Reports

The focus of this book revolves around informatics, which largely involves the utilization of business intelligence applications (e.g., OLAP, dashboards, mining) to extract actionable information from all types of data to enhance the decision-making process. One of the most basic levels of this approach is the creation of business reports that incorporate sequel-related queries of data resources to extract variables that describe a business scenario. The introduction of big data involves additional requirements for this process; namely, when devising the parameters of the report to be created, the decision maker now must consider new variables that impact that conceptual report. The volume of data that must be processed must also be considered, and finally, the currency of the report (e.g., how often a report must be updated to provide adequate information for the decision maker). However, as simple as the process of generating a report may be, creating one that provides essential information to those that receive it be a quite complex task.

Consider a request by a hospital’s administrative area to produce an analytic report that provides some information on patient volume and hospital capacity along with patient satisfaction. Although this initiative appears to be straightforward and simplistic in nature, one must consider all the variables that comprise the area to be analyzed, along with the needs of the user of the report.

Some dimensions and variables that could be included in this analysis would involve the following.

When conducting customized analytics (tailored analytics to a specific company’s activities) data experts and analysts must apply due diligence to acquire that information that provides a strategic advantage in the marketplace. This involves the storage, processing, management, and ultimate analysis of data resources that describe a particular process.

Well-designed reports that incorporate the pertinent and available variables that describe a business activity can be an important source of information to decision makers (see Figure 11.1). However, the limitation of information creation at the report level is that the user often scans a report, assimilates the information, and quickly thinks of alternative business scenarios that are essential to providing more robust information regarding a process or activity. The report is limited to its current level of data aggregation and variables depicted. The next step to analysis or business intelligence involves the application of OLAP, which gives users the flexibility to view and analyze multiple scenarios of a business process. Before we describe the application of OLAP functionality that leverages large data resources and addresses currency of data, consider the more simplistic spreadsheet application of pivot tables.

Figure 11.1 (See color insert.) Hospital patient volume and satisfaction.

Pivot Tables

A simplistic version of OLAP that many users can quickly relate to includes the use of pivot tables in a spreadsheet environment. Pivot tables leverage data in a flat, spreadsheet file to present alternative scenarios that describe a business activity. Through basic spreadsheet functionality, users can quickly generate a table view of relevant variables at a particular level of aggregation. For example, a spreadsheet of data that describes a software company’s sales activities can include numerous rows according to corresponding variables. Hypothetical data recording national sales activities of branches across the country are illustrated in Table 11.1.

Table 11.1 Hypothetical Data Recording National Sales Activities

With a simple pivot function, Table 11.2 could be calculated with ease.

Dynamic Reporting through OLAP

Pivot tables are similar to OLAP in that they provide a multidimensional view of an activity. Enterprise OLAP provides greater scale to the analytic process, as it provides the platform to address multiple levels of aggregation of data resources, can depict updated views as source data are updated, and can process extremely large volumes of data. With this flexibility OLAP can help decision makers investigate information addressing multiple descriptive scenarios regarding an operation’s activity, therefore enhancing the knowledge generation process and overall ability to generate effective strategic conclusions. The diversity of information views involves various dimensions of time, performance metrics, and descriptive variables.

These inputs must be organized to provide information (variables at levels of detail) that describes a business scenario to facilitate decision support for the end user. Consider the graphical view of a cube in Figure 11.2.

Figure 11.2 depicts an illustration of an OLAP cube that facilitates analytics of patient satisfaction rates at a cardio area in a hospital. The cube presents a multidimensional view of a few variables that could potentially affect patient satisfaction. The platform gives the analyst the ability to query data variables from different levels of detail and in different combinations, through both numeric data and visualization. The tabs at the top of the graphic depict the variables that are available to be analyzed. The scenario depicted illustrates the number of daily visits to a patient by hospital staff (nurses, nurse assistants, technicians) and the corresponding patient satisfaction rates.

Figure 11.2 (See color insert.) Multidimensional cube.

Users also have the ability to change variable views regarding patient satisfaction from different perspectives, including:

•Time (weekly, monthly, quarterly)

•Illness severity (critical, observation: according to cardio Diagnosis-Related Group [DRG])

•Length of stay (LOS; duration of excess of LOS beyond expected)

•Physician (primary physician, specialty)

By navigating the different dimensions of the cube, the analyst can quickly identify strengths and weaknesses of different operational and patient descriptive variables on the performance metric. OLAP enhances the decision makers’ ability to understand more fully some of the attributes that drive patient satisfaction. Other variables can be introduced with the deliberation of stakeholders (e.g., staff training in patient communication tactics, etc.).

So what about big data you say? Remember, big data entails not only volume of data but also the new variables (sources of data). Both of these factors are considered when conducting analytics. In other words, a conceptual model must be generated that best describes the attributes of a desired process (entity to be better understood), and then data corresponding to those variables must be applied to that analytic framework. Big data adds complexity to the generation of the conceptual model as it introduces new descriptive variables that may not have been available or incorporated in the traditional structure of the particular process (e.g., categorizing physician comments on electronic health records [EHRs] during patient visits). The value of big data follows the basic concepts just mentioned; however, it can provide even greater value to the user by providing more robust models that provide greater descriptions and understanding of what affects process performance. In the patient satisfaction scenario depicted in the preceding text, perhaps a new variable that leverages social media (e.g., hospital Facebook likes) can be incorporated for another frame of reference. When considering big volumes and velocities of data in an OLAP environment, methods such as parallel processing and map reduction of data resources must be considered.

OLAP provides a robust source of business intelligence to decision makers, as it can leverage data resources including big data volumes and also provides a platform that offers a flexible, accurate, and user-friendly mechanism to understand quickly what has happened and what is happening to a business process. The multidimensional framework will give users the power to view multiple scenarios of a given process, such as the following:

•Are there associations between DRGs and excesses in patient LOS?

•What association do nurse-to-patient ratios have with readmit rates?

•How does patient demand for healthcare change during different months of the year?

The key to a valuable OLAP cube involves the combination of a few factors. One of these relates to the concept mentioned earlier, namely, that a cube must effectively describe a business scenario. The conceptual model that is used to build the cube must include noteworthy variables (relevant) with an appropriate detailed format that give users true business intelligence. The next major factor is filling the cube with accurate, current, and consistent data. Deficiencies in either of these areas can quickly render the analytic method useless for decision making. It should be noted that using OLAP for certain healthcare applications (e.g., outcomes associated with processes) introduces greater complexity than standard industry environments (e.g., sales by product by region) and requires more complex dimension modeling.¹

Analytics at a Glance through Dashboards

In today’s ultrafast, ultracompetitive information-based economy, it seems that the more senior a manager you may be, the less time that is available for investigation and drilling around multidimensional cubes. Often the level of analytics is filtered down to a few insightful reports, ongoing insights absorbed in the marketplace, and the access to real-time dashboards that display key performance indicators relevant to a particular process. These dashboards are designed to provide decision makers with a feedback mechanism as to how an organization is performing closer to real time. In the healthcare industry, streaming data from a medical sensor depicting real-time vital information of a patient takes the importance of a dashboard to a new level (just consider the traditional EKG or heart monitor). The key elements of dashboards are the delineation of relevant key performance indicators (KPIs) to a particular process, timeliness of their readings (currency of information), and finally, a user-friendly visual that provides the decision maker with a clear way of determining whether a process is operating successfully or not. The more traditional visual platform resembles that of an odometer in an automobile, where color schemes of performance reflect that of traffic lights (e.g., green, all is well; yellow, caution; and red, something is wrong and needs to be investigated). However, dashboard technology is quickly evolving where styles can include combinations of a variety of visuals (bar, line, pie charts) according to designated scales and are being utilized by decision makers at all levels in an organization.

The key to the effectiveness of a dashboard design involves its connection to the process at hand and use for decision making. Displays must be simple to understand and interpret. Just as a simple graphic display must adhere to design conventions (e.g., coherent color scheme, axis labeling, scale), so too must dashboard design, which adds complexity to the process as it combines various visual elements. The true key to a successful dashboard is evident by its effectiveness in providing timely, easy-to-understand decision support of a corresponding process. Dashboards that are too busy (include too many visuals), that are difficult to interpret, can quickly become omitted from an analyst’s arsenal of decision support information.

Consider the dashboard example in Figure 11.3. The various graphic displays are clearly delineated from one another (separate sections) and are clearly labeled. Also, the design includes different visual displays, so the information presentation does not appear to overlap or include a blended view. Finally, complementary but distinctly different KPIs give the decision maker a well-rounded view of a human capital management application in this case.

Figure 11.3 (See color insert.) Clearly designed employee analytic dashboard. (From http://www.dashboards-for-business.com/dashboards-templates/business-intelligence/business-intelligence -executive-dashboard; Domo, Inc., http://www.domo.com.)

Robust Business and Drill-Down behind Dashboard Views

Dashboards provide an instantaneous mechanism to analyze the performance status of a process. Organizations with extensive analytic capabilities through business intelligence applications can have OLAP cubes that can be quickly drilled into from a dashboard KPI that provides descriptive analytics of underlying variables that underpin the KPI. A prime example of an e-commerce-based KPI is the bounce rate on a landing page for an organization, especially when a new marketing initiative has been launched. Perhaps an organization has initiated an Internet marketing campaign with banners listed on various complementary referral sites. A red signal indicating a higher than acceptable bounce rate would provide decision makers with a timely analytic alert mechanism to investigate the source of the problem. A real-time cube or report could quickly depict which referral site may be the greatest source of misdirected traffic.

Not all dashboard displays need to be real time, where a simple refresh of data on an interim basis provides decision makers with an accurate indication of whether a process’s performance is adequate. However, the big data era involving high velocity of streaming data resources often requires a real-time dashboard visual of a given process to provide users with a quick view of variable impacts on KPIs.

Why Things Are Happening

Data mining can provide decision makers with two major sources of valuable information. The first refers to descriptive information, or the identification of why things may be occurring in a business process. This is done through the identification of recurring patterns between variables. Cross-sectional graphic displays can add significant information to decision makers to illustrate patterns between variables. Figure 11.4 provides a simple graphical view that illustrates an advertising spend versus dollar revenue elasticity curve as identified in the mining process. The figure depicts that a recurring pattern exists between the two variables, and that a direct relationship is prominent, where an increase in ad spend yields an increase in product revenue. Many non-mining-centric analysts would quickly raise the point that this information is not noteworthy, given the natural relationship between the two variables (e.g., the more spent on advertising, the more sales that are generated); however, this criticism is quickly dispelled when posing the question: If ad spend is increased by 5% from $200,000, what is the expected increase in revenue? That question is difficult to answer without the use of mining.

Figure 11.4 Advertising spend versus revenue curve.

Mining methods can yield insightful patterns as to demographic and behavioral attributes of consumer response to marketing initiatives, the impacts of process components on performance metrics, and many more. The following are a few prominent applications where mining is often utilized:

Healthcare-related areas (outcomes measurement, treatment effectiveness, operational performance)

Consumer propensities

Marketing and advertising effectiveness

E-commerce initiatives

Fraud detection

Worker and team performance

Pricing policies

Process-related applications (throughput, workflow, traffic analysis)

Risk assessment

What Is Likely to Happen

The other main source of information where mining provides value to decision makers is in the deployment of mining results. The patterns that have been identified are often embedded in an equation or algorithmic function, which are often referred to as the model, can be used to perform a “what if” analysis or estimate future expected results based on inputs. In other words, if I market my product to a particular market segment defined by demographics, what is my expected response rate? Or, is a particular activity (e.g., credit card use) likely to be fraudulent? If the analysis is based on a time series approach, mining models can provide forecasts for product sales. The analyst in this case needs to make assumptions as to future input values.

The healthcare industry involves a number of diverse areas for which analytics are required for knowledge generation. These scenarios often involve a number of diverse variables that encompass the descriptive drivers of performance metrics. Consider all the variables that follow and the corresponding level of detail that may provide descriptive information in better understanding performance metrics.

DRGs

Physician related

Nurse and staff related

Service entity (primary care physician, hospital, accountable care organization)

Patient descriptive and behavioral

Treatment related

Prescription related

Area of care

Time related

The incorporation of data mining methods that enable analysts to discover recurring patterns in an array of variables is often an essential technique to generate vital information.

Real-Time Mining and Big Data

The evolution of the big data era has increased the utilization of the concept of real-time or streaming mining approaches. More traditional streaming mining involves the creation of models through analyzing a data sample or historical data of a given process. The resulting model then becomes a function that can be used to process streaming or real-time incoming data, and corresponding actionable outputs are generated in real time as well. Streaming mining addresses the big data concept of velocity and volume of data and is incorporated in processes in which timely results are needed to improve strategies. Streaming mining applications are commonly applied in

Website traffic analysis for real-time online marketing

Fraud detection for online transactions

Financial market risk and trading

However, with the more prominent use of medical sensors that monitor patient metrics in the healthcare industry, the future may incorporate these real-time advanced analytic techniques to identify “trouble in the pipeline” for patients given the streaming data of vital signs.

Some big data sources (e.g., sensor- and satellite-producing entities) with extreme velocity and volume sometimes render the ability to extract a sample that represents the entire data source difficult, to say the least. In these instances, the ability to create optimized quantitative models to process this streaming data is limited. Techniques such as multisampling² and the implementation of self-optimizing quantitative techniques that learn as data are encountered have evolved to address this issue.

Analysis of Unstructured Data and Combining Structured and Unstructured Sources

Up to this point, this chapter has dealt with analytics of structured data. The big data era, however, largely involves the incorporation of unstructured data resources that need to be analyzed to identify actionable information that enhances strategic initiatives. Text mining addresses the analytics of textual data (words, phrases, messages, emails, etc.). At a high level of description, text analytics seeks to create structure from unstructured sources. It does this by processing various unstructured forms and classifies them into particular categories. Processing is generally based in mathematics or linguistics.

In the realm of the vastly growing utilization of electronic communication, which includes texting, tweeting, leaving content on social media, emailing, and so forth, one can quickly see the possible value that exists in deploying analytic techniques to extract information that describes responses to marketing initiatives and product and service offerings, reactions to news and events, and general consumer behavior and sentiment.

An example involving the analysis of both structured and unstructured data for informative decision support is evident when examining patients’ EHRs to better understand treatment outcomes and patient diagnosis. More structured physiological data (e.g., blood sugar levels) can be combined with unstructured data (e.g., physician comments on treatment) to better understand a patient’s status. Analytic techniques such as semantic mining can be applied in this situation to extract actionable information.

Graphic Types

Figure 11.5 illustrates the classic pie chart that depicts how a whole unit is divided among some subcomponents (pieces of an established pie). Market share is a prime example for pie charts, where share can be delineated by product lines, regions, industry competitors, etc. Pie charts have limitations when considering negative values.

Despite the seemingly simplistic bar chart depicted in Figure 11.6, the visual actually incorporates a number of important elements in the realm of analytics. The graphic depicts a comparative view of a multicomponent process (call centers in this case) in a time series setting (quarterly views). With a quick glance, the analyst can make inferences regarding relative performance (customer satisfaction) of three different call centers over time. Bar charts are more appropriate in depicting quantities or amounts of select variables.

Figure 11.5 Pie chart depicting market share.

Figure 11.6 Bar chart (comparative view of multicomponent process).

Bar charts are also often used to illustrate variable distributions (percentages of ranges or categories of a given variable). Figure 11.7 depicts a categorical age variable and the amount of data that exists in selected ranges. This gives analysts a better understanding of the dimensions of a given data variable, and in this case enables them to determine if there is any age skew or bias (high percentage of one age range relative to the population). In conducting market research, a variable distribution view enables the researcher to determine if a target market is included in a data resource.

Variable distribution analysis can often include visuals via line graphs that are useful in illustrating scenarios involving continuous variables. Figure 11.8 illustrates the continuous data variable of mall foot traffic for a given day according to retailers.

Time series line charts provide users with a visual of potential seasonality in processes. Figure 11.9 depicts the classic holiday effect in retail as is seen in the repetitive bump in sales in Q4.

Another type of chart involves the scatterplot that is commonly used to illustrate correlations between variables, where simple plots of individual data points are depicted. Figure 11.10 depicts data points illustrating correlations between employee performance and training received.

Figure 11.7 Age distribution chart.

Figure 11.8 Line chart of continuous variable distribution of mall traffic.

Figure 11.9 Time series line charts for seasonality.

Figure 11.10 Scatterplot for correlations.

A rather insightful chart style is the bubble chart. The bubble graphic enables analysts to depict three-dimensional scenarios in a coherent fashion by incorporating bubble size to illustrate variable attributes. Figure 11.11 depicts the multidimensional scenario of organizational team performance according to workload and team size.

Yet another graphic style that has increased in importance over the evolution of the big data era is the use of maps. Map visuals are generally utilized when an analysis involving location is emphasized; however, location can also refer to a process location. Applications such as traffic analysis or population analytics are common examples. Traffic can refer to website activities, vehicular, consumer, or some type of designated activity.

In a simple web traffic visual, a map can illustrate cross sections of time and area of a webpage that are receiving high user traffic. This can provide strategists with actionable information to more effectively apply online marketing tactics (e.g., display banners in hot spots on a particular page at a particular time).

Civil engineering can leverage heat maps by incorporating GPS data to investigate hot areas of traffic incidents (congestion, accidents) and optimize new designs to alleviate existing trouble areas and in designing new roadways.

Figure 11.12 provides a standard heat map where “hot colors” depict more intense activity. In this case, the hotter areas depict areas where job vacancies are difficult to fill.

Figure 11.11 (See color insert.) Bubble chart depicting workforce team performance.

Figure 11.12 (See color insert.) Heat map that illustrates areas of hard-to-fill job vacancies. (From Wanted Analytics, http://www.wantedanalytics.com.)

Map visuals are particularly applicable in the big data era, when real-time, high-velocity analytics and voluminous sources are involved. Applications that leverage big data include geovisualization that involves the analysis of geographic specific flows of data and bioinformatics and sensor output in the healthcare spectrum. For example, the healthcare industry is increasingly utilizing streaming sensor data generated by various treatment and diagnostic technologies. For diagnosis (magnetic resonance imaging [MRI]), these data describe the characteristics of a patient. Visual displays of this source are essential to extract information on trouble areas for patients. Chapter 9 (Figure 9.5) provides more detailed information corresponding to this concept. As big data sources emerge, the application of heat maps should become a common visual technique for providing analysts with a mechanism to enhance strategic initiatives. Figure 11.13 depicts a heat map showing percentage likelihood that a primary melanoma site will show lymphatic drainage to axillary lymph-node fields. While Figure 11.14 depicts an example of computerized tomography, three dimensional image of the brain and blood vessels.

Figure 11.13 (See color insert.) Three-dimensional visualization of lymphatic drainage patterns in patients with cutaneous melanoma. (From Reynolds, H., Dunbar, P., Uren, R., Blackett, S., Thompson, J., and Smith, N. Three-dimensional visualisation of lymphatic drainage patterns in patients with cutaneous melanoma, Lancet Oncol 8:806–12, 2007 (Figure 1), http://www.thelancet.com/journals/lanonc/article/PIIS1470-2045%2807%2970176-6/fulltext?rss=yes.)

Value of Data and Analytics

We began this chapter by stressing the importance of analytics as an essential component to deriving value from data, where the era of big data adds intensity to the concept, as it adds new dimensions to the equation. Regardless of the source of data, its value is not realized unless it provides some resource to a strategic endeavor. Rarely does a decision maker reference a data source without first formulating a reason to do so. Once the conceptual need is defined, only then can data provide value.

The conceptual need involves the quest to better understand a process with the goal of enhancing its efficiency or productivity. Simply analyzing random data and coming up with associations between variables may actually generate negative returns because the analytic process requires time and resources, and the result may not add meaningful information.

Figure 11.14 (See color insert.) Computer tomography image. (From Shutterstock, http://www.shutterstock.com/pic-221642638/stock-photo-medical-illustration-of-the-brain-and-head -arteries.html.)

Consider the growing data resource in the area of sports. More variables (variety) and real-time downloads of various athletic activities at corresponding events (velocity and volume) may seemingly provide great value to understanding various attributes of different athletes and sports teams. However, to truly generate value for decision making, a conceptual model must be created. Consider the quest to better understand what leads a team to achieve a winning record. An analysis of corresponding data could yield the following result: basketball teams with winning records generally score more 3-point shots.

At first glance, this may seem to be very valuable information, but the revelation proves limited at best when looking to make a strategic decision. What does a coach do in leveraging this associative pattern—encourage players to take more shots from the 3-point zone? Does he change practice to intensify skills for increasing 3-point percentages for players? And if so, what happens to the team’s performance from the 2-point zone, and does a reduction in 2-point conversions decrease the likelihood of winning a game despite an increase in 3-point shots? In the case at hand, does the variable of number of 3-point shots really add descriptive value to what leads to a team’s success? Perhaps more appropriate variables that can provide strategic action could entail

Team practice data (frequency, drills, duration)

Player descriptions (height, speed, position, age)

Type of offensive and defensive tactics

Identifying patterns among these types of variables empowers a coach (decision maker/strategist) to implement strategic initiatives that impact a performance metric or defined objective—winning.