RPA and AI Notes

Robotic Process Automation (RPA) & Artificial Intelligence (AI)

The Future of Work

We are entering the Automation Era, which favors cognitive, social, emotional, and technologically oriented work. RPA is an opportunity to introduce software into a business, allowing organizations to reflect on internal processes and create competitive advantages. Automate repetitive tasks to free people for value-added activities and constantly assess processes for automation potential using a bottom-up approach with low-code/no-code solutions, involving citizen developers, and evolving through APA, RPA, Intelligent Automation, and AI.

The Shifting Landscape of Finance

Digitalisation is transforming the finance landscape.

Session Overview

• Robotic Process Automation (RPA)

  • Definition and functionality of RPA

  • Types of RPA Robots

  • Impact of RPA in Finance & Accounting (F&A)

  • Challenges of RPA adoption

  • Automation potential in F&A

  • Digital Transformation through RPA

  • SIPOC Framework

  • Organization Level Transformation

  • Lean and Six Sigma methodologies

• Artificial Intelligence

  • Definition of AI

  • AI Pillars

  • Value of AI

• Machine Learning

  • How it works

  • Types of Machine Learning

• Natural Language Processing

  • Definition of NLP

  • NLP in F&A

• Follow Up Information (Day 2 & StudioX)

AI Overview

Definition of AI, Machine Learning (ML), Supervised Learning, Unsupervised Learning, Reinforced Learning, Deep Learning, Application of ML, Natural Language Processing (NLP), Context of NLP, Application of NLP.

The Collective Intelligent Accountant, Automation Tools, The Future of Work, The Collective Intelligent Accountancy.

What is Robotic Process Automation?

RPA is a technology that enables computer software to emulate and integrate actions typically performed by humans interacting with digital systems.

RPA Technology Characteristics

• Swivel chair, Presentation Layer

• High Volume / Low Complexity

• Stable Processes and Stable Underlying Applications

• Processes performed by Large Teams

• Mature

RPA Robots can execute processes that are highly manual, repetitive, rule-based, with low exception rates, and standard electronic readable input. RPA Robots are able to:

• Capture data

• Run applications

• Trigger responses

• Communicate with other systems

Technology Maturity S-Curve for Insurers (Source: Accenture).

How does RPA work?

• Platform: An evolving system of interdependent pieces (modules) that can be innovated upon, comprising Architecture and Modules.

• Two fundamental technology management phenomena:

  • Increasing interdependency of products and services

  • Increasing ability to innovate by more actors

• RPA Mimics Human Action

• Quick to Implement and powerful to Scale

• Operates any Application 24/7

• No Mistakes / Non-Stop

• Structured reading and actioning Data

(Gawer and Cusumano, 2002)

What Processes Fit Best?

• Manual and Repetitive Processes: High transaction volume, high frequency processes running daily & weekly.

• Processes with Standard Highly Readable electronic Input type: Excel, Word, email, xml, PPT, readable PDFs.

• Rule Based Processes: Clear processing decision making based on standardised and predictive rules.

• Low Exception Rate: Activities with a low number of variation scenarios.

Not all processes are fit for automation. To get the best benefit of a rapid ROI, choose processes which first have passed through a transformation initiative using Lean Six Sigma methodology aiming to standardise, control and streamline first.

What Processes Fit Best? (Continued)

• High Volumes: Processes with high transaction volumes (and high frequency).

• Mature and Stable Processes: Well documented, stable, & predictable with well-known operational costs.

• Changeable Processing Method or System Change: Processes for which the processing method cannot be changed due to various reasons and do not require a fundamental change in the underlying technical architecture of the current systems

• Automation Savings – in FTEs: Automate only processes which can provide a savings in terms of human work-effort of a minimum of 2 FTEs.

Not all processes are fit for automation. To get the best benefit of a rapid ROI, choose processes which first have passed through a transformation initiative using Lean Six Sigma methodology aiming to standardise, control and streamline first.

Types of Robots in RPA

• Attended Robots: Work side by side with human colleagues, executing some tasks of a process, requiring human intervention for business activities where human input is required.

  • Characteristics:

    • Resides on an employee's workstation

    • Triggered only when needed

    • Under direct control and supervision

    • Responsive

    • User-friendly

    • Flexibility

• Unattended Robots: Operate without human touch, deployed by orchestrator, runs efficiently in batch mode, enables remote access. Work independently of any human interaction.

  • Characteristics:

    • Manual, Repetitive

    • Highly rule-based back office activities

    • Which do not require any human intervention

    • Flexible to deploy on virtual or remote environments

    • Easy to scale

    • Accurate

Why RPA?

RPA is Non-Invasive, Easy to Scale, Easy to Start With, and Future Proof compared to other approaches in digital transformation.

How is RPA impacting the work landscape?

Some companies are already going for a “digital first” strategy, and RPA can accelerate their implementation.

• Share of task hours between Humans and Robots:

  • 2018 (calculated): Robot Hours 29%, Human Hours 71%

  • 2022 (projected): Robot Hours 42%, Human Hours 58%

World Economic Forum report titled "The Future of Jobs Report 2020“

UiPath Studio

Workflow Designer: Visio-like process mapping and automation framework; simplifies training and accelerates process modeling & automation.

Automation Recorder

Specialized workflow recorders for desktop apps, web apps, Citrix and terminal emulators. Unrivaled computer vision perceives the screen like a human does. Debugging tool for analysis of automated processes: visual, step-by-step process execution, breakpoints and highlighting of target elements.

Automation Library

Hundreds of drag-and-drop actions. Pre-built process enables automation to integrate with cognitive and OCR technologies from ABBYY, IBM Watson, Google, Microsoft text analysis and APIs.

RPA Across Industries

Services, Financial, Healthcare, Education, High Tech, Manufacturing, Retail, Government

• 93% of organizations say that automation kickstarts digital transformation.

• 89% of organizations are adopting a digital-first strategy.

(Source: 2018 IDG Digital Business Survey, 2019 Economist Intelligence Unit Survey)

Personal Applications of RPA

• A robot that does the check-in for you 24 hours prior to any flight.

• A robot that checks all your emails and sends you the important ones as a digest after your vacation.

The Benefits of RPA

• Increased Turnaround Time & Increased Productivity

• Automated Knowledge Management & Reduced Compliance Risk

• Non-intrusive Integration with Enterprise Systems & Real-time Unified View

• Increased Revenue & Reduced Operational Cost

• Reduced Change Management Timelines & Contextual Information

• Increased Employee Engagement & Enhance User Experience

• Guided Support through Structured Collaboration & Improved Customer Satisfaction

• Unlock Massive Productivity Gains in an Enterprise

Challenges of Digital Transformation in F&A

Legacy Systems, Unstructured Formats, Natural Language Interactions and Fragmentation, Paper-based Documents with Unstructured Formats.

RPA Track Record in Finance & Accounting

Order to Cash Process Flowchart ("As-Is")

Manual Validate Sales Order, NO/YES conditions, Outlook and SAP interactions.

Order to Cash Process Flowchart ("To-Be")

Automated Input Structure Recognized? with NO/YES conditions, SAP interactions, OCR application to extract data, manual processing if needed.

Clarifications

• Unknown Exceptions %: Percentage of the total volume received which cannot be processed without an external factor (query/approval)

• Process Expected to Change: Are processes or applications used to process a case going to change within 3 - 6 months?

• Standard Input: Inputs are standard if the content is positioned in the same place even if the input types are different.

• Inputs are NOT Standard: Inputs are considered as non-standard when the position of the content varies from one input type to another

Complexity Factors

• Number of Screens: The higher the number of screens, the more elements must be captured and configured prior to the process automation.

• Types of Applications: Some applications are more easily automated (such as the Office suite, or Java), others heavily increase the automation effort (Mainframe, for example).

• Business Logic Scenarios: An automation's complexity increases with the number of decision points in the business logic.

• Types and Numbers of Inputs: Standard input is desirable, Non-standard input can be of different complexity grades, with free text being the most complex.

Categories of Processes

• No Robot Process Automation

• Semi Automation

• High-Cost Robot Process Automation

• Zero Touch Automation

Automation Potential in Accountancy

Accounts Payable, Accounts Receivable, General Accounting, Tax, Treasury & Compliance, Financial Planning, Analysis and Reporting.

The SIPOC Framework

Following the SIPOC methodology, you could now open conversations with upstream and downstream process owners as well as have a better understanding of the current process itself, potentially increasing the current deal size.

Why are these questions important?

• You get to know the name(s) of the process owner(s) who directly impact your current process. If their process is causing issues to your current process, now you can pitch automation to these upstream processes.

• Inputs could be through emails, paper, PDF, fax etc. Once you understand the input, you can recommend an appropriate automation solution

• Once you understand what is sent downstream and in what format, you could start thinking of automation opportunities for the current as well as the downstream processes

• Once you understand the customer of a process, you could ideate as how can the current process generate more value for them and how can automation enable that value generation?

Example of a SIPOC – Vendor Set Up Process

The COPIS Framework

(Note: The transcript mentions COPIS but repeats the SIPOC definition. Assuming this is an error and COPIS wasn't detailed.)

The Organisation Level Transformation

Four key components: PROCESS, POLICY, PEOPLE, TECHNOLOGY

Tools for an Organisation Level Transformation

• LEAN: An approach that emphasizes the smooth flow of items synchronized to demand to Identify Waste.

  • Understand 7 types of Waste

  • 5 S – Sort, Set in Order, Shine, Standardise & Sustain

• SIX SIGMA: A disciplined Methodology of improving every product, process and transaction by Minimising Variation.

  • DMAIC: Define, Measure, Analyze, Improve & Control

Transformation is a Continuous Journey and not a Destination. Sustain from Lean Principles and Control from Six Sigma.

Lean

Lean focuses on identifying those wastes in a function or even different process, the entire business. The fundamental principle is to identify and remove them using 5S

• Seven Types of Service Waste

  • Over Processing

  • Transportation

  • Motion

  • Inventory

  • Waiting

  • Defects

  • Over Production

• 5S to Remove It Focus on 7 types of waste (“Muda” in Japanese) and 5S to remove them.

  • Sort: Sort & remove unnecessary items

  • Set in order: establish an order of what needs to be where

  • Shine: establish a culture to keep the workplace pleasing

  • Standardise: eliminate variation

  • Sustain: It is about self discipline to sustain all that has been achieved in steps 1 to 4

Six-Sigma

• Define: What is the business problem?

• Measure: With data establish how big a problem?

• Analyse: Understand what is causing those issues?

• Improve: Take corrective actions and measure again.

• Control: Establish Governance.

DMAIC provides this structure.

Benefit Realisation of RPA

When assessing automation opportunities, companies should go beyond the traditional ROI

• Cost savings: RPA ensures cost savings through FTE reduction.

• Productivity Gain: Increase in processed volumes within the defined unit time, coupled with a decrease in turnaround time.

• Business Agility: Enabling businesses to act at a faster pace than before.

• Quality Improvements / Error reduction: Robots run as configured with a 0% error rate.

• Compliance: Ability to comply to regulatory requirements.

• Customer Satisfaction: Automation leading to customer satisfaction.

• Flexibility: If there is an unexpected spike in volume, robots enable you to scale up or down as required.

What is Artificial Intelligence? – Definitions

• "Artificial intelligence (AI), the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings" (Encyclopedia Britannica)

• "The field of computer science dedicated to solving cognitive problems commonly associated with human intelligence, such as learning, problem solving, and pattern recognition" (Amazon)

• "A branch of computer science dealing with the simulation of intelligent behavior in computers. The capability of a machine to imitate intelligent human behavior" (Merriam-Webster)

• "The theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages" (Oxford Dictionary)

What is Artificial Intelligence?

Artificial Intelligence in Action – Large Language Models (LLM) / Generative AI.

Generative AI in Action

PWC rolling out a Microsoft Enterprise Version (OpenAI) called ChatPWC.

• Large Language Models (Generative AI) within the Firms environment.

• Trained on internal PWC historical data.

• Currently deployed in Administrative and Research tasks

The Pillars of Artificial Intelligence

• Machine Learning (ML)

• Natural Language Processing (NLP)

Where are we currently?

Machine Learning in Action - StarCraft II AlphaStar

How does Machine Learning work?

• Machines try to learn through analysis

• Machine Learning makes predictions based on past learning

• Machines test their learning on new data introduced

The 4 Types of Machine Learning

• Supervised Learning: Data is labelled to tell the machine exactly what it should look for.

• Unsupervised Learning: The machine learning algorithm attempts to learn by itself from very little guidance and no labelling.

• Reinforcement Learning: The machine learning gains knowledge by trial and error to achieve a clear objective.

• Deep Learning: Attempts to mimic the patterns of the neural pathways that a human brain uses to understand something.

Machine Learning in Action

AXA has used machine learning for insurance underwriting, namely, to predict which car drivers are most likely to be involved in ‘large loss’ accidents. It predicted “large-loss” traffic accidents with 78% accuracy.

Practice – The FAE Perspective

• Cost of project: $2 \text{ billion }$

• AXA customers: $67 \text{ million}$

• Accident rate (>$10,000 payout): $1\% \text{ per annum}$

• Average premium for accident-prone customers: $\$700$

• Non-renewal rate after premium increase: $75\%$

• Cost of capital: $7\%$

• Project appraisal period: $3 \text{ years}$

• Tax rate: $10\%$

Your director has asked you to consider the facts and to prepare an analysis that she can use to persuade the board to adapt its pay-out strategy by adopting some technology to help predict who potentially could be involved and load premiums accordingly to offset some of this risk.

Natural Language Processing

NLP breaks down human text and voice data by performing the following:

• Speech Recognition: Reliably converting voice data into text data. Required for any application that follows voice commands. Challenging - People use speed, context, tone, accents, incorrect grammar, etc.

• Speech Tagging: Grammatical tagging - particular word or piece of text based on its use and context.

• Disambiguation: Selection of the meaning through a process of semantic analysis in the given context.

• Named Entity Recognition: Identifies words or phrases as useful entities.

• Co-reference Resolution: Identifying if and when two words refer to the same entity.

• Sentiment Analysis: Attempts to extract subjective Qualities - attitudes, emotions, sarcasm, confusion, suspicion.

Where are we currently?

• Natural Language Generation is the task of putting structured information into human language… (can be described as the opposite of speech recognition or speech-to-text).

• Optical Character Recognition (OCR) allows the processing of documentation.

Where are we currently

Large Language Models (LLM) – Current Deployment: BroadApplications across Finance andAccounting.

• J.P. Morgan – OpenAI LLM detecting trading signal from documents and speeches created by the Federal Reserve.

• Bloomberg – OpenAI LLM trained on financial data which performs research tasks.

• Technology is pushing boundaries – OpenAI powered, solution for globalAML, compliance and fraud prevention (FlagrightAI).

NLP in Finance & Accounting

Accountancy Firms performs thousands of Audits and Advisory contracts every year. The Firms must review and validate thousands of documents ranging from lease contracts to legal agreements to invoices and statements. For example, a vehicle lease agreement might contain ten pages of text, a property lease may contain up to 200 pages. A lease review will take on average 1 – 5 hours to complete depending on the profile of the reviewer, experience profile etc. Using Natural Language Processing a document could be scanned with up to 90% accuracy in a few moments. The program will query or flag content that it is unsure of and direct human attention there.

Where are we currently?

Amazon Textract is a machine learning (ML) service that extracts printed text and other data from documents as well as tables and forms.

• Introduction to Artificial Intelligence Day 2

• ACME System 1