to focus humans on
more critical problems, while it takes care of the low-value stuff. Andyouwould agree, unless you enjoy driving during rush
hour to work, every day of your life. WithAI, analysts could focus on high-quality tasks which create bottom-line impact.

5 | P a g e School of Electrical Engineering, KIIT, BBSR


1.2. Clearing the Air Around AI
1.2.1 Too many words, too much confusion
With buzzwords such as AI, machine learning and RPA floating around, it’s difficult tograsptheirtrue
meaning and how they are distinct from one another. The following definitions aimtocleartheair around
these technologies and their capabilities.

1.2.1 What is Artificial Intelligence – the parent scientific discipline “Artificial Intelligence is the science of
making machines do things that would require intelligenceif done by men.” Marvin Minsky
Artificial Intelligence or AI is the ability of computer systems to learn, reason, think andperform tasks requiring
complex decision making.

AI is capable of:
• performing complex tasks that need judgment by analyzing humans • evolving with
experience and looking for better ways to execute • handling newer inputs based on
experience.

AI is used as a parent term. It is the net application of multiple processes, two of whichareRobot
Processing Automation and Machine Learning

1.3. What is RPA - rules-based automation


Robot Processing Automation or RPA is the use of software ‘robots’ mimicking human actions toperform a well-
defined business process. RPA is best-suited for performing tasks which are repetitivein nature and executed by
following a fixed set of rules.
However, RPA is based on strict compliance of orders and highly structured input – hence, eliminating any
scope of learning.

1.4. What is Machine Learning - decision making Machine Learning is an application of AI,
capable of identifying patterns froma large set of datawiththehelp of algorithms. It is self-learning in nature and
becomes ‘smarter’ over time. It could be deployed to predict future outcomes and identify trends.

1.5. Combined effect


When an AI-based system is powered by both Machine Learning and Robot Processing Automation, itiscapable of:
1. Repetitive task automation
2. Assisted decision making
3. Autonomous decision making

Examples of the above three are highlighted below:


Repetitive task automation: With the help of algorithms, AI could access different website portals, loginto their
accounts, download remittance information, irrespective of format and type, and captureit inthesystem. It could do
this accurately without any human intervention.

Assisted decision making: Prioritizing accounts for collections requires input fromvarious A/Rdepartments and data
sources such as Credit, Deductions and Cash Application. AI could pull indata
6 | P a g e School of Electrical Engineering, KIIT, BBSR
seamlessly from various sources at regular intervals and use it to dynamically prioritize accounts. Basedon the aging and
prioritization details, AI could also automate strategic dunning for accounts suchasreminder emails, past-due notices and
calls.

Autonomous decision making: Automating trivial activities provides analysts more time to focus onhigh-risk tasks.
Executing credit reviews for low-risk accounts could be delegated to AI. AnAI-basedsystem could consider all the factors
usually considered by a human analyst in estimating the credit limit.Such an autonomous system could be used for auto-
approving credit limits and coordinatingapprovalsfor low-priority accounts while analysts focus on credit decisions for high-
value, key accounts.

2. DEEP DIVING TO EACH A/R PROCESS 2.1. Credit Management


Artificial Intelligence can help to transform credit management from a reactive process to a proactiveprocess by enabling
dynamic account review prioritization based on internal factors as well as externaltriggers. Traditionally, teams define fixed-
time intervals for performing credit reviews across customersegments. This means that credit limits often remain stagnant
while multiple account level factorscontinueto change. AI takes in real-time information on credit agency ratings, account-
level A/Randagingdetailsand notifies analysts about accounts which require immediate review. This helps analysts focus
ontheessential customer accounts by removing the non-required periodic reviews and saving moneyspent ongetting latest
credit rating agencies reports.

2.1.1. The Current Process


In typical scenarios, customers are segmented into high-value and low-value brackets. Intheperiodicreview process, high
value customers are given priority over small customers. As a result, frequencyof periodic reviews for small customers is not
sufficient.
For periodic reviews, a fixed date is provided to the system and once that date arrives, a periodicreview workflow is
triggered. The credit review requires the analysts to aggregate credit information from various credit agencies and
accordingly assign a revised credit limit.
Due to a reactive process where the credit team acts on triggers instead of taking actions beforeevents,there are plenty of
loopholes which can exploit the business operations of the company. For e.g.

7 | P a g e School of Electrical Engineering, KIIT, BBSR


according to collections information, if a customer shifts from the low-risk bucket to high-riskbucketfive months
prior to the periodic review date, then the credit system will not be able to detect it.
The limitations of existing credit review process are:
• Not triggered by external events
• Expensive due to downloading credit reports of every customer
• Negligence towards small customers
• The company is prone to high risks

How Can Robotic Process Automation Support Periodic Credit Review

Robotic Process
Automation(RPA) is another one of the new fads of accounts receivables throughwhich A/R executives can
automate all repetitive manual tasks, save operating costs andre-allocateresources to other high value tasks.
In Credit Management, RPA can do the following:
1. Credit Agency Integration:
It can automatically integrate with different credit agencies like D&B, Experian andalsoseveralFinancial
Groups such Yahoo Financials etc. in order to download all credit and financereportsofall customers
automatically. Moreover, all the essential information fromthe reports will automatically be aggregated and
stored within the system. The A/R teamwould not havetospendresources or time downloading reports and
aggregating information.

2. Credit Scoring Engine


A robust credit scoring engine that can automatically calculate the credit score for a particularcustomer
based on pre-defined metrics and parameters as required by a company for calculatingthe credit score and
credit limit of the customer.

3. Structured Collaboration and Approval Workflow In order to have faster approval and decision making on the
credit score and credit limit, thestructured collaboration and approval workflow helps the analysts to escalate
any reviewtoconcerned manger along with all previous information including reports, payment history,
earlier credit review and calculated credit score at once and get the reviewpromptly.

Yet, does it address all challenges? No.

8 | P a g e School of Electrical Engineering, KIIT, BBSR


2.1.2. The Need for AI:
Despite the benefits of RPA, Periodic Credit Review still suffers as:
The small customer accounts are still left neglected
Analysts have to wait till the next periodic review date to download all reports andreviewanaccount
during which the account might already have gone bankrupt or unstable enoughtonotpay its outstanding
A/R
Under these conditions, the A/R team would be left blindsided with no premonition of anydisaster.This may lead
to increase in the number of blocked orders which will quite drastically impact thetop line with the drop in sales
and overall drop in revenue generated. Moreover, it will alsoincreasetheoutstanding receivable. Worse, if the
account goes bankrupt the receivable will become uncollectible.
Hence, it is essential to have any advanced technology that can foretell the risks and alarmtheteamofany event
that might indicate that the credit rating of an account is taking a dive.

Blocked Order Resolution


As we discuss how lethargic and inadequately supported periodic credit reviewmight leadtoincreasein the number
of blocked orders and uncollectible receivables which immediately impact bothbottomline as well as bottom-line,
it is essential to understand if the blocked order resolution processitselfisadequate and reliable?
Following is the common process followed by most of the global companies to resolve blockedorders

As we can clearly see that in this process, any blocked order gets released based on unreliableinformation or
commitment by the customer and collections team. Now, if the customer still doesnot pay, then not only the
earlier amount remains due, the value or amount of the releasedblockedorder also become outstanding
receivable.
Hence, this becomes a huge risk for the organization. Also, a lot of effort and time is wastedbythecredit, cash
application as well as collections team to release blocked orders which couldhavebeeninvested in other high-
value tasks.
But, what if AI could predict that a particular order might get blocked before it actuallygetsblocked?It would
save time and resources as well as eliminate the risk of increasing past-due A/R

9 | P a g e School of Electrical Engineering, KIIT, BBSR


2.1.3. Credit Management Powered by AI:

Proactive Periodic Credit review:

In AI supported credit review, analysts will be able to study customer payment historyandpatternstopredict which
customers might be downgraded in their credit rating. For high-value/impact accounts,Credit analysts would be
proactively prompted to work on these accounts. It will download and keep a check on both internal as well as
external factors 24*7 andtrigger aperiodic credit review in case of dipping credit ratings. It will also automate the
credit reviewforallsmall customers. Supported with both
RPA and AI that can automatically download, aggregate all credit reports information andcalculatethe credit score
and limit, all the accounts will be ready waiting just for the approval of theconcerned analysts or manager saving
a lot of time for the analysts.

Proactive Blocked Order Management

Based on the

previous payment behaviour, credit ratings and current chareacteristics, AI can predict if thecreditlimit is going to get
exceeded and the order might get blcoekd, it will automatically triger a workflowforcreditanalyst to check it’s credit
limit and inform the collections team who then initiate an automatic correspondneceto get the the payment from the
customer directly. With this, the credit limit becomes sufficient andtheordernever gets blocked.
10 | P a g e School of Electrical Engineering, KIIT, BBSR
2.2. Cash Application
2.2.1. Cash Application - Current Process
Drilling down into the individual cash application processes isolates the areas which are still largelydependent on
human resources and require immediate automation.

Remittance information aggregation

For checks
Check remittance data mostly arrives in the form of paper check stubs. The paper-basedinformationneeds to
be manually entered into a spreadsheet by the analysts to match it with payments. Keyinginremittance data is
a time-consuming and low-value activity that relies entirely on humanresources.

For electronic payments


E-payments and remittance information are received separately. Remittance comes in multipleformats (.txt, .csv,
BAI2, .pdf) and through multiple channels (email, web-portals, fax). Thedownsideis that unlike for checks, banks
provide no service for processing e-remittances. Analysts manuallyfetch, read, and match remittances to
payment.

Linking incoming remittance and payment


Payment-remittance linking is a challenge, particularly for e-payments as banks offer nosupport forelectronic
remittance processing.
A/R teams collect the decoupled remittances from various sources such as email, PDFs, Excel files,websites, and
EDI files and read the remittance data to link each remittance with the payment file. The expectations of
electronic payments in terms of automation are completely crushedunder theweight of the manual work required
to link the payment with the remittance.

11 | P a g e School of Electrical Engineering, KIIT, BBSR


Open A/R invoice mapping
Once the payment and remittance are linked, the next part of the process is matching the payment withthe open invoices.
Analysts perform this matching based on the details present in the remittanceinformation such as invoice numbers and
purchase order or shipment details. If these details andthepaid amounts match, the open A/R invoice is closed.
Short-payments
Buyers make short-payments due to eligible trade promotions, early payment discounts, or disputedservices or goods.
Deduction analysts go through the customer master, followed by customer documentation such as Proof of Delivery and Bill
of Lading and then do back-and-forthwiththesalesand customer service teams to validate the short-payment. If the discount
is eligible, the analystsneedto map the relevant reason code for that discount to the payment before applying it to the ERP.
Allofthe above processes consume a lot of time which drive analysts to write-off discounts withinaspecificthreshold without
proper research to avoid spending time rummaging through a pile of papers.

2.2.2. The Need for AI

Why cash application is important


The number one of priority of any receivables team is applying the payment received inthefirst place. Unless the customer
invoice is closed, all the efforts directed towards credit, collections, anddeductions are in vain.

The majority of the resources in cash application are burdened with low-value manual tasks suchasrekeying, reconciling, and
data fetching. Having an automated cash application process speedupthesetasks and could aid other A/R processes such as:

Collections: Having real-time visibility of the payments received allows dunning to be more systematic and strategic

Deductions: Automated dispute coding would help analysts focus on speedy deductionresolutioninstead of figuring out the
nature of the deduction.

Why cash application needs to be automated


It is well beyond the point where companies are able to debate whether or not to automate. Cashapplication is a low-hanging
fruit requires automation in order to stay up-to-date with the dynamicpayments landscape.

12 | P a g e School of Electrical Engineering, KIIT, BBSR


2.2.3 Cash Application powered by AI
More than 80% of the cash posting exceptions resolved by analysts are repetitive in nature. Thisisaresult of
customers repeatedly sending remittance information with the same inconsistenciesorerrors. An Artificial
Intelligence based solution could easily learn the pattern in which analystsmanually handle these exceptions and
auto-resolve such exceptions without requiring analyst inputfor a similar exception in the future. Deploying such a
self-learning systemover a cloudnetworkcould enable it to learn from the resolutions performed by thousands of
analysts – therebyvirtuallyeliminating the exceptions which require manual resolution over a period of time.
2.3. Deductions

2.3.1 Current Process: RPA supporting Deductions

Earlier, the deductions team would struggle and invest huge amount of resources and timeondownloading and
aggregating claim back up documents from different sources that come indfferentformats. Also, converting it all to
a single format and linking to the right deduction was a hugeheadache. Moreover, writing new denial letter for
each customer every time their deductionwasfound invalid was again a huge effort requirement.

Here is how RPA helped to cross the hurdles:

13 | P a g e School of Electrical Engineering, KIIT, BBSR


1. Automatic Back-Up Documents Aggregation Engine
RPA empowers your deduction management to automatically download all claimdocumentsfrom
customer’s email, websites or even paper claims and aggregate all important informationwithin a single
format which is then linked to the right deduction and collated withinthesystemin case of need during
further research.

2. Structured Collaboration Workflow


While researching, analysts might need to do a lot of to- and-fro betweeen internal teamslikesales team,
logistics team as well as external teams like customer or carriers ad shippers togetmore information or to
acquire any missing information. This collaboration engines easestheflowof information and documents (if
any) between internal or external teams. It helps tomaintainsacentralized reprository of all notes and
documents for more accurate and faster researchandalsocreate a document trail for audit purposes.

3. Automatic Correspondnece
Instead of wasting time and resources in writing redundant denial letters for custoemrs withinvaliddeductions,
automatic correspondnece functionality helps to save pre-designed and formatteddenial correspondnece letters
for all customers. The analysts can select froma givenpool oftemplates, attach docunents and just click ‘send’
to correspond with the customer.
2.3.2 The Need for AI

RPA runs quite a few miles in adrressing all challenges of Deduction Management yet is it suuficient?Here are
a few arguments:
1. Even if a deduction is valid, it still requires a set of manual, time-consuming tasks tobeexecuted before an
analyst is able to determine its validity. With more than half of all deductionsbeing valid, this means that
credit and A/R teams lose productivity that could have beenspentonresolving and collecting on invalid
deductions.

2. A high percentage of deductions that are written off are false positives i.e. they are invaliddeductions but are
incorrectly found valid and hence written off. Also, there are companieswhoprefer to automatically write off
any deduction below a particular threshold value say$25tosavetime and resources. In this case, any deduction
below $25 is automatically writen off without anyvalidation. Now, if there were just one or two such cases,
companies would not needdeductionsteam at all. But, once customers realise this threshold value, they start
exploiting andeachofthese$25 deduction for say 10,000 invoices lead to $250,000 loss!

3. Moreover, due to lack of bandwidth to persue invalid deductions as the teaminvests equal amountof time for
both valid and invalid deductions, the recovery rate for invalid deductionremainsquitelow.
With unneccessary investment of time and resouces that even cut down efforts requiredfor recovering invalid
deductions, deductions team using their current process would still face$250,000loss excluding the amount of
false positives deductions.
These arguments scream how direly deductions team as well A/R executives need AI for deductions
resolution.

14 | P a g e School of Electrical Engineering, KIIT, BBSR


2.3.3. Deduction Powered by AI

In order

to reduce the time-crunch on deduction resolution, Rivana predicts the validityof adeductionby analyzing
several factors like payment history, customer account etc. and then makingpredictions.With these predictions,
70% of the deductions worklist is automatically resolved.

For the remaining 30% deductions, Rivana provides recommendations. These recommendationsarebased
on certain confidence categories. For example, Rivana might predict for a certaindeductionthat it is 80%
Invalid. This means that there is an 80% probability of this particular deductionbeinginvalid. Similarly, it
can predict 65% Invalid or 50% Invalid. On the basis of these recommendationsprovided by Rivana,
deductions worklist can then be re-prioritized to work on the highest probableinvalid deductions first.

After the list is prioritized, HighRadius Claims and POD Automation(CPA) solution is appliedtoautomatically
collate documents and aggregate information from various sources like websites/ emails/paper documents. This is
further supported by HighRadius Deductions Cloud Solution that providesastructured workflow and collaboration
engine for any approvals or escalations. Hence, providingtheanalysts more time on the actual root cause analysis
of the deduction.

The results of using AI:


Eliminates time and work lost on valid deductions
Prioritize high-probability invalid deductions
Identifies and controls inaccurate write-offs

2.4. Collections

2.4.1 Current Process: Reactive Collections

A typical collections process is largely reactive and relies heavily on due dates as the pivot for all dunningactivity. It
starts only when an invoice is due or shifts to a larger aging bucket. The majorityof collectionsoperations, including
account prioritization, correspondence strategies, and customer collaboration, arebased on static parameters such as
aging bucket and invoice value. This results in a clutteredcollectionsworklist, inefficient identification of delinquent
accounts, and wasted collections efforts. Duetotheabsence of a scalable collections process which takes dynamic
parameters into account, thecollectionsteam ends up chasing only past due A/R while the overall team productivity is
lost in labor-intensive, time-consuming, low-value tasks such as ERP data extraction,manual

15 | P a g e School of Electrical Engineering, KIIT, BBSR


worklist creation and correspondence with non-critical customers. The key fallouts includeaslowercash
conversion cycle, increasing DSO, inefficient processes and higher operational costs. Amanualcollections
process is illustrated below:

In essence, an ideal collections management process rests on four pillars:

1. Data: Data is the backbone for all collections operations and includes all of theinformationextracted from
the ERP, the data obtained from other teams such as cash application, creditmanagement, deductions
management, billing, and A/P, and the information gatheredfromthecustomer such as payment
commitments and invoice discrepancies.
2. Customer Collaboration: Undoubtedly, customer collaboration is the heart of thecollectionsprocess.
But isolating the at-risk, critical customer from the fast-paying, low-riskcustomersisno mean feat
and has a significant impact on the outcome of the customer correspondenceandaccount coverage
for a collector.
3. In-Line Correspondence Logs: With an exponential increase in collections workload, itisnotfeasible for
collectors to keep track of all previous customer collaboration throughnotepadsorspreadsheets. It is
essential to log correspondence within the collections management tooltoreduce the pre-correspondence
workload for the collectors.
4. Account Prioritization: With tens of thousands of customers and a multitude of openinvoices,spending time
and efforts on critical collections accounts is of paramount importanceforeverycollector to ensure account
coverage and positive growth for collections closed. Inthisscenario,relying on static parameters such as due-date
for prioritization, as discussed before, is simplynotanoption for collectors. The collections teams across the
industries need to consider dynamic, leadingindicators as the bedrock for worklist prioritization for a scalable
collectionsprocess.

2.4.2. The Need for AI


An ideal, scalable solution provides an efficient, compliant and cost-effective process byoptimizingthefour
basic factors of an automated collections process. Typically, the automation for eachofthefourkeys pillars of
the collections process would provide:

1. Real-time Data: Most of the trending collections automation solutions provide real-timedataintegration to
the A/R teams, ensuring availability of data that is accurate and availablevirtuallyimmediately at their
fingertips.
2. Automated Correspondence: With the progressive drive towards e-adoptionanddigitalenablement of the
customers, automated correspondence is one of the most commonfeaturesoffered by top software
available in the market today. Automated dunning via email, fax, print

16 | P a g e School of Electrical Engineering, KIIT, BBSR


and mail with easy-to-create correspondence templates is an effortless recipe toimproveaccount coverage,
achievable through technology solutions currently available. 3. Log Correspondence in a Single System:
Most automation facilitates call notes,correspondence logs, and other means to document communication
with the customer withinthesolution itself. This eases the workload of collectors and ensures a single-point
accesstoallcommunication history with all customers.
4. Prioritization - Still Stagnant: Regardless of the challenges and roadblocks offeredbystaticprioritization,
most automations are still dependent on lagging factors, such as due-dateandinvoice value, to slice and
dice the invoice data extracted from the ERP. While thistechniquemay seem reasonable and insightful
enough from a top-level view, the industrial statisticsandleaders disagree.

Static prioritization is the core of a reactive collections process which ensures that dynamicchangesincustomer
behavior are obscured in worklist prioritization. Teams that have understoodtheimportanceof proactive
collections and dynamic parameters have tried leveraging Average Days Delinquent (ADD)as a predictor and
leading indicator but failed to produce noteworthy results.
Machine Learning could be leveraged to enhance collections process as it enables paymentdatepredictions that are
up to four times as accurate, using historical A/R data. It could identifyrelevantvariables and analyze valuable
patterns in the collections cycle to make an educatedguessonthepayment date for each customer; science that is
practically impossible for humans. MachineLearninghas the ability to process, analyze, and identify patterns
amidst the enormous volume of historicaldataavailable for each customer. It could predict the payment date at an
invoice level for all customersandhelp the collections teams become proactive through improved dunning
strategies.

2.4.3 Collections Management Powered By AI

The collections process encompasses numerous factors (features) which could be leveragedtoidentifythe payment
trends of each customer and make predictions based on these features. In machinelearning,regression analysis is
used to predict the exact (continuous) value rather than a class (rangeofvalues)prediction as with classification
algorithms. With respect to collections, regression analysispredictsanexact payment date while the classification
algorithms predict the payment delay in termsofbucketssuch as [0,5], i.e. delay between one to five days.

Since the collections process has a high number of predictive variables and a huge samplesize,therandom forest
model was an ideal fit. A Random Forest consists of an ensemble of an arbitrarynumberof simple tree predictors
which are used to determine the final outcome, each capable of producingaresponse when presented with a set of
predictor values. In a regression problem, their responsesareaveraged to obtain an estimate of the dependent
variable (output) given the predictors. It couldpredictthe exact payment date for each invoice with higher accuracy

as illustrated in the graphbelow:

17 | P a
g e School of Electrical Engineering, KIIT, BBSR

Proactive Collections Management not only helps analysts de-clutter their collectionsworklistsby
prioritizing the right accounts but also enables greater output with less cost byoptimizingcollections
efforts. Insights from Artificial Intelligence could be ingrained ingrassroot-leveloperations such as
customer segmentation and collections rules and strategies. Improving customer satisfaction, ensuring
better visibility, and optimizing resourceallocationare some of the key benefits offered by an AI-enabled
collections process. Thefollowingexplores some of the simulated benefits in detail.

• Dynamic Rules and Strategies: Predicted payment date and dynamic parameters enhancethecollector’s
focus on critical accounts. For instance, the following figure illustrates asamplerulein a proactive
collections management tool:

This rule is based on three factors: ▪Invoice value: Static parameter from open A/R
▪ Number of days for invoice to be due: Dynamic parameter calculated from open A/R ▪ Predicted delay: Proactive parameter
predicted by AI/ML algorithm

• Optimize Order and Credit Management Based on Customer Behavior: PredictedDelaycouldbecomputed at


the time of order creation based on the invoice parameters and customer history.Ifthe predicted delay is
high: ▪ Request upfront payment for accounts or particularinvoices
▪ Require payment commitments at the time of order creation
▪ Updating credit terms to proactively minimize delay in payment

• Proactive Correspondence to Reduce DSO and ADD: With proactive collections management,the collectors
do not need to wait for invoices to be past-due and can take proactiveactionsbasedon predicted delay.
Taking proactive measures reduces past-due A/R and ensures visibilityintothe performance of individual
collectors as well as collections strategies.

2.5. Electronic Invoice Presentment and Payment

2.5.1. The Current Process

Current EIPP process involves customers coming to the portal and making paymentsorusetheportal for
presentment. The process is ad-hoc or manual and needs manual interventioneitherbythe merchant or
customer. E-adoption efforts are misguided and not targeted. Automated,intelligent electronic campaigns
could significantly increase adoption without drivingupcosts.

18 | P a
g e School of Electrical

Engineering, KIIT, BBSR

2.5.2. The Need


for AI

EIPP+
radiusOne
Network
combination can help increase adoption of electronicinvoicesandelectronic correspondence. Network
intelligence to increase adoption of electronic payments.It'svery likely that a customer who is buying would
be a customer for another supplier too. TheradiusOne network will use this intelligence to create targeted e-
adoption campaigns toswitchthemtoelectronic invoices and electronic correspondence. Currently companies
provide a flat discounttoall their buyers to incentivize them to switch to e-invoicing.

By using the intelligence provided by the radiusOne network, the Machine learningsystemwillbeable
to:
i. Create targeted campaigns to increase e-adoption.
ii. Track electronic usage of your customer to switch a customer frompaper/fax invoicingtoEmail invoice &
electronic payments.

2.5.3. EIPP Powered By AI

How it would work: Our portal sends electronic-invoices through mail to customers whoarealreadyreceiving
paper /fax invoices. We use our technology to track mail opens/invoice downloadsoveraperiod of time for those
customers. If our email tracking system tags themas a frequent user ofe invoices they are most likely a candidate
to make the switch to e-invoicing and also to electronicpayments. This information is used by of Machine learning
algorithm to create targetedmail campaigns (for e.g.: a mail sent a day before the due date of an invoice explaining
the benefitsof

19 |

P a g e School of Electrical Engineering, KIIT, BBSR


paying using electronic means as they would be able to get the benefits (discounts as per payment terms) of
paying on time vis-à-vis checks which would reach the recipient late.

Reduce the risk: The use of Deep Learning technologies in Payments


Risk decisions in payments are never 100% clear, therefore they are the perfect playingfieldforDeep
Learning technologies.

How are risk decisions made?


As a merchant or acquirer, you do not want to accept a payment where there is a high probabilityforafraudulent
action or a high likelihood for a chargeback. On the other hand, rejecting a “good”paymentnormally results in
conversion loss and thus, lost revenue for the merchant.
To decide on acceptance or decline of the payment, many different factors are available tobetaken into account,
for example:
• Time & date – e.g., is a credit card used at normal shopping times? • Merchant category codes – is it used
at a supermarket or for online games? • IP-addresses – where is the device used for online shopping
located? Social information – the use of social media data is quite a new application in payment
riskdetection.

Neural net learning consists of multiple hidden layers and mimics the behavior of the humanbrain.Deep
learning included multiple neural networks put one after the other.
By training a neural network with historic risk data, the network will learn and improve andthereforebe able to
better predict risk for future payments. However, it is important to point out that themostimportant driver for
deep learning quality is the grade and amount of training data – not theactualalgorithm!

Pay with your voice or your face: Upcoming applications of Deep Learning and Artificial Intelligence
The payment industry has a long history of working with data to make qualified risk decisions. Forsure, risk
analysis is the “bread and butter” application of Deep Learning technology withinapayment company.

In the upcoming years we expect many new applications, driven by AI technology that will significantly
influence the industry, both in payment and commerce in general. Just tolist a few:
1. Chatbot assisted payments/ correspondence
2. Natural voice processing
3. Smart data for offers and advertising
4. Facial Recognition identification or biometric identification

3. CONCLUSION AND FUTURE SCOPE OF AI INAR


We are living in an age where technology is skyrocketing in terms of progress. Stagnancy is athingofthepast and
unless organizations jump on the band wagon of AI automation, they won’t reap the far-reachingbenefits that
technology has to offer. The following technological advancements are in the cardsalongwiththe growing adoption of
AI across industries.

Natural Language Processing + Smart Assistants


If you own an iPhone, chances are high that you use Siri to set your morning alarm, add an dinner plantoyourcalendar or
just to know how’s the weather is going to be tomorrow. During the introduction of computers, executing a task meant
writing hundreds of lines of cryptic code. Then came GUI which allowed

20 | P a

g e School of Electrical Engineering, KIIT, BBSR

exploring features in an intuitive manner, via mouse or touch. With the recent onset of Smart Assistantscoupled with
Natural Language Processing, the system is able to understand human voice (alongwithitsunique accent, intonation,
pitch), perform the task and/or respond in a manner that is indistinguishablefrom an actual human being. NLP in A/R
will make the environment from user-friendly for theanalysts.Instead of going through millions of data points, the
analysts can obtain information directlybyinteractingwith the smart assistant inbuilt within the solution. Moreover,
being an AI-based solution, NLPwill self learn and improve with time.

Speech-to- text Synthesis


There have been numerous occasions where we wished there was a technology to capture notes duringameeting. With
AI, it is no longer a fantasy. Speech-to- text synthesis is able to capture humanvoiceandconvert it to digital text in
real-time. Speech synthesis is based on Machine Learning usingsamplesofdifferent types of voices, languages,
accents. Prior to using the technology, there is a voice calibrationsetupwhere the system gets familiar with the user
voices. Once the system is set-up, it can take notes with90%+accuracy. The use cases for this technology can be for
collectors taking call notes and payment commitments while calling customers.

Omnichannel
Nowadays people own multiple smart devices such as smartphones, laptops, tablets, TV. Internet ofTechnology or
IoT coupled with AI allows these devices to communicate with each other. Thereasonforthis is actions performed by
one device should be reflected in all other devices. Omnichannel for Accounts Receivables allows users to access
information in all their devices across processes of theOrder-to- cash cycle in real time such that they can take
corrective information instantly. Omnichannelgives managerial visibility of process- owners through easy-to-
consume reports and statistics, out-of-boxas well as customized. With the recent advent of augmented reality and
virtual reality headsets, it won’tbelong before the life of an A/R analyst would resemble being Tony Stark.
4. REFERENCES
[1]. https://www.highradius.com/integrated-receivables-solutions/receivablesradius/credit-solutions/ [2].
https://www.highradius.com/integrated-receivables-solutions/receivablesradius/cash- application-automation/
[3]. https://www.highradius.com/integrated-receivables-solutions/receivablesradius/deductions-management/[4].
https://www.highradius.com/integrated-receivables-solutions/receivablesradius/collection-solutions/[5].
https://www.highradius.com/integrated-receivables-solutions/receivablesradius/eipp-ondemand/ [6].
https://www.highradius.com/integrated-receivables-solutions/receivablesradius/radiusone-network/[7].
https://www.highradius.com/highradius-rivana-artificial-intelligence-for-credit-and-ar/

21 | P a

g e School of Electronics Engineering, KIIT, BBSR

You might also like