Accounting Information Systems outputs
XBRL, AI and in-memory technologies
Introduction
The objective of this chapter is to explore the role of three cutting-edge technologies – XBRL (eXtensible Business Reporting Language; including iXBRL), in-memory computing technology and Artificial Intelligence (AI) – in data analytics for internal decision-making. These technologies are chosen as XBRL solves the issue of data preparation and integration and enhances data usefulness; in-memory computing enables real-time data processing; AI helps extract abstract patterns of data and provides guide for data prediction.
XBRL is a financial reporting markup language that provides the financial community with a standard approach to prepare, publish in a variety of formats, reliably extract and automatically exchange information from financial statements. With XBRL tags, data items on financial statements and related footnotes can be automatically identified and described. More importantly, XBRL can play a role in decision-making. Research conducted by Hodge et al. (2004) indicates that XBRL-enhanced search engines assist investors in acquiring and integrating relevant financial information. In a similar way, the improved data usefulness provided by XBRL facilitates data analytics and provides insight for management and internal auditors. The need for real-time access to all of an entity’s data for improved internal decision-making calls for in-memory computing technology like SAP HANA, which provides more efficient and robust data processing and analysis. Using this technology, entities no longer need to access information stored in a data warehouse. Transactional data is maintained in memory, allowing data analytics to be conducted in real time. The SAP HANA platform, therefore, can be used to conduct real-time data analytics based on huge size of data. Artificial Intelligence provides an intelligence layer for data to efficiently process tedious and complex analytical tasks and has become one of the hottest trends in business. AI capabilities are spreading into nearly all industrial applications, ranging from banking systems that detect attempted credit card fraud, to software agents that track and pay for orders for thousands of goods, to marketing analysis systems that predict sales for products in a specific area.
This chapter explore all three in terms of analytics and internal decision-making. The final part of the chapter proposes the possible applications of Machine Learning and Deep Learning as two important AI technologies to business decision-making.
eXtensible Business Reporting Language (XBRL) and Inline XBRL (iXBRL)
XBRL
To internal business decision-makers, a key to achieve advantageous position against competitors is the availability of insightful data and the efficiency of obtaining the data (Reyes et al., 2007). Although it was originally designed for financial statements of public companies, XBRL tags can be attached to financial documents of varying formats, allowing internal decision-makers to acquire and integrate relevant financial information.
Executives may need to compile information from different departments into a single spreadsheet to analyse the performance of a business strategy. When performing analytical procedures for audit planning, an internal auditor may want to compare the number of unfilled orders to inventory and cost of goods sold, for example. However, the financial information within or without the entity usually has varying forms (i.e., HTML, Excel, Word document, PDF and Edgar filings in plain text) and some forms do not allow their content to be indexed by search engines or other automated analytic tools. Due to the inconsistency of information format among different systems, it is inefficient and vulnerable to errors when collecting/extracting data fields and when preparing it for automated data analysis. As a result, the need for a uniformed and flexible language for business communication with high semantic value has grown (Debreceny et al., 1998) and the potential benefits that XML (eXtensible Markup Language) plays in facilitating the retrieval of rich datasets on the web has been discussed (Debreceny and Gray, 2001).
eXtensible Business Reporting Language (XBRL) developed by the AICPA (and other international accounting bodies) provides a standard method for preparing, analysing and exchanging financial information. It is a standard set of specifications for web-based business reporting. XBRL is based on XML, a markup language that defines the way in which the documents can be encoded to be human-readable and machine-readable. As a global standard for encoding semantic information, XBRL (with XML-based tags attached to the reported financial items) provides financial information users with free and interactively-available data (DataTracks, 2012) as well as structured metadata1 (text, graphic, or video data) at the most primitive level to intelligently search, read, share, exchange, analyse, validate and present (Harris and Morsfield, 2012).
The XBRL instance document is a XML file that contains the values and contexts, such as the measurement units and reporting period of financial reporting data, providing data and structure to enable machine and human readability. Thus, an instance document could be the financial records of a company displayed in an HTML file with various XBRL items embedded. The instance is linked to the XBRL taxonomy. There are two categories of taxonomy, the standard taxonomy and the extension taxonomy. The XBRL standard taxonomy defines financial reporting concepts (also called “elements,” such as “cash and cash equivalent”) used to report (tag) facts about specific financial statements line items. However, as extant research reveals that the standard taxonomy cannot fulfil the needs of companies with specific characteristics for financial reporting (Bonsón et al., 2009), taxonomy extensions were developed, allowing organizations to disclose concepts that are not included in the standard taxonomy by creating elements unique to their own business. The extensibility distinguishes XBRL from most other XML-based standards. XBRL elements consists of schema and a linkbase. While the schema defines underlying taxonomy elements that are being reported, involving element name, data type, balance type and period type, the linkbase combines labels and references to the elements and defines the relationships between those elements as well as relationships between elements and other sources.
Despite those benefits that XBRL brings, concerns remain relative to the quality of the information reported in XBRL, including filing errors and XBRL-based financial statements’ comparability. For instance, it was reported that one-quarter of all SEC-XBRL filings under rule 33-9002 up to September 1, 2009 contained calculation errors (Debreceny et al., 2010). In July 2014, the SEC highlighted the need to eliminate calculation errors by sending a letter to alert CFOs of missing calculation relationships in the XBRL files, asking them to “take necessary steps to ensure that you are including all required calculation relationships”.2 Software systems were thus developed to identify XBRL errors. For example, a Consistency Suite tool that covers more than 30 categories of XBRL filing errors is provided by XBRL.US to examine problems in XBRL files before submission. In addition, some errors cannot be identified by a software. For instance, the incorrect positive or negative sign of the data value, the use of an old extension element, the tag selected represents a different meaning than what is disclosed in the traditional paper-based document, etc. The detection of such errors relies on the expertise of the XBRL individual (Financial Execution Research Foundation, 2014).
The flexibility provided by an XBRL extension taxonomy results to a compatibility issue (Cohen, 2004; Zhu and Wu, 2011; Chen and Sun, 2011). It is reported that almost two-thirds of the elements in XBRL financial reports are not comparable with those of similar companies (Zhu and Wu, 2011). Researchers pointed out that the preparing process of XBRL-tagged information should be cost-effective and XBRL extensions make automated comparisons across companies and industries inefficient as it may need human intervention to understand the elements in individual XBRL disclosures (Debreceny et al., 2011). There are discussions on the importance of how to balance the trade-off between the flexibility and the issue of comparability, costs and reliability (Cohen, 2004; Piechocki et al., 2009). Since 40% of XBRL extensions are found unnecessary (Debreceny et al., 2011), regulators need to carefully address XBRL taxonomy extensions to guide filers in reducing the number of extensions in financial statements.
The information integrity within XBRL-based documents is another critical problem (Perdana et al., 2014). No and Boritz (2004) suggested the concept of XBRL assurance to ensure the information integrity by implementing assertions to comply with the XBRL taxonomy and guidance. Boritz and No (2008) observed the lack of assurance of XBRL documents in the SEC’s voluntary XBRL filing program and to address this issue, they implemented mock assurance as a surrogate to provide assurance on XBRL filing for United Technologies Corporation (Boritz and No, 2009). Srivastava and Kogan (2010) proposed a set of assertions, consisting of statements for ensuring XBRL assurance and assurance processes. Plumlee and Plumlee (2008) addressed three issues: assurance guidance, XBRL-tagging validation and audit risks, for XBRL assurance.
iXBRL
Since 2009, the SEC has required listed companies to report two versions of financial information – a plain-text document and a XBRL document – due to the difficulties of the human readability in XBRL documents. A XBRL document often presents poorly when converted to a human viewable format, as the standard rendering solutions are unable to effectively replicate the presentation layout and formatting of traditional paper-based financial report. The problem becomes even worse when the published taxonomies need to be extended to include facts that were not predefined in the company’s extension taxonomy. Currently, the audited or reviewed financial filings (e.g., 10-K and 10-Q) are stored in HTML format in the “Document Format File” section of the EDGAR database for human consumption, while the unaudited XBRL-based documents are stored in the “Data Files” section of the EDGAR database and for consumption by software. This separate reporting leads to unnecessary duplication of work for filers (Basoglu and White, 2015) as well as extensive training and detailed technical knowledge required of financial statement users (Harris and Morsfield, 2012). As a result, researchers argue that the current implementation of XBRL-based financial reporting is not as useful and transparent to analysts and investors as it was expected (Basoglu and White, 2015).
Fortunately, Inline XBRL (iXBRL), a recent development of XBRL-based reporting, tackles this problem. iXBRL embeds XBRL metadata within an HTML document to integrate both human- and machine-readable instance documents (Cohen et al., 2014). Unlike the XBRL-based document that can be viewed only with specialized viewers and focuses on automated machine readability of data, the iXBRL document can be viewed on standard browsers and emphasizes data rendering (DataTracks, 2012). The preparer of an iXBRL document is able to control the presentation format of the document to make it look like a printed financial report or an interactive website. For example, the preparer is able to make the text bold or italicized, and the order and alignment of facts in a table can also be controlled by the preparer. As a result, preparing financial documents with iXBRL could benefit the user by making the data much more readable for humans. Furthermore, the implementation of iXBRL opens new topics of future research, such as the validation of the potential benefits and values that iXBRL could bring, as well as the assurance of the integrity of iXBRL instance document (Basoglu and White, 2015).
Data analytics and internal decision-making
By improving information relevance, representation, comparability, consistency and understandability, XBRL3 enhances the usefulness of financial information and more disaggregated data (Vasarhelyi et al., 2012) and facilitates data analytics, helping entities investigate past performance, optimize business processes and gain insight for internal decision-making. On the other hand, thanks to the frequent use of GPS-enabled devices, wireless sensors and RFIDs, business is in a real-time connected world where the real-time data captured by the embedded sensors communicates to data centre networks for processing. In such a digital world, business growth is sustained on a real time (or near real time) basis by speed decisions made from complete information, with insights based on real-time data analytics (Russom, 2013).
The real-time data analytics refers to data that is able to capture, process and analyse the instant it flows, or streams, into the system (Pittman, 2013). The benefit of real-time data analytics is obvious: the manager is able to make informed business decisions rapidly and with more complete information. After shortening the response time for data analysis, a company can take prompt strategy to prevent customers from churning. Based on the responses of 2,106 senior executives, PwC’s Global Data and Analytics Survey of 2016 indicates that by 2020 executives expect more data-driven, faster and more sophisticated decision-making with a good mix of human judgment and machine algorithms, rather than merely rely on intuition and experience (PwC, 2016).
Real-time analysis of large volumes of data: modern in-memory database technology
Given today’s exponential growth of real-time data collected by sensors during business operations, speed is crucial for a company to succeed. Executives have to make timely decisions based on the real-time data. To maximize the profit of the entity, the manager needs to obtain competitive products or services sooner, exploit market opportunities faster, identify business risk and threats more rapidly and manage the customer lifecycle at every touchpoint more efficiently (Hunley and Foley, working paper). The internal auditor needs a real-time monitoring of the accounting information system and detects anomalies more promptly. Therefore, the need for deeper, mobile and real-time access to all the data of the entity for improved internal decision-making at an operational, tactical and strategic level call for more efficient and robust data management infrastructures as opposed to traditional architectures of database systems, where it relies on hardware refreshes and improved memory caching to support complex data management tasks.
With the ability of replicate, store and perform analysis on real-time data, in-memory computing technology has great potential to enable real-time decision-making on digitized business operations. Businesses formerly relied on traditional relational database management systems (RDBMS) with hard drive based memory to store our data. With the new in-memory computing technology we are able to store the data in “connected” main memory (RAM) across a cluster of computers and process it in parallel. This technology performs roughly 5,000 times faster than the traditional way.
SAP HANA4 serves as an outstanding example of modern in-memory computing technology and presents itself as a first step towards a holistic data management platform supporting high performance analytics applications at unprecedented speed. It is fully in-memory, which means it keeps data in a server’s RAM, with both row stores and column stores supporting transactional (OLTP5) and multidimensional (OLAP6) workloads. Its OLAP capabilities are virtual, having no requirement on caches, aggregation, indexes or physical cubes. It is unnecessary to duplicate the reporting process by posting data to a separate system. Data analytics are conducted simultaneously when AIS transactions are performed, providing instant feedback on AIS transactions and data (Smith, 2016). SAP HANA and other in-memory computing techniques are driven by technology innovations including (1) high-speed processing due to hardware advances and increasingly affordable memory, (2) increased amount of memory as a result of the mainstream availability of 64-bit processors, (3) the popularity of multicore processor (SAP, 2013).
Since SAP HANA has evolved into a platform in 2013, organizations are able to build and deploy a growing number of on-demand applications in areas of customer engagement, finance, human resources, manufacturing, procurement, logistics, IT, etc., and perform retrospective and predictive analysis and deliver real-time insights from both transactional and analytical processing. As data is stored and analysed in local memory, the latency of data transferring and loading is eliminated, which leads to more prompt reactions to anomalies and irregularities. For example, the SAP Fraud Management analytic application powered by SAP HANA processes large volumes of real-time data, helps analysts detect, prevent and deter fraud and provides continuous monitoring on fraud. SAP HANA platform also enables millions of fans around the world to visit NBA’s statistics website (www.NBA.com/Stats) to instantly access and interact with the NBA’s entire history of official statistics. Premium automotive tuning company Mercedes-AMG GmbH employs the SAP HANA platform to support real-time analysis of data from five different engine tests, enabling ad hoc analyses and simulations at any time. Whenever a potential engine issue is detected, the system reports the corresponding patterns, enabling AMG to immediately terminate expensive tests and start analysing the situation.
In the academic arena, researchers analysed the capabilities of SAP HANA for analytical processing. Rudny et al. (2014) examine the analytical possibilities of SAP HANA by applying it to forecast the energy demand in the energy sector based on a number of experiments, and demonstrated its effectiveness in time series forecasting as compared to classical solutions. Bracher et al. (2015) apply SAP HANA to marketing by scoring churn risk within a telecommunication industry. Keshava et al. (2015) focus on audience discovery and targeting and identify prospective clients for sales agents based on data from various sources (such as market trends, blogs, news, marketing catalogues) and demonstrate the superior text analysis capability of SAP HANA.
The application of AI to internal decision-making
AI is “the effort to make computers think” (Haugeland, 1989). In other words, AI is a machine which uses cutting-edge techniques to competently perform or mimic “cognitive” functions that are usually performed by human minds, such as “learning” and “problem solving” (Norvig and Russell, 2009). Internal business decision-makers have an urgent demand for insights from real-world big data to assist them in obtaining better knowledge of the operational and financial situation and to serve their customers. Nevertheless, on one hand, management and internal auditors have to deal with mountains of data, forcing them to employ data analytical tools to process and analyse it. On the other hand, they have to rely on human experts to identify the features of the real-world big data as most of the data is semi-structured (e.g., text) or unstructured (e.g., video or audio). Fortunately, current advances in AI may help them solve this dilemma and support faster and better decision-making (MIT Technology Review, 2017). In this section, two key AI technologies, Machine Learning and Deep Learning, are discussed. While Machine Learning assists us in predicting patterns by the model trained with historical data, Deep Learning focus on information extraction of big data.
Machine Learning
Machine Learning is defined as a “field of study that gives computers the ability to learn without being explicitly programmed” (Samuel, 1959). It evolved from the research of pattern recognition and computational learning theory in AI. Machine Learning attempts to make predictions through the construction of algorithms that learn from data. The algorithms are generated by developing a model from a subset of data called “training set” and testing the model with the “test set,” another subset of data. By recognizing complex patterns of data, Machine Learning enables cognitive systems to automatically learn, make intelligent decisions and improve itself through interactions with data, devices and people (Lakshmi and Radha, 2011). Machine Learning has been prevalently applied to data analysis in fields of finance, marketing, telecoms, taxation, litigation, insurance and web analysis.
To date, Machine Learning has been widely adopted to predict events, especially in the field of finance. Examples include forecasting bankruptcies (Wilson and Sharda, 1994; Ben Jabeur and Fahmi, 2014), defaulting loans (Messier and Hansen, 1988; Moin and Ahmed, 2012), stock prices (Barr and Mani, 1994; Huang and Lin, 2014), interest rate (Kanevski and Timonin, 2010) and credit ratings (Buta, 1994; Hájek, 2011). More importantly, Machine Learning has been incorporated into fraud detection, which is of special interest to auditors. As financial fraud perpetrators attempt to conceal fraud and financial fraud often involves collusion, it is difficult to discover these types of fraud tactics merely by employing traditional analytical modelling methods. Various Machine Learning algorithms were applied to solve the fraud detection problem. Lin et al. (2003) evaluate an integrated fuzzy neural network for financial fraud detection and conclude that it outperforms most statistical models and prior artificial neural networks. Whiting et al. (2012) explore the role of statistical learning and data mining in detecting financial fraud, with the goal of advancing fraud discovery performance and proactively detecting or even mitigating financial fraud. They establish four different models, including Probit, Logit, partially adaptive and ensemble models and compare their predictability. Using endogenous financial data through the evaluation of analytical procedure expectations, an early study conducted by Green and Choi (1997) develops a neural network fraud classification model to detect the risk of management fraud. Bell and Carcello (2000) propose a logistic regression model for estimating the likelihood of fraudulent financial reporting for an audit client. Cecchini et al. (2010) propose a methodology for detecting management fraud based on support vector machines. The results show that the support vector machine based methodology is a useful method for discriminating between fraudulent and non-fraudulent companies. Machine Learning applies pattern matching and deviations or variations detection to evaluate risks (Ramamoorti et al., 1999). Issa and Kogan (2014) propose a methodology for reviewing the assessment of internal controls risk by internal auditors and business owners. This methodology is based on developing an ordered logistic regression model that employs historic data on internal controls risk assessments. The Machine Learning applications that have been addressed in academia could be used to leverage internal decision-making in several areas, which are now outlined.
Financial planning and budgeting
Business planning, especially financial budgeting, is critical for a company. Although modern budgeting software solutions such as ACCPACCFO, BIZBENCH and iLumen can assist management to create budgets without the hassle of managing error-prone spreadsheets, these products provide limited analytical capability in discovering potential unviable budgets (Yip, 2012).7 Chief executives are responsible for the overall process of a company’s financial planning and budgeting, determining the effectiveness of compliance with corporate policies and procedures and ensuring the financial planning and budgeting process operates as planned. Facing the problem of how to determine the accuracy of budget, Machine Learning may be a good approximation approach. Previous research shows that certain financial variables can indicate the future performance and sustainability of a business. Examples of these variables include cash flow, profit margin, net profit and return on equity, etc. They can be used to predict future events of a business in a particular industry (Nijhuis and Westerhuis, 2013) and, therefore, are associated with the company’s budget. As a result, by investigating a company’s historical variation between the forecast and actual data on these budget-related financial variables, internal auditors are able to uncover the characterized change pattern of them and use this knowledge to label every historical forecast as a viable or unviable budget8 (Yip, 2012). Machine Learning techniques, by analysing the historical data, could effectively help recognize the existence of pattern change and establish a classification model for the accuracy inference of the company’s business budgeting and planning. This model is designed to alert management for budget inaccuracy, so that a company can adjust its method and policy of financial planning and budgeting.
Operational process monitoring, control and diagnosis
The current economic climate encourages cost-cutting activities, increases risk exposure and fosters organizational changes. Organizations are implementing Continuous Monitoring, a feedback mechanism used by management, to ensure that controls are operated as designed and transactions are processed as prescribed (Vasarhelyi et al., 2010). To enable continuous monitoring, an organization’s operational process should be controlled and the data from the process should be analysed to obtain insights on the patterns underlying the data. A possible solution to this could be the combination of a Machine Learning technique (such as Artificial Neural Network) as an expert system, which is a knowledge-based AI technology (Uraikul et al., 2007). Since Machine Learning can automatically extract knowledge (e.g., the pattern of data) from data, symbolic information can be integrated into an artificial neural network learning algorithm (Kasabov, 1996), rules can be generated (this is the bottleneck of Machine Learning in knowledge acquisition application) (Uraikul et al., 2007), knowledge modelled and extraction achieved.
Retrieving information from weblogs
Weblogs providing commentary on businesses and companies have rapidly gained in popularity over the past few years. Information from weblogs can be an important and more objective source of data. The problem regarding how to effectively search and collect related data from countless weblogs opens up new opportunities for developing blog-specific search and mining techniques. A study by Chen et al. (2008) proposes probabilistic models for blog search and mining using two Machine Learning techniques, latent semantic analysis (LSA) and probabilistic latent semantic analysis (PLSA). The former applies Machine Learning to business blogs. It demonstrates that the proposed model can present the blogosphere in terms of topics with measurable keywords, hence tracking popular conversations and topics in the blogosphere. Potential applications of this stream of research may include retrieving data from other sources, such as newspapers and analysts’ reports. In addition, since this study focuses on information retrieving, more advanced Machine Learning techniques can be used to design a system to automatically monitor and identify trends in data from these sources.
Deep Learning
The value of Deep Learning in internal decision-making primarily lies in the fact that it is able to extract highly abstract data features from unstructured or semi-structured data without human intervention, providing a great volume of analysable, real-world big data to support decision-making. Deep Learning (also called Deep Neural Network) is a new area of Machine Learning invented in 2006 by Hinton et al. (2006). Its objective is to represent input data and generalize the learned patterns for the future use (Hinton et al., 2006; Najafabadi, 2015). Instead of teaching machines what to do, Deep Learning technology allows them to learn how to do it for themselves based on the data provided and ultimately they will tell us what to do (Maycotte, 2014). In simpler terms, Deep Learning is about learning multiple levels of representation and abstraction, which helps acquire the meaning from data such as images, voice and text. Both Deep Neural Network and Artificial Neural Network contain an input layer, hidden layer and output layer, whereas Deep Neural Network has multiple hidden layers and Artificial Neural Network only has one or two hidden layers.
During the past ten years, we have observed numerous successes of Deep Learning in diverse applications of image feature coding, natural language processing, handwriting recognition, audio processing, information retrieval and multitask learning and robotics (Deng and Yu, 2013; Mezghani et al., 2010; Xiong and Zhao, 2014). The successful applications of Deep Learning to voice, image and text processing and analysing benefit the internal decision-making of entities by providing valuable insights from additional sources of data. Speech recognition is one of the first successful applications of Deep Learning methods at an industrial scale. It is the process of translating spoken words into text, which is challenging due to the high viability in speech signals, such as accents, dialects, pronunciations, emotion and speech styles. The presence of environmental noise, reverberation, different microphones and recording devices results in additional variability. For a machine to correctly recognize speech, it needs cognitive computing – a system with architecture that imitates how the human brain understands information. Deep Learning aims to automatically identify features from raw input data typically through analysing the primitive spectral or possibly waveform features (Deng et al., 2013). With speech recognition tools boosted by Deep Learning, such as IBM Watson, at hand, conversations and speech from a company’s call centre, management meetings and other sources could be identified and translated automatically and the result can be used to establish trends or clusters of opinion with similar characteristics.
Empowered by deep conventional neural networks (CNN), a classic Deep Learning model, along with the mainstream availability of prodigious data sets, image recognition is gaining momentum. Image recognition involves two core tasks: image classification and object detection. In the “image classification” task, the system is taught to recognize object categories, such as “book,” “tree,” or “car,” while in the “object detection” task, precise position of the object in the image should be identified (He et al., 2014). A deep neural network, in this way, recognizes visual patterns of objects. It begins with the input layer, where the images are input to the model, followed by the first hidden layer where the very essential components such as pixels are identified. As the layer inside the deep neural network proceeds further, it recognizes more advanced and abstract features of the image, such as edges and then shapes. Each successive layer in a neural network uses features in the previous layer to learn more complex features. Each hidden layer going further into the network is a weighted, non-linear combination of the layers in the lower level. The entire Deep Learning process is about refining the weights representing what was learned during unsupervised training. In 2013, Google announced its visual search engines. The basic idea of this search engine is to “use the visual content of an image to generate searchable tags for photos combined with other like text tags and EXIF metadata to enable search across thousands of concepts like a flower, food, car, jet ski, or turtle” (Rosenberg, 2013).
One possible source of data for business analysis is the video captured from surveillance cameras. A face detection system based on image recognition technology can conduct a serious of tasks – face detection, parsing, verification and face attribute recognition. Such a system is helpful for gathering evidence for fraud detection as investigators could easily identify the perpetrator or search for a particular individual’s activities from sequential digital images of a CCTV system. In addition, entities may use electronic commerce or image processing systems to scan and convert source documents (e.g., purchase orders, bills of lading, invoices and checks) into electronic images to facilitate storage and reference.
Text is another source of data. Examples include emails, conference calls, newspaper articles, conversations and comments like Tweets and Facebook posts, product reviews from Amazon and so on. Such types of data contain useful information that could provide insights for business decision-making in customer targeting and segmentation and other areas. Since the vast majority of the text is semi-structured or un-structured, it is difficult to extract the feature with regular text mining or Machine Learning techniques (i.e., traditional artificial neural networks). Deep Learning is a solution to unlock the potential of text data. With a deep neural network, text data can be easily analysed and the feature underlying can be extracted without human intervention, thanks to its multiple hidden layers where huge numbers of neurons are connected with each other based on complex mathematical calculation. The deep neural network is capable to identify “who,” “what,” “when,” “where,” “why” and the sentiment from the text data and transforms the unstructured text into structured data with identified features, which can then be incorporated into the organization’s existing data analytics applications.
Summary
This chapter discusses three technologies that facilitate internal decision-making in data analytics: XBRL, in-memory computing and Artificial Intelligence. By providing a standard business communication language, XBRL improves the effectiveness and efficiency of data preparing, analyzing and exchanging from incompatible sources. In-memory computing enables real-time analysis of large volumes of data to support timely decision-making. Lastly, AI solves the dilemma of big data processing that, on one hand, decision-makers have had to deal with mountains of data, forcing them to employ data analytical tools; on the other hand, it has to rely on human experts to identify the features of real-world big data as most of the data is semi-structured (e.g., text) or unstructured (e.g., video or audio). In addition to introducing the basic idea of the three technologies, this chapter analyzes their benefits and shortcomings and explores the current (or possible) applications in different aspects of internal decision-making.
Notes
1 The metadata defines the reported terms and the relationships between the terms.
2 See the sample letter at www.sec.gov/divisions/corpfin/guidance/xbrl-calculation-0714.htm
3 It also includes iXBRL and XBRL GL.
4 HANA stands for High-Performance Analytic Appliance.
5 OLTP stands for Online Transaction Processing.
6 OLAP stands for Online Analytical Processing.
7 For methodology details see Yip, 2012.
8 For example, label “0” is given if the actual value of a test variable is worse than the predicted one, indicating an inaccurate forecast of budget. A label “1” is given if the actual value of a test variable is better than the predicted one, indicating an accurate forecast of budget.
References
Barr, D. S. and Mani, G. (1994). Using neural nets to manage investments. AI Expert, 9(2), 16–21.
Basoglu, K. A. and White Jr., C. E. (2015). Inline XBRL versus XBRL for SEC reporting. Journal of Emerging Technologies in Accounting, 12(1), 189–199.
Bell, T. B. and Carcello, J. V. (2000). A decision aid for assessing the likelihood of fraudulent financial reporting. Auditing: A Journal of Practice & Theory, 19(1), 169–184.
Ben Jabeur, S. and Fahmi, Y. (2014). Predicting business failure using data-mining methods. Working paper. Department of Research, Ipag Business School.
Bonsón, E., Cortijo, V. and Escobar, T. (2009). Towards the global adoption of XBRL using International Financial Reporting Standards (IFRS). International Journal of Accounting Information Systems, 10(1), 46–60.
Boritz, J. E. and No, W. G. (2008). The SEC’s XBRL voluntary filing program on EDGAR: a case for quality assurance. Current Issues in Auditing, 2(2), A36–A50.
Boritz, J. E. and No, W. G. (2009). Assurance on XBRL-related documents: the case of United Technologies Corporation. Journal of Information Systems, 23(2), 49–78.
Bracher, S., Holmes, M., Mischewski, L., Islam, A., McClenaghan, M. Ricketts, D., Neuber, G., Jeung, H. and Vijayarajendran, P. (2015). Advanced analytics on SAP HANA: churn risk scoring using call network analysis. 2015 IEEE 31st International Conference on Data Engineering, 1400–1403.
Buta, P. (1994). Mining for financial knowledge with CBR. AI Expert, 9(2), 34–41.
Cecchini, M., Aytug, H., Koehler, G. J. and Pathak, P. (2010). Detecting management fraud in public companies. Management Science, 56(7), 1146–1160.
Chen, Y., Tsai, F. S. and Chan, K. L. (2008). Machine learning techniques for business blog search and mining. Expert Systems with Applications, 35(3), 581–590.
Chen, H. and Sun, W. (2011). Study on rational application of eXtensible Business Reporting Language. Journal of Software, 6(2), 257–264.
Cohen, E. E. (2004). CAP Forum on E-business: compromise or customize: XBRL’s paradoxical power. Canadian Accounting Perspectives, 3(2), 187–206.
Cohen, E. E., Debreceny, R., Farewell, S. and Roohani, S. (2014). Issues with the communication and integrity of audit reports when financial reporting shifts to an information-centric paradigm. International Journal of Accounting Information Systems, 15(4), 400–422.
DataTracks (2012, March 20). What is IXBRL? Differences between XBRL and IXBRL. Retrieved from www.datatracks.co.uk/ixbrl-blog/what-is-ixbrl/.
Debreceny, R., Farewell, S., Piechocki, M., Felden, C. and Graning, A. (2010). Does it add up? Early evidence on the data quality of XBRL filings to the SEC. Journal of Accounting and Public Policy, 29(3), 296–306.
Debreceny, R. and Gray, G. L. (2001). The production and use of semantically rich accounting reports on the internet: XML and XBRL. International Journal of Accounting Information Systems, 2(1), 47–74.
Debreceny, R., Gray, G. and Barry, T. (1998). Accounting information in a networked world-resource discovery, processing and analysis. American Accounting Association Annual Meeting, New Orleans, Louisiana.
Debreceny, R. S., Farewell, S. M., Pechocki, M., Felden, C., Graning, A. and d’Eri, A. (2011). Flex or break? Extensions in XBRL disclosures to the SEC. Accounting Horizons, 25(4), 631–657.
Deng, L., Li, J., Huang, J. T., Yao, K., Yu, D., Seide, F., Seltzer, M., Zweig, G., He, X., Williams, J., Gong, Y. and Acero, A. (2013). Recent advances in deep learning for speech research at Microsoft. 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, 8604–8608.
Deng, L. and Yu, D. (2013). Deep learning: methods and applications: foundations and trends. Signal Processing, 7(3–4), 197–387.
Financial Execution Research Foundation (FERF). (2014). Beware XBRL errors in SEC filings – including those not detectable by software. Retrieved from www.financialexecutives.org/ferf/download/2014%20Final/2014-023.pdf.
Green, B. P. and Choi, J. H. (1997). Assessing the risk of management fraud through neural network technology. Auditing, 16(1), 14.
Hájek, P. (2011). Municipal credit rating modelling by neural networks. Decision Support Systems, 51(1), 108–118.
Harris, T. S. and Morsfield, S. G. (2012). An evaluation of the current state and future of XBRL and interactive data for investors and analysts. Columbia University Academic Commons. Retrieved from http://hdl.handle.net/10022/AC:P:20283.
Haugeland, J. (1989). Artificial Intelligence: The Very Idea. Cambridge, MA: MIT Press.
He, K., Zhang, X., Ren, S. and Sun, J. (2014). Spatial pyramid pooling in deep convolutional networks for visual recognition. European Conference on Computer Vision, 346–361.
Hinton, G. E., Osindero, S. and Teh, Y. W. (2006). A fast learning algorithm for deep belief nets. Neural Computation, 18(7), 1527–1554.
Hodge, F. D., Kennedy, J. J. and Maines, L. A. (2004). Does search-facilitating technology improve the transparency of financial reporting? The Accounting Review, 79(3), 687–703.
Huang, C. Y. and Lin, P. K. (2014). Application of integrated data mining techniques in stock market forecasting. Cogent Economics & Finance, 2(1), 929505.
Hunley, E. J. and Foley, R. (working paper). Delivering Intelligence for Real-Time Decision Making. https://support.sas.com/rnd/papers/sugi31/10031.pdf.
Issa, H. and Kogan, A. (2014). A predictive ordered logistic regression model as a tool for quality review of control risk assessments. Journal of Information Systems, 28(2), 209–229.
Kanevski, M. F. and Timonin, V. (2010). Machine learning analysis and modeling of interest rate curves. European Symposium on Artificial Neural Networks – Computational Intelligence and Machine Learning, 47–52.
Kasabov, N. K. (1996). Foundations of Neural Networks, Fuzzy Systems and Knowledge Engineering. Cambridge, MA: MIT Press.
Keshava, S., Kiran, P. and Nithin, S. (2015). Audience discovery and targeted marketing using SAP HANA. Advance Computing Conference (IACC), 2015 IEEE International, 749–753.
Lakshmi, R. D. and Radha, N. (2011). Machine learning approach for taxation analysis using classification techniques. Machine Learning, 12(10).
Lin, J. W., Hwang, M. I. and Becker, J. D. (2003). A fuzzy neural network for assessing the risk of fraudulent financial reporting. Managerial Auditing Journal, 18(8), 657–665.
Maycotte, H. O. (2014, December 16). Why big data and AI need each 0ther – and you need them both, Forbes. Retrieved from www.forbes.com/sites/homaycotte/2014/12/16/why-big-data-and-ai-need-each-other-and-you-need-them-both/#5577062458dc.
Messier Jr., W. F. and Hansen, J. V. (1988). Inducing rules for expert system development: an example using default and bankruptcy data. Management Science, 34(12), 1403–1415.
Mezghani, D. B. A., Boujelbene, S. Z. and Ellouze, N. (2010). Evaluation of SVM kernels and conventional machine learning algorithms for speaker identification. International Journal of Hybrid Information Technology, 3(3), 23–34.
MIT Technology Review (n.d.). 10 Breakthrough Technologies 2017. https://www.technologyreview.com/lists/technologies/2017/.
Moin, K. I. and Ahmed, D. Q. B. (2012). Use of data mining in banking. International Journal of Engineering Research and Applications, 2(2), 738–742.
Najafabadi, M. M. (2015). Deep learning applications and challenges in big data analytics. Journal of Big Data, 2(1), 1.
Nijhuis, S. and Westerhuis, J. (2013). What factors determine long-term solid financial performance? Performance, 5(1): 54–61.
No, W. G. and Boritz, J. E. (2004). Assurance reporting for XML-based information services: XARL (eXtensible Assurance Reporting Language). Canadian Accounting Perspectives, 3(2), 207–233.
Norvig, P. and Russell, S. J. (2009). Artificial Intelligence: A Modern Approach. Upper Saddle River, NJ: Prentice Hall.
Perdana, A., Robb, A. and Rohde, F. (2014). An integrative review and synthesis of XBRL research in academic journals. Journal of Information Systems, 29(1): 115–153.
Piechocki, M., Felden, C., Graning, A., and Debreceny, R. (2009). Design and standardisation of XBRL solutions for governance and transparency. International Journal of Disclosure and Governance, 6(3), 224–240.
Pittman, D. (2013, December 16). Learn about real-time analytics through real-world examples. IBM Big Data & Analytics Hub. Retrieved from www.ibmbigdatahub.com/blog/learn-about-real-time-analytics-through-real-world-examples.
Plumlee, R. D. and Plumlee, M. A. (2008). Assurance on XBRL for financial reporting. Accounting Horizons, 22(3), 353–368.
PwC (2016). PwC’s Global Data and Analytics Survey 2016: Big Decisions. Retrieved from www.pwc.com/us/en/advisory-services/data-possibilities/big-decision-survey.html.
Ramamoorti, S., Bailey, J. and Traver, R. O. (1999). Risk assessment in internal auditing: a neural network approach. International Journal of Intelligent Systems in Accounting, Finance & Management, 8(3), 159–180.
Reyes, E., Rodríguez, D. and Dolado, J. (2007). Overview of XBRL technologies for decision making in Accounting Information Systems. Journal of Information Systems, 19(4), 19–41.
Rosenberg, C. (2013). Improving photo search: a step across the semantic gap. Google Research Blog. Retrieved from https://research.googleblog.com/2013/06/improving-photo-search-step-across.html.
Rudny, T., Kaczmarek, M. and Abramowicz, W. (2014). Analytical possibilities of SAP HANA – on the example of energy consumption forecasting. Advances in Systems Science, 141–150.
Russom, P. (2013). Operational intelligence: real-time business analytics from big data. TDWI Checklist Report, 1–8.
Samuel, A. L. (1959). Some studies in machine learning using the game of checkers. IBM Journal of Research and Development, 3(3), 210–229.
SAP (2013). Detect, Prevent and Deter Fraud in Big Data Environments. Retrieved from www.sap.com/bin/sapcom/en_us/downloadasset.2013-09-sep-17-10.detect-prevent-and-deter-fraud-in-big-data-environments-pdf.html.
Smith, J. (2016). Accounting Information Systems: ethics, fraudulent behavior and preventative measures. University Honors Program Thesis. Georgia Southern University. Retrieved from http://digitalcommons.georgiasouthern.edu/cgi/viewcontent.cgi?article=1188&context=honors-theses.
Srivastava, R. P. and Kogan, A. (2010). Assurance on XBRL instance document: A conceptual framework of assertions. International Journal of Accounting Information Systems, 11(3), 261–273.
Uraikul, V., Chan, C. W. and Tontiwachwuthikul, P. (2007). Artificial intelligence for monitoring and supervisory control of process systems. Engineering Applications of Artificial Intelligence, 20(2), 115–131.
Vasarhelyi, M. A., Chan, D. Y. and Krahel, J. P. (2012). Consequences of XBRL standardization on financial statement data. Journal of Information Systems, 26(1), 155–167.
Vasarhelyi, M. A., Alles, M. and Williams, K. T. (2010). Continuous Assurance for the Now Economy: A Thought Leadership Paper for the Institute of Chartered Accountants in Australia. Queensland, Australia: Institute of Chartered Accountants.
Whiting, D. G., Hansen, J. V., McDonald, J. B., Albrecht, C. and Albrecht, W. S. (2012). Machine learning methods for detecting patterns of management fraud. Computational Intelligence, 28(4), 505–527.
Wilson, R. L. and Sharda, R. (1994). Bankruptcy prediction using neural networks. Decision Support Systems, 11(5), 545–557.
Xiong, H. Y. and Zhao, J. (2014). An image retrieval method based on Machine Learning and SVM. Applied Mechanics and Materials, Trans Tech Publications, 631, 474–477.
Yip, K. L. (2012). Determining the accuracy of budgets: a machine learning application for budget change pattern recognition. Master’s thesis. Unitec Institute of Technology.
Zhu, H. and Wu, H. (2011). Interoperability of XBRL financial statements in the US. Mobile Applications and Knowledge Advancements in E-Business, 129.