The term artificial intelligence was first coined by John McCarthy, in 1956, during an interdisciplinary workshop of researchers at Dartmouth. This team of researchers developed the concept of “thinking machines”, which included automata theory, complex information processing and cybernetics.
In the mid-to late 50s, General Electric became the first company to purchase a computer to process its payroll system, the UNIVersal Automatic Computer(UNIVAC). UNIVAC ran payroll in all of GE’s factories and stored data on magnetic tape instead of punch cards. The UNIVAC took 40 hours to complete the entire payroll process. During the 60s the U.S. transportation industry developed an electronic data interchanges to standardize transactions between vendors and customers. This was followed by the introduction of the first computer-based software, LANguage for Programming Arrays at Random (LANPAR), began to offer electronic spreadsheets. After LANPAR, VisiCalac and SuperCalc were released in 1970 and 1978 respectively. With each iteration, more functionality was incorporated into the programs.
At the start of 70s, it become apparent that the Dartmouth project would require far more personnel and funding to achieve any degree of success. As these challenges surfaced, the US Department of Defense took an interest into AI and began training computers to emulate basic human reasoning. In 1973, in response to criticism from Congress, funding to the Dartmouth project and other undirected AI programs was halted. Congress redirected all funding into the creation of the Defense Advanced Research Projects Agency (DARPA). DARPA’s first successful AI project resulted in the creation of a street mapping program. The second project developed by DARPA provided the infrastructure for the integration of automated accounting and spreadsheet software. In 1983, as these advances developed in the public and private sectors, Intuit was founded and released its flagship accounting software Quicken.
In the late 80s, the Japanese government joined the US in its endeavors by commissioning several new AI projects and investing billions of dollars into developing hardware and software capable of applying AI technologies. By the 1990s and early 2000s, with these investments and the rise in popularity of the Internet, cloud hosted accounting software was introduced. The advent of cloud hosting accounting systems created a new challenge in combining mobility with cross device usability. This challenge was dealt with through the creation of enterprise resource planning (ERP) and customer relationship management (CRM) systems. The ERP system integrated the management of business processes in real-time, mediated by software. The CRM systems introduced a novel approach to manage the interaction between the company’s current and potential customers. Data analytic tools were introduced with the CRM system which reviewed customer history in the context of the company’s performance in order to improve business relationships with customers. Optical character recognition and intelligent data capture was also introduced in the early 2000s, which enabled the automation of the accounts payable process. For the first time, key bits of data such as the purchase order number, amount or data was digitally instead of manually captured and populated in a variety of index fields. Pre-set templates were produced to enable information to be automatically entered into the system. Businesses were also able to customize what type of automation they required to perform routine company operations. In 2003, DARPA produced the first intelligent personal assistant. These personal assistants led to the eventual development of Siri, Alexa and Cortana. Personal assistant technology can even be found in current accounting software packages.
Today, most sciences classify Artificial Intelligence (AI) as a subfield of computer science, analogous to how machines can imitate human activities. AI can be broadly defined as the simulation of human intelligence processes by machines, especially computer systems. These processes include learning (the acquisition of information and rules for using the information), reasoning (using the rules to reach approximate or definite conclusions), and self-correction. Applications of AI include expert systems, speech recognition and machine vision. It is important to emphasize that while AI can imitate human activities it is not capable of replicating human intelligence. AI’s usefulness in the accounting field has created a strong demand within the sector. From 2014 to 2016 venture capital funding for artificial intelligence companies increased by more than 50%. Clearly, AI has an increasingly important role to play in the daily activities of an accountant.
There are seven broad categories for Artificial Intelligence (AI). The first category is Reactive Machines (RM). RM is the oldest type of AI and has very little ability to mimic the human mind. These machines have limited memory-based functionality and can only be used for generating automatic responses to a limited number of combination inputs. One example of a RM is IBM’s Deep Blue, a machine that was used to defeat Chess Grandmaster Garry Kaparov in 19977. A second type of AI is the limited memory machine (LMM). LMM have the same abilities as RMs but can also learn from historical data to make decisions. Most AI applications today, including virtual assistants and self-driving vehicles, utilize LMM AI. Theory of mind (TOM) and Self-Aware AIs are two types of AI that are currently being developed or are only in the conceptual stage. TOM AI will be able to understand the entities it is interacting with by determining their needs, expectations, emotions and thought processes, through a process called artificial emotional intelligence. Self-Aware AI is purely hypothetical. This type of AI will be able to formulate ideas and potentially develop its own emotions, needs and desires. This type of AI is aware of its internal states, can predict the feelings of others and can make abstractions and inferences. Artificial Narrow Intelligence (ANI), Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI) are the final three categories of AI. ANI represents the existing AI in use today. This type of AI has the foundation for even the most complex AI that uses machine learning and deep learning to educate itself. AGI is the AI’s ability to perceive, function, and understand scenarios like a human being. AGI can perform the same activities as humans by mimicking our multifunctional capabilities. Finally, ASI will exhibit the characteristic of singularity, which would allow AI the ability to process and analyze data, develop and execute decisions, and possibly replicate the intelligence of human beings.
AI offers a variety of technological tools including expert systems, neural networks, robotics, fuzzy logic, genetic algorithms, and intelligent agents. Expert systems are programs that simulate decision making in a manner consistent with an expert in a field. The expert system provides a shell in which a software programing environment is created thereby allowing the construction of expert or knowledge-based systems. Expert systems require that a definable group of choices be available with a decision formulated based on following a series of logical steps. Neural networks, another tool provided by AI, simulate the electronic models of human brain neural structures. Neural networks allow machines to perform the functions of a human brain, including but not limited to, learning and critical thinking. Robotics is a technology concerned with developing a system that contains, sensors, control systems, manipulators, power supplies and software all working in tandem to perform a task.
Fuzzy logic is another Artificial Intelligence (AI) technology, and is the method of reasoning that emulates human reasoning and decision-making processes. Fuzzy logic is useful in several commercial and practical purposes, as its main purpose is to control machines, and provide solutions to complex problems encountered in accounting. Genetic algorithms balance the need for exploitation (choosing a suboptimal solution given a set of imperfect option) and exploration (developing a novel solution to a problem). Finally, intelligent agents provide the ability to handle the problem of information overload and autonomously reason through a complex set of possible scenarios.
AI has become an “innovation disruptor” and has found several major areas. AI technology has been especially useful in automating businesses processes to comply with new rules and regulations and creating personalized financial reports for individuals. IBM’s Watson can understand complex regulations and ensure that reports are produced meeting the current regulatory criteria. For example, IBM’s Watson can design reports, in a fraction of the time that it would take an accountant, to meet the requirements of the Market in Financial Instruments Directive or the Home Mortgage Disclosure Act. In addition, AI will allow bankers to make loan decisions in seconds rather than months as it will be able to provide an analysis of the risks, spending patterns, and payment history of the prospective clients. By automating this process accountants will be able to reduce the risk of default loans and improve customer service.
Report preparation and analysis is another area where Artificial Intelligence (AI) has grown in influence. Now, if a client wants to examine their financial health, they can utilize AI based software like Envestnet or Yodlee’s AI Fincheck which can provide the client with an assessment of how they are performing financially and make recommendation for actions in the future. This AI based tool can provide recommendations to help the client with saving and budgeting. Also, cash, credit and investment accounts can be analyzed using advanced analytics to determine the financial health of an institution. Envestment, for example, can extract information from comprehensive financial data sets and detect any trends in the profitability, expenses, or revenue generation.
AI has also improved the efficacy of Transaction Data Enrichment (TDE), a key component of financial management. TDE uses machine learning and natural language processing to decipher unintelligible strings of characters which represents transaction and merchants and converts the data into a more user-friendly readable format. The ease of transformation of unintelligible transaction data to a readable format allows banks and customers to determine where they spent their money and with which merchant. This also assists customers in quickly identifying fraudulent activity.
Predictive analytics is the fourth area where Artificial Intelligence (AI) has gained significant traction. Tools with machine learning AI have the power of predicting future outcomes critical to the survival of a company. For example, determining if the company will have the necessary funding to cover their outstanding debts in the upcoming ten years given their past and present spending patterns can be predicted using AI.
Finally, the creation of Artificial Intelligence (AI) enabled chatbots has significantly impacted the accounting field. These chatbots can perform routine tasks such as opening new accounts, transferring money between accounts, or assisting clients with basic problems, allowing accountants more time to focus on tasks that are more complex and client-professional oriented.
Given the vast capabilities of artificial intelligence, there is no question as to the critical importance that AI can play in the accounting profession. There are several accounting firms and companies that have invested heavily in developing AI integrated accounting projects. IN fact, a survey performed by Cognilytica on 3000 different US and International corporations found that 80% of corporate finance teams spend a majority (>80%) of their time currently on manually gathering and verifying financial data. A striking 97.6% of bookkeepers expect their jobs to be automated in the next decade and 73% of executives see AI improving their company’s financial health and efficiency.
The benefits of Artificial Intelligence (AI) in accounting are vast. According to the Associate of Certified Fraud Examiners, 13 percent of organization currently use AI or machine learning to detect fraud. An additional, 26 percent of organizations use biometrics and 25 percent of organization plan to implement AI technologies in the next three years. There are a multitude of benefits afforded to accountants with the use of AI today. One benefit of artificial intelligence includes that accountants will be able to gain a new sense of freedom. Machines will not replace the insight and judgements provided by accountants, but they will be able to solve complex financial problems and respond to simple queries, thereby saving accountants a large amount of time. Chatbots for example, can answer everyday questions from customers and are increasingly being integrated to handle tier 1 level support, answer questions about latest account balances, notify customers when bills are due, and the report the status of their accounts. AI also offers a way to deal with the large swaths of data available to accountants today. AI can parse, collect, analyze, manage, and sort through the data in a fraction of the time. Natural language processing capabilities implemented to navigate deeds, contracts and other documents are examples of how AI can effectively manage large swaths of data.
- Deloitte is one of many firms who have attempted to incorporate AI into their accounting practices. Deloitte developed a Limited Memory Machine AI-enabled document review process in 2014. This process utilized an expert systems AI tool to automate the process of review and extracting key information from various source documents. The benefit of Deloitte’s system has been to reduce the time spent reviewing legal contract documents, financial statements, meeting minutes, and invoice review by 50%. In addition, with the help of IBM Watson, Deloitte has been developing an LMM-cognitive-technology-enhanced business solution for its clients. This system uses intelligent agents and has been utilized to develop leasing portfolio analyses using LeasePoint and IBM Tririga. IBM Maximo technology was utilized with Automated Cognitive asset inspection to improve the efficiency of asset inspection. Deloitte also is the project leader for Catalyst, an exciting global initiative formed from a network of startups that are working to transplant AI technologies into business solutions for client firms. This AI enabled technology assists with the automation of document review and planning the infrastructure of start up companies. Catalyst utilizes natural language processing, natural language generation, machine learning, distributed ledger technology, cognitive automation and data wrangling to achieve its objectives. Working in conjunction with Kira Systems, the Catalyst system was able to read thousands of complex documents, extracting and structuring information for analysis by Deloitte’s professionals. Another arm of the Catalyst project is called Narrative Science. This arm uses natural language programing (NLP) to help assemble reports based on regulatory compliance, inventory management, financial analysis, and audit and tax fillings.
- EY services, formerly known as Ernst & Young, has also applied AI to the analysis of lease contracts. EY state that they can use AI to make it easier to capture relevant information including lease commencement dates, amounts to be paid and renewal or termination options. According to EY, these projects make it possible to review about 70-80% of simple lease contracts electronically. In the case of more complex leases, such as real estate, about 40% of the leases are capable of being reviewed by this AI system. EY has also developed AI-enabled auditing technology. The auditing system can accept and confirm audit requests, process them and provide auditors with relevant documentation for final analysis and judgement. Drones to monitor inventory during the auditing process are yet another AI enable technology launched by AI. The drones can count the number of inventory items in a production plant and communicate data directly into the global audit digital platform termed, EY Canvas. Finally, EY is using deep learning to analyze unstructured data such as email social media posts and conference call audio files. The data collected by this system reduces the administrative time spent on reviewing audit documents, giving employees more time for judgement and analysis.
- PricewaterhouseCoopers(PwC) has also engaged in multiple AI projects. For example, PwC is collaborating with an AI enabled (GLai) system capable of analyzing and preparing documents. This system learns and becomes more capable with every audit and has already been trained on audit data from Canada, Sweden, the United Kingdom and Germany. One application of PwC’s GLai system is working with Google’s optical character recognition (OCR)program to make it simpler for clients to recognize and properly claim both goods and services tax (GST) and value added tax refunds (VAT). The OCR provides an electronic conversion of images for printed or handwritten text files and converts it into machine readable files. Information is extracted from data fields form invoices/receipts via Google’s database. The information is validated against local currency VAT/GST document requirements. Expenses are classified for VAT via PwC’s GLai system. The new validated tax claim is produced with the amount per transaction reconciled to the source company’s expense system. Another PwC AI enabled project involves the use of natural language processing (NLP), an AI-enabled technology that processes unstructured data efficiently. This NLP system can analyze complex lease agreements, revenue contracts and board meeting minutes to generate insightful inferences.
- KPMG also has a portfolio of AI tools to enhance business decision and processes. Call center analytics use NLP to convert customer calls to unstructured text, which is then streamlined to identify keywords, predict future trends and gauge customer sentiment. This system can also anticipate anomalous events and determine the necessary information to extract from documents. Also, KPMG is working with Microsoft and IBM Watson to develop tools to integrate AI data analytics, cognitive technologies and RPA. The goal of this project is to consistently deliver high quality audit and accounting services. KPMG is also creator of Ignite, which is a system of AI tools that can be implemented into a firm to enhance business decisions. Ignite can detect patterns and create tools that are capable of developing and delivering AI created solutions. Also, Ignite will enable AI to have access to open source tools, libraries and APIs to build and delivery solutions.
- Another AI enabled accounting platform is called Vic.ai, which is currently being used in numerous accounting firms. This platform can process and understand documents through computer vision and document interpretation algorithms. For example, when an invoice is entered into the system the platform can combine the information to make predictions about the company’s future revenues or expenses. The Vic.ai platform can also classify general ledger accounts and predict how to post bills to the general ledger. Automated approval of bills is yet another feature of this platform. This can be especially useful when employees who normally oversee the process are on vacation or busy doing other tasks. Vic.ai can also execute and monitor business payments, learn from data accumulated over time and even modify its behavior based on human feedback.
- Botkeeper is another AI integrated platform. It provides automated bookkeeping support to businesses utilizing machine learning and the integration of systems and software. This platform acts as a virtual robotic bookkeeper that supports accountants needs. Botkeeper can access information necessary to make journal entries, track and schedule deferred revenues and expenses, handles payroll accounts, performs bank reconciliation, and sends invoices. The platform can also manage documents and develop monthly close and financial reports.
- Dokka is an AI powered instant bookkeeping that offers the ability to automate the journal entry process. It also produces, and manages financial statements, and facilitates messaging collaboration. Dokka is touted as one of the only fully automated accounting and smart document cloud solution for bookkeepers that brings instant data extraction together with smart document management.
- Sage, the market and technology leader in accounting has also invested heavily in providing its users with AI enabled technologies. In 2016 Sage announced five new accounting cloud solutions across seven global markets for businesses of all sizes. New Sage customers will be able to adopt accounting chat bot Pegg and by 2020 Sage expects to launch “invisible accounting.” Invisible accounting will function to automate the back-office functions and empower accountants to focus on building their businesses. Sage intends to launch their Invisible Accounting Program in five stages. Sage One will provide a cloud accounting solution for start up businesses. It will be focused on purchasing activities, sales activities, cash flow and taxes. Sage One with Pegg will be the worlds first accounting chat bot that will work in tandem with Sage One. Pegg, the smart assistant, will be able to track expenses, manage finances, and sift through popular messaging apps with little assistance from the user. Sage Live will provide a customizable environment that will allow users to manage multiple business locations and currency transactions all on a handheld device. Sage 50c will provide users with the option of integration with MS Office 365 to automate the steps needed to develop quarterly and annual financial statements. Sage 50c will be able to produce prepopulated Excel spreadsheets, graphs and charts. Finally, Sage People will provide human capital management, which is a set of procedures related to people resource management. Sage People will provide specific competencies that need to be implemented within the organization to facilitate the appropriate workforce acquisition, workforce management and workforce optimization.
- Intuit’s QuickBooks Online Advanced platform also offers AI enabled tools. Fathom is the smart reporting tool developed by QuickBooks, which Intuit says will help users of the software transform their data into “dynamic reports, allowing them to track their business results. Fathom provides a suite of in-depth analysis tools and metrics which help users see how well their business is performing. This tool utilizes NLP and machine learning to assess profitability, cash flow, growth and other key performance indicators (KPIs).
As the number of AI projects increase, several challenges have surfaced. One of the major challenges in incorporating AI based programs into the accounting profession includes the effects on the reputation of the accounting firm. One of any firm’s primary responsibilities is to present financial statements in an honest, accurate and transparent manner. With the advent of AI based technologies, business practices are moving into the virtual hands of AI systems. Businesses must rely on their programmers to properly program AI robots to perform the most critical functions. Any error could potentially be disastrous. Additionally, AI programs are being given the freedom to draw conclusions and make complex judgements. With the reassignment of responsibilities to AI based programs, the firm is becoming less involved in the process.
The high investment and slow growth of AI also limits its integration into the accounting field. Due to the customized nature of AI enterprises need a large amount of capital in the initial application and operation of the system. Once implemented training and hiring of professionals must begin and adjustments must be made to properly incorporate the AI into the business. The short-term returns may be limited given the need to align the AI system to conform with the businesses objectives.
AI integration within the accounting profession also faces the issue of meeting compliance standards. Government regulation on what AI is capable of accomplishing is ill-defined. Given the lack of a formal set of standards, businesses could take advantage of certain privacy policies. A report released from Stanford University addressed this concern. Their report concluded that attempts to regulate AI in general would be misguided as there is no single definition of AI and the risks and considerations are unique to each domain. It is clear that as the technology continues to evolve, the government will continuously need to adapt to keep up.
Another important barrier to the integration of AI into the accounting profession is the lack of understanding by regulators regarding how institutions have integrated AI-based systems into their practice. Specifically, regulators within the audit and accounting fields will need to communicate directly with the institutions who have implemented AI-based technologies to examine corporate financial statements and effectively discharge their duties. Regulatory agencies are already discussing the impact of data analytics and their impact on current accounting standards. One important point of contention is concerned with addressing the issue of transparency. The lack of a complete understanding in deep learning models and the way in which their outputs are derived has hampered regulators from certifying that the firm’s statements have been prepared in a transparent manner. As institutions and audit firms continue to increase their reliance on novel AI systems, regulators will need to develop protocols to ensure that their analyses are both uniform and compliant with regulatory compliance standards.
Adam Smith once stated that business decisions are made on the notion that there must be an exchange of information between human beings. AI circumvents this notion, as ideas cannot be shared, questions cannot be raised, and problems cannot be solved through an open dialogue or line of communication. Essentially, AI limits the collaborative process required in making everyday business decisions. Beyond the loss of the collaborative process, AI has the potential to inhibit the creativity of visionaries within the organization. With a shift in focus on making AI programs capable of taking on more human responsibilities, a potential reduction in the efforts of what the human mind is capable of is a pivotal concern.
Managing the large amounts of data volumes and their quality are crucial to the successful integration of AI systems within the field of accountancy. AI systems require accurate data to develop models and learn. For example, transactional data, which is repetitive and structured, can play an essential starting point for the development of models. However, given long-standing philosophies embedded with the internal environment of a firm, creating valid data to structure an AI-system model will likely continue to be an obstacle. In addition, smaller firms may encounter the problem of not having enough data to develop accurate models or determine the appropriate action in dealing with a specific problem. Powerful, complex, and multifaceted models may require external sources of data to properly analyze and resolve a problem. However, the high cost to acquire this information may limit these systems from acquiring the external data. The success of machine learning is also difficult to predict. Models learn by performing tasks in repetition. Both the success and failures of the models in producing the requested output can help build a more adaptive model.
Another important limiting factor in AI integration is the privacy and ethical issues that are inherent to the use of AI systems. Fraud detection tools, for example, may access information stored in the form of text messages or emails sent and received by employees. The acquisition of this information may require both a legal and ethical examination.
Lastly, for AI integration to be successful within the accounting profession, the firm will need the right personnel with the necessary training and technical skills to utilize machine learning tools. Personnel will not only require the necessary technical skills in utilizing AI, but they will also need to understand the implementation of the tools within the context of business data analytics. For example, accountants will need to be involved in training or testing models, or auditing algorithms. In addition, accountants will need to outline problems and solutions and determine how they fit into the context of a business process, and others may be needed to deal with data preparation. The success of these activities will require accountants to have an in depth understanding of machine learning techniques so that they can design the best models and speak intelligently with experts and consultants on how best to utilize these models.