virtual auditing agents: the edgar agent challengeaccounting.rutgers.edu/miklosvasarhelyi/resume...

13
Ž . Decision Support Systems 28 2000 241–253 www.elsevier.comrlocaterdsw Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelson a, ) , Alex Kogan b , Rajendra P. Srivastava c , Miklos A. Vasarhelyi b , Hai Lu a a School of Accounting and Information Systems, DaÕid Eccles School of Business, The UniÕersity of Utah, Salt Lake City, UT 84112,USA b Department of Accounting and Information Systems, Faculty of Management, Rutgers UniÕersity, Newark, NJ 07102-1895,USA c Ernst and Young Center for Auditing Research and AdÕanced Technology, DiÕision of Accounting and Information Systems, School of Business, The UniÕersity of Kansas, Lawrence, KS 66045,USA Abstract Intelligent agents can be used as agents of organizational change. This potential exists in the domain of accounting audit, where much of what is currently done manually in batch mode could be done continuously and on-line. We discuss the use of intelligent Internet agents as a way of changing and expanding audit practices in the virtual world. A qualityrservice framework is presented that suggests ways that accounting firms can evolve in this era of on-line opportunities. The EDGAR Agent is presented as an example of an intelligent Internet agent that gathers financial information. The challenges involved in the development of the EDGAR Agent are analyzed, providing insight into the practical aspects of agent technology designed for a specific business domain. A test of the agent is presented, with comments and suggestions from financial practitioners that will be integrated into the research stream. q 2000 Elsevier Science B.V. All rights reserved. Keywords: Intelligent agent; Accounting audit; EDGAR; Agent technology 1. Introduction The virtual world presents many challenges and opportunities to accounting firms. The Internet has exponentially increased the speed and availability of information to consumers. Many of these consumers are the clients of accounting firms that in some respects have previously been taken for granted. This ) Corresponding author. Ž . E-mail addresses: [email protected] K.M. Nelson , Ž . [email protected] A. Kogan , Ž . [email protected] R.P. Srivastava , Ž . [email protected] M.A. Vasarhelyi , Ž . [email protected] H. Lu . explosion of information has resulted in many of these clients questioning the value of standard ac- counting services such as auditing 1 that provide accurate yet greatly delayed information. Organiza- tions need to know on a real time basis how they are performing. The investors and management of these organizations would also like to have this knowledge 1 Here, we are implying financial audit by the term auditing. Ž. Traditionally, there are three types of audit services: 1 Audit of the financial statements of a company to provide assurance on the Ž. financial statements, 2 Operational audit to check for the opera- Ž. tional efficiency and effectiveness of a process, and 3 Compli- ance audit to find whether the company has complied with the required laws and regulations. 0167-9236r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved. Ž . PII: S0167-9236 99 00088-3

Upload: tranduong

Post on 07-Oct-2018

227 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

Ž .Decision Support Systems 28 2000 241–253www.elsevier.comrlocaterdsw

Virtual auditing agents: the EDGAR Agent challenge

Kay M. Nelson a,), Alex Kogan b, Rajendra P. Srivastava c, Miklos A. Vasarhelyi b,Hai Lu a

a School of Accounting and Information Systems, DaÕid Eccles School of Business, The UniÕersity of Utah, Salt Lake City, UT 84112,USAb Department of Accounting and Information Systems, Faculty of Management, Rutgers UniÕersity, Newark, NJ 07102-1895,USA

c Ernst and Young Center for Auditing Research and AdÕanced Technology, DiÕision of Accounting and Information Systems,School of Business, The UniÕersity of Kansas, Lawrence, KS 66045,USA

Abstract

Intelligent agents can be used as agents of organizational change. This potential exists in the domain of accounting audit,where much of what is currently done manually in batch mode could be done continuously and on-line. We discuss the useof intelligent Internet agents as a way of changing and expanding audit practices in the virtual world. A qualityrserviceframework is presented that suggests ways that accounting firms can evolve in this era of on-line opportunities. The EDGARAgent is presented as an example of an intelligent Internet agent that gathers financial information. The challenges involvedin the development of the EDGAR Agent are analyzed, providing insight into the practical aspects of agent technologydesigned for a specific business domain. A test of the agent is presented, with comments and suggestions from financialpractitioners that will be integrated into the research stream. q 2000 Elsevier Science B.V. All rights reserved.

Keywords: Intelligent agent; Accounting audit; EDGAR; Agent technology

1. Introduction

The virtual world presents many challenges andopportunities to accounting firms. The Internet hasexponentially increased the speed and availability ofinformation to consumers. Many of these consumersare the clients of accounting firms that in somerespects have previously been taken for granted. This

) Corresponding author.Ž .E-mail addresses: [email protected] K.M. Nelson ,

Ž [email protected] A. Kogan ,Ž [email protected] R.P. Srivastava ,

Ž [email protected] M.A. Vasarhelyi ,Ž [email protected] H. Lu .

explosion of information has resulted in many ofthese clients questioning the value of standard ac-counting services such as auditing1 that provideaccurate yet greatly delayed information. Organiza-tions need to know on a real time basis how they areperforming. The investors and management of theseorganizations would also like to have this knowledge

1 Here, we are implying financial audit by the term auditing.Ž .Traditionally, there are three types of audit services: 1 Audit of

the financial statements of a company to provide assurance on theŽ .financial statements, 2 Operational audit to check for the opera-

Ž .tional efficiency and effectiveness of a process, and 3 Compli-ance audit to find whether the company has complied with therequired laws and regulations.

0167-9236r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.Ž .PII: S0167-9236 99 00088-3

Page 2: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253242

Table 1Ž w x.Requirements for virtual transactions Bhimani 3

Confidentiality Communications are restricted to only the parties involved in a transactionAuthentication Assurance that parties are communicating with whom they think they are doing business.Data Integrity Data sent in a transaction should not be modifiable in transitNonrepudiation Neither party can deny participation in a transaction after the factSelective application of services Part of a transaction is hidden from view while another part is not

more frequently than is currently provided by thew xannual audit 9 . Accounting firms that do not re-

spond to these client needs will find themselves at acompetitive disadvantage. The very existence of thevirtual global environment will cause the currentaccounting audit model to evolve or die.

w xCommenting on the audit function, Elliott 10 , p.106 says:

The information revolution is greatly changing theattest environment, and we must try to understandthe nature of the changes in order to determinehow best to respond to them. These changes andprospects call for responses to adapt internal andexternal accounting, accounting education and re-search, accounting firms, and attest work.

The accounting profession is responding. Re-Žcently, the AICPA American Institute of Certified.Public Accountants Board has formed a Special

Committee on Assurance Services to develop astrategic plan for identifying and implementing anexpanded assurance function2 resulting from the in-

w xformation revolution 14 .There is a great need for new types of services.

Some of these services will be built on the corecompetencies that already exist and some will haveto be developed. These services will be built aroundthe demand for confidentiality, authentication, dataintegrity, non-repudiation, and selective applicationof client services that are dictated by the virtual

2 Assurance function, in general, deals with providing assur-ance on a process, service or a piece of information. An assuranceservice is a generalized version of the audit service. Similar toaudit services, the auditor collects, evaluates and aggregates itemsof evidence relevant to the objective to provide assurance. Forexample, the auditor can provide assurance on the quality of aservice provided by a hospital.

w xenvironment of the Internet 3 . Table 1 illustratesthese transaction needs in the virtual world.

2. Intelligent agents

Much of the work done in the virtual world willbe done by intelligent agents. These agents can bedefined as programs that operate autonomously toaccomplish unique tasks without direct human super-

w xvision. Minsky and Riecken 16 has described intel-ligent agents in a more pragmatic way, describingthem as ‘‘go-betweens’’ that possess specializedskills, similar to travel agents or insurance agents.By having intelligent agents do the more mundaneand manual tasks of the audit function, the opportu-nity exists for auditors to become more knowledge-able about and responsive to their clients. Agentsshould enhance human intelligence and help to make

w xpeople smarter 18 . In the case of auditors, this cantranslate to the ability to provide new and innovativeservices to clients that extend the practice of auditingbeyond the traditional scope.

This paper addresses the issue of intelligent agencyand virtual auditing by providing an example of anagent that renders a value-added service that canpotentially be offered in an auditing practice. We callthis program the EDGAR Agent since it searches the

Ž .EDGAR Security and Exchange Commission SECdatabase for the current 10Q report of listed compa-nies. This paper outlines the challenges faced whendeveloping an intelligent agent for a specific busi-ness domain, in this case, the accounting domain offinancial reporting.

3. The auditing function in the virtual world

Traditionally, the audit function has been limitedto the verification and accuracy of financial reporting

Page 3: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253 243

in an organization. An exception to this has been thearea of EDP auditing. This practice began as anadditional financial reporting verification process,but in many cases has been extended to verifying theoperation of computing systems in the organization.By looking at the EDP function in this way, it seemsonly natural to assume that the Internet would pro-vide opportunity to extend the financial reportingverification process to provide new and hopefullyprofitable services to clients in the virtual world.

These new services can be classified into twow xgeneral categories, quality and service 13 . Tradi-

tionally, the practice of auditing has focused on thequality dimension. Through the verification of finan-cial reporting3, the quality of information providedby the firm to its stakeholders is assured. This assur-ance is similar to a radiologist assuring that a medi-cal X-ray is accurate and provides a quality diagno-sis, or a building inspector assuring that a house isbuilt to specifications and will hold up the roof.

While doing auditing assurance, many accountingfirms have discovered the lucrative practice of busi-ness and information systems consulting. This is anatural extension of the assurance process, sincebusiness processes and information systems can di-rectly affect the quality of organizational informationw x8 . These consulting practices focus on the servicedimension. Both the quality and service dimensionscan be extended under the auditing umbrella in thevirtual world, as shown in Table 2.

Internet security is one of the biggest barriers tow xdoing business electronically 5,7 . For intelligent

agents to succeed and be accepted, auditors, theirclients, and the customers and stakeholders of theirclients must believe the Internet is secure. Agents

w xneed to be seen as competent and trusted 15 . If thetransaction environment is not seen as secure, agentswill not be accepted. However, the security issue can

3 In a financial audit, the auditor collects, evaluates, and aggre-gates various items of evidence relevant to the five assertionsŽexistence or occurrence, completeness, rights and obligations,valuation and allocation, presentation and disclosure, see Arens

w x .and Loebbecke 9 , for definitions made by the management foreach account on the balance sheet. Based on all the evidencecollected, the auditor determines whether each account balance isfairly stated and then provides an overall opinion on the financialstatements whether they are fairly presented.

Table 2Virtual auditing qualityrservice framework

Quality Service

Verification of transactions Nonrepudiation ofand data transactionsAuthentication of transactions Proxy intelligenceand data searchesIntegrity of transactions Real time database searchand data and reportingCompleteness of transactions Translation torfromand data EnglishTimeliness of transactions Competitive intelligenceand data

be turned into a profit opportunity for auditing prac-tices. Auditors, and their intelligent agents, can ex-tend the quality dimension of the audit practice toinclude the verification of confidentiality, authentic-ity, integrity, completeness, and timeliness of infor-mation found on the Internet. These security-orientedservices would be an extension of the existing corecompetency of financial verification that already ex-ists in audit practices. By furnishing these verifica-tion services, the providing accounting firm wouldalso be seen as competent and trusted, and thesetraits would be endowed on the intelligent agent thatis a proxy for the firm.

Electronic commerce also provides the opportu-nity for many new consulting services that can beoffered under the auditing practice umbrella. Theseservices include the acceptance of agents workingunder the proxy of an established and trusted ac-counting firm, in other words, the nonrepudiation ofthe services or information provided by or soughtafter by these agents. These agents will act as inter-mediaries between the auditing firm and its clients,as well as a proxy for the client in the virtual world.These proxies can perform text-based searches forinformation within the client’s databases and withinother databases on the Internet. These database searchcapabilities provide the opportunity for real timedatabase reporting and updates. Intelligent agents canalso provide translation capabilities to clients thatmay not be English speaking. Finally, agents canconstantly search the virtual world to gather informa-tion that auditing practices can convert to valuableand salable client competitive knowledge.

Page 4: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253244

Besides these new opportunities, even the tradi-tional audit function will change in the Internet era,especially in terms of the nature of evidence and theway it is accumulated.4 Intelligent agents will play

w xan important role in this process 19 . For example,in performing analytical procedures5 as required by

w xthe AICPA 1 , the auditor can use an agent to gatherindustry information through the Internet and per-form the analysis irrespective of the place and time.

4. The EDGAR Agent

The EDGAR Agent is a first step toward on-linecontinuous auditing and expanding the role of the

w xaudit function in the virtual world 19 . The EDGARAgent is a category of agent that assists in informa-

w xtion access and management 2 . The EDGAR Agentis an easy-to-use intelligent agent that searches theSEC EDGAR database for the current cash balanceof a user-specified corporation. The user then isasked to choose the industry group for the companybefore hershe submits the query. The agent thensends the query including the corporation and indus-try names to the EDGAR database server. There are

Page 5: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253246

Table 3A comparison of Intel and Ford consolidated balance sheets

Intel Ford

Assets AssetsCurrent assets AutomotiveCash and cash equivalents Cash and cash equivalentsShort-term investments Marketable securitiesTrading assets Total cash and marketable securitiesAccounts receivable, net ReceivablesInventories InventoriesRaw materials Deferred income taxesWork in process Other current assetsFinished goods Net current receivable from Financial Services

Deferred tax assets Total current assetsOther current assets Equity in net assets of affiliated companies

Total current assets Net propertyProperty, plant and equipment Deferred income taxesLess accumulated depreciation Other assetsProperty, plant and equipment, net Total Automotive assetsLong-term investments Financial ServicesOther assets Cash and cash equivalentsTotal assets Investments in securities

Net receivables and lease investmentsOther assetsNet receivable from AutomotiveTotal Financial Services assetsTotal assets

Liabilities and stockholders’ equity Liabilities and stockholders’ equityCurrent liabilities AutomotiveShort-term debt Trade payablesAccounts payable Other payablesDeferred income on shipments to distributors Accrued liabilitiesAccrued compensation and benefits Income taxes payableAccrued advertising Debt payable within one yearOther accrued liabilities Net current payable to Financial ServicesIncome taxes payable Total current liabilities

Total current liabilities Long-term debtLong-term debt Other liabilitiesDeffered tax liabilities Deferred income taxesPut warrants Total Automotive liabilitiesStockholder’s equity Financial ServicesPreferred stock PayablesCommon stock and capital in excess of par value DebtRetained earnings Deferred income taxes

Total stockholder’s equity Other liabilities and deferred incomeTotal liabilities and stockholders’ equity Net payable to Automotive

Total Financial Services liabilitiesStockholders’ equityCapital stockPreferred StockCommon StockClass B Stock

Capital in excess of par value of stock

Page 6: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253 247

Ž .Table 3 continued

Intel Ford

Liabilities and stockholders’ equityForeign currency translation adjustments and otherEarnings retained for use in businessTotal stockholders’ equity

Total liabilities and stockholders’ equity

code, considering that many people are the enthusias-tic believers of using scripting language along withCrCqq or Java to improve performance. In theEDGAR Agent, we found this treatment did notsignificantly improve speed.

Ž .3 Choose a better local serÕer. The EDGARAgent code is expected to be platform independent.Initially we worked on an IBM RS 6000r350 serverwhere the CPU time has been widely distributed tothe extensive Email users. We moved to a DECAlphaServer 1000A 5r400 server that is used pri-marily for research computing, and the running speedimproved as expected.

5.2. Format of the financial statements

There are significant differences in the financialstatements between companies in the EDGARdatabase, even when those companies are in thesame industry. These format differences directly im-pact the accuracy of the information returned by theEDGAR Agent. In order to illustrate the point, asimple comparison of the balance sheets of twocompanies, Intel and Ford Motor, is shown in Table3. Both are manufacturing companies, but they are indifferent industries. We can see the obvious differ-ences in format even though they both follow stan-dard accounting format. We performed an analysis of66 companies across a variety of industries andfound that the format variety is reduced for thecompanies within the same industry. For example,both Motorola and Intel manufacture electronic chips,and their financial statements are very similar. Thatis the primary reason the user is prompted to inputthe industry at the beginning of an EDGAR Agentinquiry. Another reason is that it is then easier for aselected company to be compared with its peer com-

panies. Through our industry survey, we found thisclassification helps eliminate the difficulties causedby the format differences in retrieving and process-ing the financial statements.

5.3. The Õariety of the accounting terms

The most challenging issue in the EDGAR Agentdevelopment was how to deal with the variety of theterms used in the financial statements. Neither theSEC nor the AICPA impose standards on theseterms. Companies have complete freedom in choos-ing the names used to describe financial items suchas cash. At the current time, we are most concernedwith the terms used in the ratio calculations. Some ofthese have few variety, for example, the term ‘‘Cash’’is being used by 9.1% companies of our sample,‘‘Cash and cash equivalents’’ by 69.7%, ‘‘Cash anddue from banks’’ by 12.1%, with the remaining9.1% of the companies using other terms such as‘‘Cash and interest bearing equivalents’’. In contrast,some terms have significant variety, such as‘‘Accounts receivable’’, which has more than 20synonyms in the 66 balance sheets analyzed.

The first approach adopted to deal with the termvariety was the use of a direct pattern matchingfunction. The approach worked very well for a fewspecific tested companies. However, each time weencountered a new synonym that was not included inthe judgement criteria written in the agent code, weexpanded the criteria. After a while we began to hit awall because it was almost impossible to make oneor two exclusive matching criteria. Using complexjudgement criteria greatly decreased the speed of theagent since every component of the criteria had to bechecked over the whole file line by line.

Page 7: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253248

Our second approach to this problem was statisticw xinformation retrieval 6 . We decomposed each term

into single words and assigned a certain probabilityto each word according to the possibility of itsappearance in the corresponding standard term. Thecorresponding term with the highest probability wasthen selected. For example, the word ‘‘Cash’’ almostcertainly appears in ‘‘Cash and cash equivalent’’,and a very high probability should be assigned.However, it is less likely for the term to appear inthe synonyms of ‘‘Accounts receivable’’, so a verylow probability should be assigned in this case.Furthermore, ‘‘Cash’’ is impossible to be included inthe ‘‘Current liability’’, so the probability is zero.Using Cqq to write the test code, and with limitedtests, it was found the statistic retrieval conceptcould be adapted to agent development. The draw-back to this approach is that it is hard for a developerto learn the correct assignment of probability withoutan extensive investigation into accounting terminol-ogy.

Our chosen approach to solving the terminologyproblem is a combination of Perl code and a rela-

w xtional database 24,25 . With the assumption thatsynonyms are close to exclusive if a fairly largesample can be maintained, we believe a table in arelational database will help us overcome the termrecognition obstacle. This table has a one-to-manyrelationship. Carrying the necessary words, the agentgoes to the database to search for an exact match.The process is triggered from the agent script to thedatabase interface, then to the ODBC driver andfinally to MS Access or Oracle. When the agentdetermines an exact match in the synonym field ofthe table, it will retrieve the corresponding standardterm and its value. Without an exact match, the agentwill retrieve nothing and return. An alternativemethod being tested is having the agent make anautonomous decision about which standard term ismost likely used when there is no exact match. Thisapproach utilizes intelligence within the agent, andthe implementation can be the combination of thestatistic retrieval approach and this database ap-proach or other artificial intelligence methods.

While the above syntactic problems are real andchallenging, there are two even deeper challengesunderlying them. These challenges are semantic and

w xcontextual 21 . Semantic challenges exist in the way

information is interpreted by auditors, and how intel-ligent agents will deal with this. Context challengesarise when agents search the web at large and findfinancial information on web sites, but need to deter-mine if the context of this information is relevant forretrieval. These issues are currently being studied byother members of our research team in parallel withthe Edgar agent development.

The challenges to building the EDGAR Agentdemonstrate the constraints placed on the technicaldevelopment of intelligent agents. While our agent isrelatively simple from a technical perspective, the‘‘messy’’ world of financial reporting required somecreative solutions to problems not found in the purelaboratory environment. Our next step in dealingwith these ‘‘messy’’ problems was to conduct alimited, preliminary field test of the agent.

6. Preliminary field test of the EDGAR Agent

The Edgar agent was reviewed by thirty profes-sionals representing the management, accounting, andfinance areas. These respondents were members of abusiness school graduate program and all had somecurrent or previous work experience. The instrumentused to evaluate the agent is included in Appendix Aof this document, and was based on previously vali-

w xdated measures 22,23 . While the statistical datapresented below in Table 4 is interesting and usefulfor design purposes, the feedback we received incomment form was even more insightful.

One of the things that came through strongly wasthe issue of trust in Internet data. While most of ourrespondents thought the information looked goodand could be useful, they indicated that they had noway of knowing if the data was accurate. For exam-ple:

The responses of NrA are indicative of the factthat I did not audit the results for accuracy;presuming that the results are accurate, the infor-mation is very useful.

I did a comparison to Bloomberg data on a fewcompanies. The data pulled from Bloomberg var-ied slightly in several areas. We would like to

Page 8: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253 249

Table 4EDGAR agent evaluation results Ns30, seven-point Likert scale

Item Mean STD

The quality of data produced 5.69 0.84The format of the output 3.62 1.53The accuracy of the data 4.10 2.17The timeliness of response 5.54 1.24The speed of response 5.69 1.16The structure and organization of the screens 4.62 1.02The look and feel of the agent 4.00 1.70Overall quality 5.00 0.80Ease of use 5.62 1.02User friendliness 5.00 1.33Understandability of on-screen instructions 5.54 1.03Understandability of output 5.31 1.16Usability of the agent 5.31 0.93Usefulness of data produced 4.92 1.89Commercial applicability of the agent 5.50 0.98

know why they vary. The comparison was withADM after the close on November 18.

Based on the above comment, we checked thatparticular data point and found that we were accu-rately reporting numbers from Edgar and the stock

w xmarket. However, Bloomberg 20 was reporting dif-ferent data, showing that even ‘‘official’’ sources onthe Internet seem to be varying slightly. These com-ments and our follow-up investigation indicate thatInternet security and accuracy are issues that must beaddressed for agent technology to be used success-fully in real business domains such as auditing.

The other comments were primarily around for-matting issues.

I like the fact that you always have the option of alink back to the search page, rather than having touse the back key in the search engine. The pro-gram is very plain, while it is easy to read, a‘‘nicer package’’ would make it seem more pro-fessional. The format of the output forced me toscroll over an inch to see the last column and thenI couldn’t see what the first column said. Maybelines would help readability.

Based on these comments, we are currently updat-ing our user interface using focus groups of financialprofessionals as respondents.

7. Discussion and future research

In addition to financial information, intelligentagents have the capability of searching the Internetfor environmental and competitive information abouta firm or its clients that is present in the media. Theability of agents to search in full text mode willallow them to be trained to search for competitivethreats and opportunities. Different types of artificialintelligence can be used so that agents learn as theysearch. Neural networks are one type of tool thathave the potential to build continuously learning

w xagents 17 .Intelligent agents that can provide not only finan-

cial information and calculations about a company,but also can reconcile this information with environ-mental data have the potential to be marketed as avalue-added service to clients of accounting firms.This service can be used internally by the clientcompany for auditing and decision-making, or couldbe used externally by the client to assess and makedecisions about the competitive environment. In thisway, accounting firms can extend existing audit andconsulting services under the qualityrservice frame-work. As accounting firms enable their clients togather information and participate in the virtual globalworld, they will also profit from providing securityand authentication of transactions as they are per-formed.

Direct research resulting from this study will ex-tend and refine the EDGAR Agent’s capabilities forgathering, isolating and analyzing key financial datafrom the EDGAR database. Additional refinementswill include intelligence on the home computer thatcalculates financial ratios in a way that is useful andmeaningful for corporate decision-makers. TheEDGAR Agent lays the foundation for additionalresearch in the area of on-line financial reporting andvirtual auditing. For example, one can integrate suchan agent with artificial neural networks to develop

w xmodels for predicting bankruptcies 11 , potentialsw xfor management fraud 12 , or for going concern

w xjudgments 4 with timely information of the clientand the industry.

We plan to gradually build up the EDGAR AgentŽinto the FRAANK Financial Reporting and Audit-

.ing Agent With Net Knowledge Agent capable ofdemonstrating numerous features of a universal on-

Page 9: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253250

line auditor’s assistant. We envision the FRAANKŽ .Agent Fig. 3 as having multi-tier architecture with

the agent’s logic clearly separated from the end userinterface on the one hand, and from the data andknowledge sources on the other hand.

The current implementation of the EDGAR Agentas a monolithic Perl program will be redesigned forthe modular architecture presented above. TheFRAANK Agent will include the programming logicwritten in PERL, KQML and Java, and the data andknowledge sources implemented in an SQL database,with the internal communications implemented usingODBC over TCPrIP. The heterogeneity of the im-plementation is the result of choosing the most ap-propriate tool for each task. Thus, the analysis andinformation extraction from natural text documentswill be implemented in Perl and KQML, while mo-bile applets that FRAANK might need to spawn willbe written in Java.

We plan to continuously develop and enhance theAI capabilities of the FRAANK Agent. It is howevernot realistic to expect that a single agent can embodyall the sophisticated stand-alone expert systems de-veloped for the accounting and auditing professions.Important steps on the way leading from the EDGARAgent to the FRAANK Agent will include:

Ø More detailed analysis of SEC filings to extractfinancial statements and compute financial ratios.

Such analysis will require the creation and incor-poration into our agent of a data source of ac-counting terms and their synonyms.

Ø Interaction with additional information sources,e.g., querying a stock quote server to compute thecurrent market value of a company, which can beused to compute Altman’s Z-factor.

Ø Formal representation and learning of the struc-ture of external information sources aimed atfinding important financial and accounting infor-mation in companies’ Web sites of arbitrary struc-ture.

Ø Collaboration with external agents, e.g., the useof existing ‘‘news agents’’ for gathering recentnon-financial information about companies.

This stream of research could lead to the develop-ment of intelligent agents that continuously accessand search client databases for information and re-port this information back to the auditing firm. It isclear that such agency would require a high degreeof competency on the part of the agent as it acts as aproxy for the human auditor. This type of agencyalso requires a high level of authentication and veri-fication that the information found in the database isup-to-date and factual. For the client of an account-

Fig. 3. The FRAANK Agent architecture.

Page 10: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253 251

ing firm to allow such an agent into its database, ahigh level of trust is required. However, the rewardsof such a relationship would be worthwhile to boththe accounting firm and its client. A close partner-ship would be developed whereby the accountingfirm no longer provides only historical informationand verification to the client, but also provides valu-able decision-making and competitive knowledge.The intelligent agent becomes the third partner in arelationship whose goal is to make the client organi-zation more profitable.

Appendix A. User questionnaire

A.1. The UniÕersity of Kansas study of intelligentinternet agents

The following questionnaire asks a series of ques-tions about a developmental intelligent agent,FRAANK. It is important that you answer the ques-tions in the survey carefully and honestly. Thisresearch will only be of value if you tell us what youreally think. Please contact Kay Nelson at KU if you

Ž .have questions 785 865-7529 or [email protected]

Your participation in this research is voluntary. Ifyou object to the questionnaire or to a specificquestion, you may chose not to respond. Your coop-eration is strongly desired, and the KU research teamhas taken measures to ensure that your responsesremain confidential. All answers to this questionnairewill be kept strictly confidential. At no time will thisquestionnaire be shown to anyone else in your orga-nization. Responses will be reported only at thecompany level, not at the individual project level.Once the data has been collected and entered into thecomputer, completed questionnaires will be de-stroyed, leaving only the coded data as a record ofresponses.

Thank you for your participation in this study.Please e-mail responses to Kay Nelson, [email protected].

The following demographic information is forpurposes of project tracking only. This information,

as well as all information in this questionnaire willbe seen only by the researcher and will be keptstrictly confidential under The University of Kansasresearch guidelines.

1. Name________________________________

2. Employer_____________________________

3. Employer UnitrDepartment_________________________

4. Academic program______________________

5. Credits completed toward degree ______6. Age ______7. Gender ____F ______M8. Number of Years Using Internet ________

Ž .9. Internet Expertise Check one _______Novice__________Average _________Expert

10. Average number of hours spent weekly on theInternet _________

11. Have you ever purchased anything via the Inter-net? ____Y ____N

12. Have you used a credit card on the Internet?_____Y _____N

13. Name of Internet provider ______KUŽ .______Other please specify

A.2. Instructions

To complete the survey, please go to the follow-ing URL: http:rrpink7.busmis.ukans.edurFraankragent.pl

Please test the companies listed below followingthe directions provided on the agent. When you arefinished, please complete the following survey andemail it to [email protected]. If possible, pleaserespond by Friday, November 20th. All respondentswill receive a token gift with the KU Logo.

Company list:

Rainbow TechnologiesRainbow RentalsSunquest Information SystemsSunglass HutSunrise Assisted Living

Page 11: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253252

SanifillTexacoComputer MotionComputer Task GroupGreen Mountain CoffeeCharles RiverExpress Scripts

This following questions ask for information abouthow well the Edgar Agent performed when youtested it. Please read each question individually and

Žanswer it to the best of your ability place theappropriate number or NrA after the question in the

.space provided .Ž .1 Rate the performance of the Edgar Agent on

the following characteristics:

Not Poor Average Excellentappli- perfor- Perfor- perfor-cable mance mance manceNrA 1 2 3 4 5 6 7

a. The quality of data produced: ___b. The format of the output: ___c. The accuracy of the data: ___d. The timeliness of response: ___e. The speed of response: ___f. The structure and organization of the screens:___g. The look and feel of the agent: ___h. Overall quality: ___i. Ease of use: ___j. User friendliness: ___k. Understandability of output: ___l. Understandability of on-screen instructions: ___m. Usability of the agent: ____n. Usefulness of data produced: ____o. Commercial applicability of the agent: _____

Please add any additional comments aboutFRAANK.

References

w x1 American Institute of Certified Public Accountants, in:AICPA Professional Standards Vol. 11996, AU Section 329.

w x2 G. Aparico, The Role of Intelligent Agents in the Informa-tion Infrastructure, IBM, 1995.

w x3 A. Bhimani, Securing the commercial internet, Communica-Ž . Ž .tions of the ACM 39 6 1996 .

w x4 S.F. Biggs, M. Selfridge, R. Krupka, A computational modelof auditor knowledge and reasoning process in the going-concern judgment, Auditing: A Journal of Practice and The-

Ž .ory, Supplement 1992 .w x5 N. Borenstein et al., Perils and pitfalls of practical cybercom-

Ž . Ž .merce, Communications of the ACM 27 7 1996 .w x6 L. Egghe, R. Rousseau, Topological aspects of information

retrieval, Journal of the American Society for InformationŽ . Ž .Science 49 13 1998 .

w x7 K. Crowston, Market-enabling internet agents, in: Proceed-ings of the Thirtieth Annual Hawaii International Conferenceon System Sciences, 1997.

w x8 T.H. Davenport, Process Innovation: Reengineering Workthrough Information Technology, Harvard Business SchoolPress, Boston, MA, 1992.

w x9 A.A. Arens, J.K. Loebbecke, AUDITING: An IntegratedApproach, Prentice Hall, NJ, 1994.

w x10 R.K. Elliott, Confronting the future: choices for the attestŽ . Ž .function, Accounting Horizons 8 3 1994 .

w x11 K. Fanning, K. Cogger, A comparison analysis of artificialneural networks for financial distress, International Journal of

Ž .Intelligent Systems in Accounting and Management 3 4Ž .1994 .

w x12 K. Fanning, K. Cogger, R. Srivastava, Detection of manage-ment fraud: a neural network approach, International Journalof Intelligent Systems in Accounting and Management 4Ž .1995 .

w x13 P. Konana et al., Pricing of information services usingreal-time databases: a framework for integrating user prefer-ences and real-time workload, in: Proceedings of the Seven-

Žteenth International Conference on Information Systems De-.cember, 1996 , 1996.

w x14 R. Lea, Opportunities for assurance services in the 21stcentury: a progress report of the special committee on assur-ance services, in: Proceedings of the 1996 Deloitte andToucherUniversity of Kansas Symposium on Auditing Prob-

Ž .lems May 1997 .w x15 P. Maes, Agents that reduce work and information overload,

Ž . Ž .Communications of the ACM 27 7 1994 .w x16 M. Minsky, D. Riecken, A conversation with Marvin Minsky

Ž . Ž .about agents, Communications of the ACM 27 7 1994 .w x17 T. Mitchell et al., Experience with a learning personal assis-

Ž . Ž .tant, Communications of the ACM 37 7 1994 .w x18 D. Norman, How might people interact with agents, Commu-

Ž . Ž .nications of the ACM 37 7 1994 .w x19 M.A. Vasarhelyi, The CPASrCCM experiences: prospec-

tives for AIrES research in accounting, in: Proceedings ofthe 1996 Deloitte and ToucherUniversity of Kansas on

Ž .Auditing Problems May 1996 .w x20 Bloomberg Financial Services, http:rrwww.bloomberg.

comrwelcome.html.w x21 S. Madnick, Are We Moving Toward an Information Super-

Highway or a Tower of Babel? The Challenge of Large-Scale

Page 12: Virtual auditing agents: the EDGAR Agent challengeaccounting.rutgers.edu/MiklosVasarhelyi/Resume Articles/MAJOR... · Virtual auditing agents: the EDGAR Agent challenge Kay M. Nelsona,),

( )K.M. Nelson et al.rDecision Support Systems 28 2000 241–253 253

Semantic Heterogeneity, Working paper, Sloan School ofManagement, Massachusetts Institute of Technology, 1997.

w x22 K.M. Nelson, M. Ghods, Technology flexibility: conceptual-ization, validation, and measurement, European Journal of

Ž .Information Systems 1998 December.w x23 K.M. Nelson, M. Ghods, Measuring quality during software

maintenance: an empirical analysis, Decision Support Sys-Ž .tems 1998 October.

w x24 C. Wong, Web Client Programming with Perl: AutomatingTasks on the Web, O’Reilly and Associates, Sebastopol, CA,1997.

w x25 L. Wall, T. Christiansen, R.L. Schwartz, Programming Perl,2nd edn., O’Reilly and Associates, Sebastopol, CA, 1996.

Dr. Kay Nelson is an assistant professorof Information Systems at The Univer-sity of Utah. Dr. Nelson has industryand academic experience both in theU.S. and overseas. She works exten-sively in the area of organizational andinformation technology flexibility, strat-egy, business value, and measurement.Dr. Nelson holds a Ph.D. from the Uni-versity of Texas at Austin in Manage-ment Science and Information Systemswith minors in Technology Management

and Organizational Behavior. Dr. Nelson has published in theareas of software engineering and ISrBusiness partnership inpublications such as MIS Quarterly and Decision Support Sys-tems. Dr. Nelson was the recipient of the International Conference

Ž .on Information Systems ICIS best paper award in 1993 andrecently won another best paper award at the Workshop on

Ž .Information Technologies and Systems WITS .

Dr. Alexander Kogan received the Ph.D.in Computer Science from the USSRAcademy of Sciences, Moscow, and anMS degree in Applied Mathematics andoperations research from the MoscowInstitute of Physics and TechnologyŽ .Phystech , in 1988 and 1984, respec-tively. He is currently an Assistant Pro-fessor of Accounting and InformationSystems with the Faculty of Manage-ment, Rutgers University, Newark. Hisresearch interests are in the areas of

expert systems, artificial intelligence, knowledge-based decisionsupport systems, accounting information systems, accountingproblems of Internet infrastructure and electronic commerce, pro-ductivity accounting, logical analysis of data, Boolean functions,and reasoning under uncertainty. Dr. Kogan has published over 40technical papers in these areas.

Dr. Rajendra P. SriÕastaÕa is Ernst andYoung Distinguished Professor of Ac-counting and Director of the Ernst andYoung Center for Auditing Research andAdvanced Technology at the School ofBusiness, University of Kansas. He holdsa Ph.D. in accounting from the Univer-

Ž .sity of Oklahoma, Norman 1982 and aPh.D. in physics from Oregon State

Ž .University, Corvallis 1972 . ProfessorSrivastava’s publications have appearedin The Accounting Review, Journal of

Accounting Research, Auditing: A Journal of Practice and Theory,and International Journal of Intelligent Systems. He received the1996 Award for Notable Contribution to AI and Expert SystemsResearch in Accounting from the AIrES Section of the AmericanAccounting Association. His current research interests are eviden-tial reasoning in auditing, and on-line continuous auditing.

Dr. Miklos. Vasarhelyi is the director of the Rutgers AccountingResearch Center and was area chairman for three years. Prof.Vasarhelyi visited at the Theseus Institute in France and is ‘‘Pro-fesseur Vacataire’’ at that institution. He was Associate Professorand Director of the Accounting Research Center of the GraduateSchool of Business, Columbia University and Assistant Professorof Accounting at the University of Southern California. Creatorand coordinator of the MBA program at the Catholic University ofRio de Janeiro, as well as Executive Director of the Rio DataCenter. Prof. Vasarhelyi has been associated with the AT and TBell Laboratories since 1985. His work has concentrated in theareas of financial systems architectures, internal audit and expertsystems.

Hai Lu is a doctoral student in Account-ing Information Systems at the Univer-sity of Kansas. He received his MS andPh.D. in Optoelectronic Engineeringfrom Chinese Academy of Sciences. Hiscurrent research interests include the ap-plication of information technology inauditing domain and the informationdisclosure related to the internet. He wasa visiting scientist at CSIRO Division ofTelecommunication and EngineeringPhysics in Australia from 1991–1993

and a visiting scholar at Marshall School of Business in Univer-sity of Southern California in Fall 1998. He has previouslypublished several articles in Review of Scientific Instruments.