LEARNING OBJECTIVES

After completing this chapter, you should be able to do the following:

•     Distinguish among Big Data concepts that apply to accounting operations.

•     Identify how data analysis can be used in the accounting department.

INTRODUCTION

Now that you have a good understanding of Big Data, how should it be applied specifically to accounting concepts? Accountants are familiar with accessing and manipulating structured accounting data, such as the following:

•     Account name

•     Account general ledger code

•     Transaction amount

•     Vendor name

•     SKU number

Accountants are generally not familiar with unstructured data that are contained in memo fields, miscellaneous fields not accessed by traditional reports or databases, or systems (like email) that are built with the data in an unstructured format.

Nor are accountants familiar with streaming data in all of its forms nor in retrieving the data for analysis purposes. Examples of streaming data include social media discussions and machine sensor data. Some accountants may doubt that unstructured or streaming data will have any value. However, in the expanded role of the financial function, the concept of adding value to the organization is beyond traditional financial roles. This is best illustrated with the following image:

Source: Lindell, James, Controller as Business Manager. (AICPA, 2014).

Structured data exists at the detail or commodity level, business strategy, and industry strategy level. The industry strategy level data will most likely have to be found, downloaded, massaged and analyzed. It is possible that structured data will also exist at the broadest trend level although it will need to be manipulated in the same fashion as industry-level data.

Unstructured and streaming data can exist at all levels. It is incumbent on accountants to determine the following:

•     What data are available?

•     How can it be accessed?

•     What tools are necessary to analyze and transform the data into useful information?

The interesting point to keep in mind is that the data that are not being accessed in structured, unstructured and streaming arenas have the potential to add significantly to the value created in the accounting department.

This chapter will illustrate methods for applying Big Data approaches to daily accounting functions as well as ways to use Big Data and data analysis for operational teams.

BIG DATA FOR THE CFO

According to Infochimps, a subsidiary of IT consulting giant Computer Sciences Corporation, Big Data is essential for today’s enterprise financial strategies.

In past practices, control or compliance issues, treasury issues, and financial reporting were the primary focuses of financial executives. "Today, CFOs are expected to be strategic business partners with operating units and close confidants to the CEO."¹ Today’s CFO will be involved with strategy, board issues, revenue development, cost control, and profitability, among many other tasks. On many occasions, CFOs and controllers will be required to assume additional duties in the areas of legal, human resources, and information technology. It is no surprise that the CFO’s job description has changed more than any other corporate executive position.

Big Data is applicable to enterprise financial strategies more than any other innovation seen by the CFO in recent years. Big Data empowers the CFO to access and understand areas that have previously been inaccessible. This will result in the CFO getting ahead of the business’ needs. Some samples of how Big Data can help the CFO include the following:

•     Planning and forecasting with information-driven planning, rolling forecasts and multi-year plans overall dimensions of the business.

•     Minimizing risk and fraud by constantly filtering through the points of interest of each occasion and exchange looking for the frequently inconspicuous marks of extortion or some different business risk factors.

•     Advanced financial and management analytics for monthly books, statutory reporting, and variance analysis.

•     Profitability modeling and optimization with cutting-edge cost analysis, product or customer viability, and allocations.

•     Financial system administration with comprehensive score-carding using innovative data visualization.

•     Previously impossible financial, ratio, and related information analysis that can lead to new insights, applications, and enhanced company profitability and value.

KNOWLEDGE CHECK

1.     Based on the text, which statement is most accurate?

a.     Big Data is second in importance to an established culture.

b.     Big Data is second in importance only to lean management techniques.

c.     Big Data is applicable to enterprise financial strategies more than any other innovation seen by the CFO in recent years.

d.     Big Data will play an important role in seeking out wasteful practices within the organization. Interestingly, the CFO, CIO, and CEO can cooperate to improve corporate performance with the ability to implement Big Data arrangements rapidly and without the expenses of employing new groups of developers.

BIG DATA AREAS OF FOCUS

In the previous chapter, we looked a whole host of applications for Big Data from traffic flow management to baking better bread. The opportunities to apply Big Data analytics to accounting are similarly broad. So what are some of the specific areas where Big Data can be applied?

•     Accounts receivable

•     Accounts payable

•     Duplicate payment detection

•     Sampling

•     Data imports, extractions, and analysis

•     Continuous auditing and monitoring

•     Fraud detection and monitoring

•     Analysis of procurement cards

•     Payroll and time sheets

•     Joins and comparisons

•     Inventory audits

We’ll look at each of these areas in more depth in the sections that follow.

Accounts Receivable (AR)

According to Ventana Research, Big Data applications can be used in accounts receivable to advance consumer satisfaction and loyalty.² For example, an organization that does an initial investigation of payment patterns can have a good idea of when particular clients will pay. Let’s assume, for example, that one customer who routinely pays his or her balance between the certain days of the month has not paid a week after his or her usual payment date. By applying Big Data analytics, an alert would be automatically generated with triggers to follow-up actions. A call to the customer might be made, or an automated email notification sent to notify the customer regarding the delayed payment. There are several advantages to this approach. If there is no exception, other than poor follow-up by the customer’s accounts payable (AP) department, the notifications should prompt corrective action. This is also a more timely and preferable approach to other more punitive actions. Resolving accounts receivable (AR) issues sooner enhances cash flow, and if your organization made a mistake, the client would be upset when you demand payment.

Another use for Big Data in receivables is the identification of clients who are routinely late in paying their bills. This can result in internal company discussions about possible solutions such as restricting credit or discovering approaches to elicit quicker payments.

As valuable as data analysis is for estimating, it may be significantly more profitable when connected to audits and cautions. Data analysis can identify a pattern against which actual payments can be compared. This can create an early warning system when payments deviate from expected timeframes. When customers are in financial difficulty, it is normal procedure for them to ignore their vendors (in this case—your organization.)

What are some ways that an organization could support AR with Big Data that the organization may not have used in the past?

•     Search streaming media (social media, google alerts, and the like) for any comments about the financial health of the customer.

•     Track additional remittance information that may suggest that payments are being held by the customer.

•     Determine customer’s financial position as opposed to industry benchmarks.

•     Evaluate historical change in credits that were taken or disputed items.

KNOWLEDGE CHECK

2.     Which item was NOT listed as a way to help process and collect AR?

a.     Search streaming media.

b.     Search unstructured remittance information.

c.     Create an effective dunning process.

d.     Stratify AR and focus on the top 80 percent of value.

Accounts Payable (AP)

Given the sheer amount of data contained within AP, predictive analysis is the most value-adding process. Many have said that they are assessing data inside the vendor master document, although others said that the focus of their predictive analysis was the areas of purchasing-to-payment analysis, review of operating disbursement (with the primary attention being given to AP), and validation of payments to vendors.³

Duplicate Detection

One often overlooked procedure that can be applied to asset recovery is a duplicate payment check. It is a method to discover recovery dollars, accounting mistakes, outliers, fraud, or anomalies. Big Data can drastically reduce the amount of time needed to investigate duplicate payments, invoices, transactions, and vendor returns for credit.

An annual download is recommended for the disbursement data. The data should be sorted in a spreadsheet based on invoice numbers, dates, amounts, and the like. Keep in mind that traditional safeguards built into AP systems have long been overridden by AP clerks with the simple addition of a "dash" or addition of an "A" to fool the software that the invoice is original.

KNOWLEDGE CHECK

3.     What tool was recommended to check for duplicate payments?

a.     An outsourced financial person.

b.     A database.

c.     A spreadsheet.

d.     Streaming data.

Sampling

Sampling is a major aspect of audit work, and with such a variety of approaches to use, it is no surprise that sampling is one of the leaders in adding value to data analysis. One popular tool is using IDEAs statistical sampling techniques to set the scope. With good statistical analysis, the results can be extrapolated to the overall population of data. Statistical functions, summarizations, and stratifications help leadership analyze data to make projections and assess past performance.

Stratified sampling can be used to look at losses and gains on backdated trades, to stratify invoices and payment, and to execute random sampling. Data analytics is likewise used to test and perform unique word searches. One of the numerous advantages data analysis technology offers the capability to break down the data of an entire population, in contrast to sampling that only looks at a percentage. Regardless, data analysis permits you to join, sort, and summarize information to analyze smaller sets of data.

Data Imports, Extractions, and Analysis

For most auditors and accounting experts, the greatest difficulty is knowing where to apply data analysis—getting the needed information and importing it into an analysis tool. Data analysis programming engineers have made vast innovations to disentangle data imports, using tools like PDF converters, drag-and-drop abilities, and the extension of data capacity limits. New technology also allows for extraction within specific criteria (such as name or client codes). And tools like IDEA and ODBC can save hundreds of work hours over MS query or SQL because they allow for more efficient downloads, examinations, and summarized information.

Auditors are using data analysis to perform large downloads of financial transactions, alongside synopses and analytics, to help scope and test amid the review. For example, an internal auditor in the health services industry is using data analysis to acquire payment data from an enterprise resource planning application to review for payment dates on weekends, duplicate record search, and account coding errors.

Continuous Auditing and Monitoring

Data analysis is used to automate manual procedures and regularly test systems via continuous monitoring. Special programs can be implemented to establish nonstop review scripts to detect data entry errors as well as exceptions to travel and entertainment, budget and financial statement exceptions, procurement cards, HR, and accounts payable.

Though monitoring can identify real-time exceptions, it is usually more feasible for accountants to process only monthly or quarterly. Continuous monitoring will be a function of the likelihood of an error, the magnitude of an error, or the other controls in place.

Other areas that should be considered for continuous monitoring are regulatory compliance, credit information, market data, financial information, and any other major changes.

Also, consider monitoring major industry trends and company trends via the following:

•     Google alerts

•     Social media posts, tweets, and the like

•     Criminal postings

•     Management and discussion analysis from SEC reports

KNOWLEDGE CHECK

4.     Which of these was NOT listed as a difficulty in Big Data analysis?

a.     Where to apply data analysis.

b.     Acquiring the necessary information.

c.     Having the right tool to analyze the data.

d.     Importing the data into the system.

Fraud Detection and Monitoring

Though you may discover fraud when looking for anomalies or oddities, many accountants are using data tools as a scientific device to hunt down fraud and schemes. Data analysis tools permit the review of information from diverse perspectives to identify the real cause of the fraudulent situation. Some of the ways data analysis is being used to look for patterns of fraud include trend analysis, behavioral analysis, and changes in spending patterns.

Analysis of Procurement Cards

Procurement cards have reduced the amount of administrative work required to handle the processing of small purchases and posting of transactions. However, procurement cards have resulted in additional control risks. Although procurement cards have helped "lean" accounts payable departments, they must be managed with appropriate controls to diminish abuse and waste. Accountants must examine the activity continuously to analyze trends and spending patterns. The same processes to manage accounts payable should be conducted for aggregate procurement cards and individual procurement cards. If inappropriate behavior is occurring, it should be detected in the analytics process.

Payroll and Time Sheets

Chasing down errors or fraud related to timesheets and payroll can be time-consuming for auditors. Respondents to a survey conducted by Audimation Services stated that they execute "weekly payroll dollar or total reasonableness testing," or they perform 30+ examination systems on a quarterly premise covering purchase-pay, revenue, journal entries and payroll.⁴ Data analysis is also adding value by surveying electronic time and attendance records for consistency and compliance with existing arrangements, systems, and work regulations.

A respondent said they had reduced "the time required to review payroll prior to [our] weekly transmission to ADP from several hours to less than 30 minutes."⁵

Joins and Comparisons

One of the greatest qualities data analysis conveys the capacity to join databases and records to sort, summarize, and investigate information. This enables the ability to look at information from distinctive points to track exceptions, misrepresentations, blunders, and other data. When databases or files are joined, they can be used to check for anomalies, perform inventory turnover analysis and stock analysis, extract data from PDFs to create a new analysis, reconcile outstanding checklists, and replace manual conversion processes.

Some examples of ways that joining processes can be used to compare sets of data are:

•     Active application user accounts against a list of terminated employees

•     Physician scheduling system compared to the billing system

•     Vendor terms including days paid early and aged receivables

•     Missing invoice numbers

•     Employee and HR records compared to vendor address

•     Tax ID numbers and conflict of interest

•     Review of billing and pricing agreements against contract agreements

Inventory Audits

•     Analytic tools are also being used to analyze inventories. They can be used in the following ways:

•     Inventory audits

•     Identify slow moving items

•     Identify obsolete items

•     Reconcile the inventory counts

•     Conduct test counts

•     Fixed asset inventories

Centralized Accounting and Data Mining

When companies have a centralized database, it is easier to perform data mining analysis. Consider vendor analysis, customer analysis, expense analysis, journal entry analysis, and the like.

Compliance

Data analysis is valuable in meeting industry and documentation requirements. An internal audit function can use data analysis for arranging, executing, and reporting SOX compliance as well as reviews of other areas and procedures.

Data Warehousing

Data analytics is being used to help create data warehousing strategies incorporating legacy systems.

Population Testing

In the past, data analysis was limited in its ability to gather large volumes of data that could be accessed, massaged, and analyzed. Data analytics today allows this ability and, as a result, the amount of data can be increased—possibly to the entire data population as opposed to relying on sample data and a corresponding extrapolation.

Reconciliation

The power of data analytics allows greater analysis and reconciliation of data. Some governmental contracts require great accuracy. However, the volume of data and the reporting levels made this requirement difficult to maintain prior to data analytics.

Revenue Reporting

Using data analysis for revenue reporting enables the combination of disparate databases, making corporate-wide reporting quicker and more efficient. Similar to the way it can enhance and improve government reporting, data analytics has also improved the ability to calculate revenue recognition at a level that was previously not feasible.

Creative Uses

Respondents to the Audimation survey also noted some other creative uses for data analytics, including the following:

•     Hourly energy bidding analysis

•     Analyzing client behaviors around periods of renewal

•     Analyzing surveys provided by customers

•     Measuring the adequacy of a promotional program

•     Analyzing of student enrollment

•     Reviewing billing

•     Monitoring the Foreign Corrupt Practices Act

•     Integrating of student data with online learning systems

•     Appraising risks

•     Analyzing bad medical debt

•     Testing data in different system logs

•     Analyzing wire exceptions in banking systems

•     Evaluating loans and analyzing portfolios⁶

PREDICTIVE ANALYTICS AND ACCOUNTING

What is the role of predictive analytics and the accounting department? Review the following definition from chapter 1:

Predictive analytics is the practice of extracting information from existing data sets to determine patterns and predict future outcomes and trends. Predictive analytics does not tell you what will happen in the future. It forecasts what might happen in the future with an acceptable level of reliability, and includes what-if scenarios and risk assessment.⁷

How Nissan Used Predictive Analytics to Survive a Natural Disaster

Predictive analytics can create a competitive advantage for your company that could even help it weather a natural disaster. Consider Nissan and the 2011 tsunami. Japan was crushed by an earthquake and the subsequent tsunami on March 11, 2011, that killed more than 20,000 people and severely damaged the Fukushima Daiichi nuclear plant.

As the country began to recover, it was clear that the effect on the worldwide economy and Japanese organizations would be massive. Harm to organizations in Japan would bring about enormous supply chain disturbances all through the world.

The data analysis of John Wilenski, CPA, CGMA, at Nissan Motor Co. Ltd played a significant part in helping Nissan remain afloat amidst the aftermath.

More than 45 of Nissan’s critical suppliers suffered serious harm as a result of the catastrophe, according to research by the Massachusetts Institute of Technology (MIT) and Pricewaterhouse Coopers.⁸ Fortunately for Nissan, analytics gathered before the quake of its supply chain helped the organization use sound judgment afterward in the rebuilding process. This success was part of the process created by Wilenski to monitor Nissan’s vendors.

Nissan used budgetary information and data provided by supplier CFOs, often updated weekly or even daily. Wilenski’s group created different models for evaluating suppliers’ financial health, including a cash flow assessment tool, a stress test with "what-if’ scenarios of suppliers and a break-even tool. These tools were based on supplier information and global economic data. This process allowed Nissan to determine which suppliers would survive the calamity and which suppliers would need help.

MIT and PwC report stated that solid risk management and viable countermeasures helped Nissan end 2011 with a 9.3 percent improvement in productions, contrasted with the stark 9.3 percent decrease across the entire industry.⁹

Ten Keys to Executing Data Analysis

Wilenski is building a predictive investigation tool for higher education. Here are his 10 suggestions to implement data analysis:¹⁰

•     Distinguish the business objective.

—     Predict bankruptcy or default?

—     Minimize risk?

—     Increase profitability?

—     Lessen costs?

—     Retain employees?

—     Draw in new clients?

•     Find data sources.

—       Company-owned systems

—       Outside sources

•     Government

•    

•     Association

•     Public database

•     Unstructured or streaming data

—       Surveys

•     Build at the lowest level of detail.

•     Verify that information is accurate, timely, and helpful.

•     Determine the quality of data.

•     Figure out which information may be predictive. It may be necessary to aggregate data and determine its correlation.

•     Automate and computerize with the possibility of human interaction and intervention.

•     Communicate simply and in the language of your audience.

•     Be collaborative. Get support and insight.

•     Continually improve the model.

ANALYTICAL PROGRAMS AT THE SEC

The following excerpt was from the speech "Has Big Data Made Us Lazy," delivered by Scott W. Bauguess, Deputy Director and Deputy Chief Economist, DERA to the American Accounting Association in the Fall of 2016. Note the ways that the SEC is utilizing Big Data in his comments.

Why are these different perspectives important for how we approach data analytics at the SEC? I’ve been at the Commission for nearly a decade. During that time, I have worked on a large number of policy issues. The economic analyses that have supported these policy decisions are predominantly grounded in the theory-driven research of social scientists. They rely on carefully constructed analyses that seek to address causal inference, which is crucial to understanding the potential impact of a new regulation.

But in the last few years, I’ve witnessed the arrival of increasingly complex data and new analytical methods used to analyze them. And some of these analytical methods are allowing analyses of previously impenetrable information sets—for example, those without structure, such as freeform text. This has been of particular interest to the SEC, where registrant filings are often in the form of a narrative disclosure. So, as a result, we have begun a host of new initiatives that leverage the machine learning approach to behavioral predictions, particularly in the area of market risk assessment, which includes the identification of potential fraud and misconduct.

Today, the SEC, like many other organizations, is adopting these new methodologies at a very rapid pace. Of course, this is not to say that we are letting go of classical statistical modeling. And, as I would like to focus on now, none of our analytical programs, whether grounded in classical statistical modeling or machine learning, can replace human judgment, which remains essential in making the output of our analytical models and methods actionable. To understand why let me give you some examples.

Let me begin with the Corporate Issuer Risk Assessment Program, also known as CIRA, which relies on classical statistical modeling developed by DERA economists and accountants in collaboration

with expert staff in the SEC’s Division of Enforcement. This program grew out of an initiative originally referred to as the "accounting quality model," or, AQM, which was itself rooted in academic research. In particular, AQM focused on estimates of earnings quality and indications of inappropriate managerial discretion in the use of accruals. As former DERA Division Director and SEC Chief Economist Craig Lewis noted, "[a]jcademics in finance and accounting have long studied the information contained in financial statements to better understand the discretionary accounting choices that are made when presenting financial information to shareholders."[iv]

Today, the CIRA program includes these modeling measures of earnings quality as part of more than two hundred thirty (230) custom metrics provided to SEC staff. These include measures of earnings smoothing, auditor activity, tax treatments, key financial ratios, and indicators of managerial actions. Importantly, they are readily accessible by SEC staff through an intuitive dashboard customized for their use. Referencing DERA’s collaboration with the Division of Enforcement’s FRAud Group, Enforcement Division Director Andrew Ceresney noted earlier this year, "CIRA provides us with a comprehensive overview of the financial reporting environment of Commission registrants and assists our staff in detecting anomalous patterns in financial statements that may warrant additional inquiry."[v]

However, this was not how the press first reported on the original initiative when it coined the term "Robocop" to describe it—as if a machine makes the important decisions in identifying potential market risks. As our current DERA Director and Chief Economist Mark Flannery recently noted, "this implied perspective is at best inaccurate and at worst misleading. While these activities use quantitative analytics designed to help prioritize limited agency resources, the tools we in DERA are developing do not—indeed cannot—work on their own."[vi]

But at the same time, some of the most exciting developments at the Commission have centered on machine learning and text analytics. While machine learning methods have been around since the 1950s,[vii], it is the arrival of big data and high-performance computing environments that have advanced their uses. At the Commission, this has taken on several forms. At the most basic level, and consistent with methods that are now commonplace in academic research, we have extracted words and phrases from narrative disclosures in forms and filings. For example, by applying a programming technique that uses human-written rules to define patterns in documents, referred to as "regular expressions," [viii] we are able to systematically measure and assess how emerging growth companies are availing themselves of JOBS Act provisions through what they disclose in their registration statements.

More recently, we have adopted topic modeling[ix] methods to analyze tens of thousands of narrative disclosures contained in registrant filings. For those of you not familiar with topic modeling, when applied to a corpus of documents, it can identify groups of words and phrases across all documents that pertain to distinct concepts ("topics") and simultaneously generate the distribution of topics found within each specific document. We are also performing sentiment analysis using natural language processing techniques to assess the tonality[x] of each filing—for example, identify those with a negative tone, or a tone of obfuscation. We then map these topic and tonality "signals" into known measures of risk—such as examination results or past enforcement actions—using machine learning algorithms. Once trained, the final model can be applied to new documents as they are filed by registrants, with levels of risk assigned on the basis of historical findings across all filers. This process can be applied to different types of disclosures, or to unique categories of registrants, and the results then used to help inform us on how to prioritize where investigative and examination staff should look.

While this machine-learning approach to text analytics has provided a new and exciting way to detect potential market misconduct, just as with classical modeling methods, it does not work on its own. In particular, while a model may classify a filing as high risk, the classification does not provide a clear indicator of potential wrongdoing. To the contrary, many machine learning methods do not generally point to a particular action or conduct indicative of fraud or other violation. The human element remains a necessary part of the equation.¹¹

Note: the roman numeral references included in the speech are as follows:

[iv] Craig Lewis, Chief Economist and Director, Division of Risk, Strategy, and Financial Innovation, U.S. Securities & Exchange Commission, Financial Executives International Committee on Finance and Information Technology, Dec. 13, 2012.

[v] Andrew Ceresney, Director, Division of Enforcement, U.S. Securities & Exchange Commission, Directors Forum 2016 Keynote Address.

[vi] Mark Flannery, Director, Division of Economic and Risk Analysis, U.S. Securities & Exchange Commission, Global Association of Risk Professionals Risk Assessment Keynote Address.

[vii] See Arthur L. Samuel, Some Studies in Machine Learning Using the Game of Checkers, IBM Journal, Vol. 3, No. 3, July 1959.

[viii] Thompson, K. (1968). "Programming Techniques: Regular expression search algorithm." Communications of the ACM. 11 (6): 419-422. doi:10.1145/363347.363387.

[ix] See, for example, David Blei, "Probabilistic Topic Models," Communications of the ACM. 55, April 2012.

[x] See, e.g., Tim Loughran and Bill McDonald, 2011, "When is a Liability not a Liability? Textual Analysis, Dictionaries, and 10-Ks," Journal of Finance, 66:1, 35-65.

Exercise: How many Big Data techniques are being used by the SEC and how could they be utilized by the accounting department?

Practice Questions

1.     What type of data should be accessed to increase value?

2.     What were some of the sources for industry trends?

3.     Describe the steps to creating a predictive analytics program.