Big Data and Analytics Demystified
If analytics is any mathematical or scientific method that augments data with the intent of providing new insight, aren’t all data queries analytics? No. Analytics is often thought of as answering questions using data, but it involves more than simple data (or database) queries. Analytics involves the use of mathematical or scientific methods to generate insight from the data.
Analytics should be thought of as a progression of capabilities, starting with the well-known methods of business intelligence, and extending through more complex methods involving significant amounts of both mathematical modeling and computation. Reporting is the most widely used analytic capability. Reporting gathers data from multiple sources, such as business automation, and creates standard summarizations of the data. Visualizations are created to bring the data to life and make it easy to interpret. As a generic example, consider store sales data from a retail chain. The data is generated through the point of sale system by reading the product bar codes at checkout. Daily reports might include total store revenue for each store, revenue by department for each region, and national revenue for each stock-keeping unit (SKU). Weekly reports might include the same metrics, as well as comparisons to the previous week and comparisons to the same week in the previous calendar year. Many reporting systems also allow for expanding the summarized data into its component parts. This is particularly useful in understanding changes in the sums. For example, a regional store manager might want to examine the store-level detail that resulted in an increase in revenue from the home entertainment department. She would be interested in knowing whether sales increased at most of the stores in the region, or whether the increase in total sales resulted from a significant sales jump in just a few stores. She might also look at whether the increase could be traced back to just a few SKUs, such as an unusually popular movie or video game. If a likely cause of the sales increase can be identified, she might alert the store managers to monitor inventory of the popular products, reposition the products within a store, or even reallocate inventory of the products across stores in her region.
Descriptive and Predictive Analytics
Reporting, also known as descriptive analytics, is focused on reporting what has happened, analyzing contributing data to determine why it happened, and monitoring new data to determine what is happening now. Business intelligence (BI) software provides capabilities for descriptive analytics. BI helps enterprises understand how their organization is operating by giving them a clear picture of the recent past. The deployment of BI software also requires an enterprise to carefully think about its key performance metrics, to understand what it wants to measure and monitor, and to develop some process for exploring potential causes underlying changes in measurements. Sometimes it is obvious what should be measured. Other times it is less obvious, such as when someone discovers a leading indicator of revenue through analysis. Once these capabilities have been mastered, many organizations seek to look forward and get “headlights” on what will happen in the future. They turn to predictive analytics, which uses techniques such as statistics and data mining to analyze current and historical information to make predictions about what will happen in the future. Predictive analytics typically produces both a prediction of what will happen and a probability that the prediction will happen.3
In many ways, the difference between descriptive analytics and predictive analytics is like the difference between a weather report and a weather forecast. Whereas a report describes what has happened, a forecast tells what is likely to happen and how likely it is to happen. An accurate prediction has greater business value, which is why considerable effort is applied to finding data that contains valuable signals about the future and developing analysis methods that can extract the signal effectively. In the past, performing predictive analytics had a reputation of requiring deep analytic skill. Today, modern tools have broadened the population of users who can leverage predictive analytics to augment decision making. No introduction to predictive analytics is complete without the caution that a prediction can be no better than the data that is used to make it. Thus incorrect data will likely lead to incorrect predictions, and events that have never been observed and captured in the data will never be predicted from analysis of that data.
Predicting what will happen based on the past can be quite useful. It can, for example, help a regional retail manager understand demand for a frozen dessert in the stores he manages as a function of the weather and the local competition. But, more importantly, it can also show him the relationship between past sales of this dessert and promotional pricing, coupons, and advertising. If he assumes that the same relationships will hold going forward, he can then estimate (or predict) future sales that will occur under different pricing and advertising schedules. Predictive analytics that finds relationships between actions and outcomes is particularly useful.
We use the term prescriptive analytics to cover analytics methods that recommend actions. In general, the goal is to find a set of actions that is predicted to produce the best possible outcome. To do this, you need to understand the relationship between actions and outcomes. In many cases, that relationship is clear and (more importantly) constant. If producing and distributing a newspaper circular costs 30 cents per copy, then the cost of distributing it to any specified number of customers is clear: 30 cents multiplied by the number of customers. However, the value, in terms of increased revenue, of the circular can only be inferred by looking at past data and, for example, comparing sales in weeks during which no circular was distributed to sales in weeks in which a circular was distributed. But the inferred relationship may be different in different towns, and it may vary by week of the year. That is, the computed value must be recalculated whenever new data becomes available. Despite the limitations, it seems clear that value can be derived through careful and appropriate use of predicted relationships to make decisions. Mathematical optimization has been used for decades to recommend decisions based on known, constant relationships between actions and outcomes. It is used extensively in supply chain and logistics decisions involving scheduling and allocation of resources. More recently, it has been used for decisions where some relationships are predicted from historical data rather than being based on physical limits (for example, transportation time) or business rules (for example, economic order quantities). Examples include setting prices based on predicted price elasticity, advertising based on predicted views of the ad and predicted sales lift per viewer, and targeted promotional offers (for example, coupons at checkout) based on customer segmentation.
Social Media Analytics
Analytics can also be applied to data that does not come from within an enterprise and to data that is not easily interpreted as transactions, such as data from Twitter. Several emerging areas within analytics are already providing business value. One area is social media analytics, which analyzes, or “listens,” to social media data to assess public opinion, or sentiment, on a variety of topics.4 Examples of social media include blogs, micro-blogs (such as Twitter), social networking (such as Facebook), and forums. It is possible to mine social media for comments on a particular subject and analyze them for positive and negative sentiment, how people feel about a topic, or the “pulse” of a community related to a topic. A common way to communicate sentiment is to use a dashboard with views for volume trends over time, sentiment trends over time, and geographic distribution of sentiment. A frequent use of social media is to mine customer data to get feedback on products.5 Measuring the return on investment (ROI) of using analytics for social media can present a challenge. Value driver trees (that is, visualizations of the hierarchy of influences of different value drivers) have been used to measure the value of using social media analytics for a sports clothing manufacturer, a city government, and an automotive manufacturer, among other industries.6
Another emerging area is entity analytics, which focuses on sorting through data and grouping together data that relates to the same entity. As a simple example of the power of entity analytics, consider three customer records, two of which have no data in common, and a third that has a driver’s license number which is the same as that of customer record one as well as a credit card number which is the same as customer record two. These three records can be combined into one record that has more complete data than each of the three individual records. Entity analytics is a powerful technique for recognizing context and detecting like and related entities across large collections of data. Jeff Jonas, IBM Fellow and Chief Scientist, Context Computing, contends that the data enterprises have to deal with is growing so fast that enterprises are developing amnesia; that is, using the same techniques to cleanse and analyze data causes them to get behind because the amount of data is increasing so fast.7 Jonas uses a puzzle as a metaphor for organizations’ need to sort through data to make sense of it and develop context; he says that an ever-growing pile of puzzle pieces represents the ever-growing amount of data, and until you try to assemble the pieces of the puzzle or data, you do not know what you are dealing with.8 When teaching MBA students about big data and analytics, Maureen Norton and Emily Plachy include an exercise where the students work enthusiastically to put together hundreds of puzzle pieces using multiple tables, shouting out insights as they discover them.
Although entity analytics is helpful in finding relationships between pieces of data and can incorporate new data to either confirm or negate previously found relationships, additional methods can be applied to garner insight from unstructured data. Cognitive computing, computing systems that interact with people in new ways to provide insight and advice, is emerging just when we need it to help us uncover insight from the explosion of big data. Chapter 11, “Reflections and a Look to the Future,” describes this exciting new era of computing.
The most easily available source of data for analytics is an enterprise’s own internal transaction data. For our retail example, this includes inventory data, sales data, employee data, and promotion and advertising data. Increasingly, enterprises are augmenting their internal data with data from external sources, including social media data. Whereas enterprise transaction data tells the retailer what has been bought, social media data can give the retailer early insight into what customers intend to buy. With the adoption of social media tools such as Twitter and the growth of blogs and forums on the Internet, the amount of social media data is growing very large, and analyzing it is providing significant insights and benefits to businesses.
Big data has four dimensions, known as the four Vs:9
- Volume: The size of data, which can range from terabytes to petabytes of data
- Variety: The forms of data—structured, text, multimedia
- Velocity: The speed at which data is available and analysis of streaming data
- Veracity: Data quality—managing the reliability of data
Consumers who are active in the big data social media world are driving enterprises to create collaborative systems, known as systems of engagement.10 These new systems have led to the creation of engagement analytics to measure the value of engagement.11 Engagement analytics can be used to measure employee engagement and customer engagement. Systems of engagement and engagement analytics are covered in more detail in Chapter 5, “Enabling Analytics Through Information Technology.”
This book shows how IBM has solved a wide variety of business problems by leveraging analytics to elevate business results. The types of analytics described include predictive analytics, prescriptive analytics, social media analytics, and entity analytics. Some of the analytics is performed on big data. The use of analytics not only drives cost savings and revenue growth but also provides more accurate and timely information to improve decision making and reduce complexity, which helps better manage the business. Analytics gives you the ability to anticipate, and that’s very powerful. So what does this mean for you? Whether you are in business for yourself or within a large company or a nonprofit or a government entity or a classroom preparing for your future, knowledge of the possibilities that analytics opens up will give you a competitive advantage. Understanding the journey that IBM has taken will shine a light on where you can get value from big data and analytics and illuminate your path to success in business.