Evan Stubbs, SAS

If life ever feels mundane, find some friendly local data scientists and ask them what they’re working on. For bonus points, ask them what they’d like to be working on.

Then, sit back and bask in the unadulterated glory of undiluted jargon. If they’re worth half their salt you’ll at least hear about big data, dark data, structured data, unstructured data, semi-structured data, and probably more types of data than you knew there were words for.

You know how they say Eskimos have over a hundred words to describe snow? That’s nothing compared to data scientists when they get going.

They’ll probably talk about decision trees, MapReduce, SQL, gradient boosting, support vector machines, supervised learning, machine learning, and lots of other things that sound really cool but also confuse the heck out of you. As far as mind-benders go, the next step is probably string theory.

Who doesn’t love a quark?

Facing a firehose of jargon, it’s easy to get lost in the noise. Let’s be honest; most of us weren’t that great in Algebra 101, let alone Intro to Calculus. Business analytics is hard, right?

Not really. I’ll let you in on what’s probably the best-kept secret in the discipline. It’s actually a two-fer:

1) The benchmark isn’t how smart something is. It’s whether or not it’s better than what you’re currently doing. There are only two ways to drive a better outcome. Make it more sophisticated or make it more timely. Cut through the confusion and it really is that simple.

Sophistication is a seductive temptress. It’s always possible to make things more targeted, more effective, or more efficient. In the long run, perfection isn’t just an ideal; eventually, it becomes the goal.

Unfortunately, in the long run we’re also dead. It’s better to get something working, even if it’s not perfect, than it is to hold out for the ideal solution. If the current benchmark is walking to the shops, don’t try to fly to the moon on day one. Just try to make a scooter. You’ll have more success: people like a working scooter more than they like an imaginary rocket. That takes us to the second rule…

2) Making things better only happens in two ways. Either make it smarter or make it faster.

If you treat all your customers exactly the same, go and stand in the corner. You’re not customer-centric; bad marketer!

Just split them, even if it’s based on simple rules. If you’ve already split your customers, try using analytical segmentation to better target. Just remember this: while the simple things usually make the biggest difference, it’s the complex things that eventually create competitive differentiation.

If you’re trying to flag fraud and you’re reviewing transactions on a weekly basis, your exposure is whatever you didn’t pick up in the last week. If you look at things daily, your exposure is limited to a day’s worth of transactions. If you review in real-time, you don’t have any exposure. The more timely you can make your decision-making, the better the outcome you’ll achieve.

And that’s pretty much it! Your data scientist certificate is now in the post. Just don’t try and get into an argument about the relative merits of Bayesian or frequentist statistics. You’ll regret it.

Besides, we all know frequentists rule.

Evan Stubbs is the author of The Value of Business Analytics, a book that explains why teams fail or succeed. His most recent book, Delivering Business Analytics explains the link between business analytics and competitive advantage, outlines the Data Scientist’s Code (a series of management principles that move organisations towards best practice), and provides solutions to twenty-four common business analytics problems.

David Loshin, President of Knowledge Integrity, Inc.

It would be unusual to suggest that some businesses are not continuously seeking better ways of increasing revenues, decreasing operational costs, and extending profitable customer relationships. A closer inspection of the popular approaches for achieving these goals centers on what could be called “common” analytics that are not specific to any particular industry. Some examples include:

• Customer profiling and segmentation – Divides the customer community into customer categories based on key variables as a way of developing predictive models for behavior analysis
• Customer/product affinity analysis- Examines which customer segments have affinities to specific products (or products within organized categories)
• Market basket analysis – Looks at predispositions to purchasing certain products at the same time

All of these are examples of analytical approaches to drive increased product sales via upselling, cross-selling, understanding customer price sensitivity, or through the purchase of product bundles with higher profit margins.

And while the ability to execute projects enabling these analyses has typically been reserved by only the largest organizations with the biggest analytics budgets, a combination of factors is increasingly enabling a much broader spectrum of companies to be able to benefit from analytics, including:

1. Data volumes: Not only are the volumes of data expanding, the rate of expansion of newly created digital content continues to increase.
2. Positive marketing: The information management industry has done a very good job in marketing the purported benefits of analytics, effectively generating a blossoming demand.
3. Feasibility: Larger organizations may have already had the resources to implement large-scale analytics programs, but with high-performance platforms deployed on collections of easily acquired commodity hardware components, the barrier to entry for implementing an analytics program has been significantly lowered.
4. Right-time delivery: As the time windows for responding to emerging opportunities continues to shrink, there is a growing appetite for close to real-time delivery of actionable knowledge to drive trustworthy decision-making.

The result is that a greater number of smaller organizations are seeking to employ more sophisticated analysis techniques over a broader variety of digital content that spans both structured and unstructured sources. For example, including these types of digital content among others:

• Structured data sets acquired either directly through the World Wide Web, or  through data aggregator vendors
• Social media data, such as the unstructured comments and posts streamed through Twitter or Facebook, among others
• Machine-generated data, such as periodic reading of smart energy meters installed across a residential network
• Mixed-format content, such as documents and web sites containing text, photo images, graphic images, video, etc.

Analytic applications such as customer profiling, segmentation, and classification can be greatly enhanced with data from a wider variety of source. But as the demand grows for applications incorporating different types of data sources, the data management environment must be able to scale with the size and complexity of the data, and not just from a strict throughput performance perspective. There must be processes for extracting entity data from an unstructured source, identifying that entity, and augmenting the entity’s profile with discovered or learned characteristic attributes.

All aspects of data utility have to be taken into account, aligned with the idea of data management as a “dial-tone” service. Enabling predictive analytics implies that data management services must meet a base level of expectations: the performance for data delivery must be predictable, the framework must provide trustworthy information, there must be ways to ensure that commonly-used terms are not confused by downstream reinterpretation, and that data and business rules can be effectively incorporated directly into developed applications as part of the system development life cycle.

Boyce Byerly, Ph.D., Cofounder and Chief Scientist of Capital Analytics

Collecting and measuring data for human capital can be fraught with debate about the process as well as how to read the results. I caught up with Boyce Byerly, Ph.D., Cofounder and Chief Scientist of Capital Analytics, on this subject:

What is your number one hurdle you have to jump when collecting data from various outlets?
Politics, far and away, is always the largest and least tractable problem. Political problems have ways of dragging out for weeks, and can delay everything else. The whole goal is to get the data analyst on your team talking to the data analyst where the information lives – those people are super great at what they do, and any problems can be solved pretty quickly, even thorny technical ones. Solving the political problems up front is a real art of setting up stakeholder workshops that get everyone excited about the common goals; we have people who are phenomenal at that working with us, and it makes everything easier.

What is your number one argument to the Correlation/Causation debate?
I don’t think you can ever say with absolute certainty that one thing causes another. Science and history are full of examples of mistaken beliefs that were upheld by lots of brilliant people for centuries until a new insight or new way of looking at a problem came along and changed everything. What you can do is marshal credible evidence, rule out as much as possible, to convince a reasonable person. Of the factors you need to consider for proof, prior performance is the most important factor to consider. How people were performing in the past is the best predictor of how they’re going to perform in the future, regardless of how good a particular training program might be, and it’s very likely there are performance differences that will affect what you’re trying to measure.

What is your one piece of advice for young professionals looking to enter the field of human capital analytics, specifically as a statistician or data analyst?
Remember the old piece of advice that “when you have a hammer, all your problems look like nails.” You will need a variety of tools, from advanced statistics, to customer surveys, to human resources data. Most problems can, and should, be solved in more than one way. You’ll not only understand the problem better if you go after it in more than one way, but you will also have more ways to communicate with your intended audience.

Boyce Byerly, Ph.D., is Cofounder and Chief Scientist of Capital Analytics. He is the co-author of Human Capital Analytics: How to Harness the Potential of Your Organization’s Greatest Asset (John Wiley & Sons, 2013).

Nevertheless, the real trick is how to uncover that value and filter the wheat from the chaff to make any sense out of the numerous tweets, Facebook posts, and other bits of information floating around the social ether. The path to “making sense” can be paved with technologies that are starting to come into vogue for businesses large and small – big data. However, it is not just the generic term “big data” that brings value, it is truly the processes that make up big data analytics that can deliver answers that can drive business processes.

It may not be so easy to see how big data analytics can garner value out of the minutia floating around in the ether of the internet. However, the elements that make up big data analytics has the muscle to plow through the twitterverse, the realm of Facebook, and numerous other worlds of social interaction sites. Afterall, big data analytics is all about mining information from huge piles of data.

Nevertheless, value is a subjective element – simply put, when mining for gold, you have to know what gold is to begin with. That said, the tools that make up big data analytics do have their limitations and a human touch is definitely needed to make sense out of all the noise. The trick is to know what to look for. Arguably, business marketing and product development has the most to gain from applying big data analysis to social data sources. Those business processes can benefit from customer sentiment analysis, product adoption trends and common complaints that float around in the ether.

That process can focus on mining social media data sets for key words, such as a product name or company name and then cross referencing all occurrences to that mention to create a sentiment index, which can lead to insights. While it may sound simple, the truth is it is anything but. Businesses will need to turn to data scientists and others to really make big data analytics work for them and, of course, educate themselves on how big data works.

Frank J. Ohlhorst is the author of Big Data Analytics: Turning Big Data into Big Money, John Wiley & Sons, 2012.

Today, the term big data can be loosely defined as the processing of extremely large data sets that contain structured and unstructured data, which is used to uncover previously hidden insights. Simply put, big data is all about obtaining knowledge. Nevertheless, the type of knowledge delivered from the analysis of big data counts for much more than traditional business knowledge – big data analytics has the ability to deliver predictions, insights and expose relationships that were once unrealized – but only if the analysis is done correctly.

Therein lies the biggest challenge with big data – doing it right! From a technological standpoint, many of the challenges associated with big data have been solved. After all, platforms such as Hadoop and applications such as MapReduce have mitigated the problems associated with the basic processing of large data sets, creating a different challenge – how does one derive value out of large amounts of data?

Deriving that value can be a tedious and troublesome process, relying on gut instincts and the overall availability to uncover information that has analytic application – a challenge that depends heavily on designing algorithms and batch processing to uncover the value. It all comes down to knowing what to look for, using an approach that starts with a simple question, “what am I looking for?”

For most businesses, that question evolves into something that aligns with growth – such as “how do we increase sales in the western states” or “how can we predict product demand.” Ultimately, it will be the questions that need answers that will drive the big data process.

 Frank J. Ohlhorst is the author of Big Data Analytics: Turning Big Data into Big Money.

David Loshin, President of Knowledge Integrity, Inc.

In the previous post, we examined the “new normal” for the CIO and the need to demonstrate more immediate value from data.  However, introducing analytical appliances or implementing Hadoop – while ignoring the critical information infrastructure aspects that support the collection and management of data – may lead to questionable results when the level of trust in the usability of the information can easily be challenged.

That means that increased attention to fundamental capabilities for information management must accompany any adoption of new technology. As opposed to implementing data management components on a project-by-project basis, the time has come to view information management as an organizational business imperative.

Business user expectations for data accessibility, availability, and quality are approaching the sustained need for standard services, like telephony and network access. This “dial-tone” approach to information management services establishes a baseline, enterprise-wide capability for data utility, and includes components for:

• Data integration – What used to be called extraction, transformation, and loading (ETL) has evolved beyond the original scope of data warehouse population to include the end-to-end mechanisms for data sharing, access, and delivery.
• Data federation and virtualization – The desire for real-time integrated analytics has ramped up the demand for high-speed data access to heterogeneous sources. Data federation enables semantically correct mappings across data assets and makes heterogeneous data access transparent to the end users. Virtualization smooths the delivery and presentation of federated as well as provides caching to make access times predictable.
• Event stream processing – With the desire to absorb data from numerous sources, the business may want to apply filters or trigger actions based on streaming data. Event stream processing provides the infrastructure to support these types of actions.
• Managed metadata – Merging a variety of data sources without a common agreement to definitions and meanings will always lead to confusion. Establishing a metadata management practice using the right components will help alleviate some of these concerns.
• Data quality management – Any business environment will be compromised without establishing a level of trust in the usability and quality of the data. Parsing, standardization, and cleansing all contribute to a predictable level of data quality.
• Data governance – These technologies enable inspection, monitoring, and reporting of compliance with data quality rules and policies. In addition, tools to enable data stewards to be alerted to data issues and to monitor the progress of remediation help operationalize the deployment of corporate data policies.

In my next post, we will look at the demand for analytics incorporating a wide variety of data sources, including social media data, machine-generated data, as well as the desire to extract information from mixed-format content (such as documents and web sites containing text, images, video, etc.). We will also examine how that demand drives the need for predictable and trustworthy information management.

Also by Ivan Oliveira, SAS Director, Advanced Analytics Research and Development and Emmanuel Pacheco, Gaming, Hospitality and Entertainment Sales Lead for SAS Canada.

“A dollar won is twice as sweet as a dollar earned,” said Paul Newman in The Color of Money. But with growing competition for the entertainment dollar, winning over customers has never been more challenging.

When it comes to slot floor planning, bad decisions can mean significant losses in customer loyalty and potential revenue. According to the Canadian Gaming Association, legalized gaming has nearly tripled in size since 1995, from $6.4 billion in gaming win to about $15.1 billion in 2010. Moreover, as other sources estimate up to 85% of casino revenue stems from slot operations, this is considered a vital component of the business.

When deciding which games to offer or replace, casinos may look at historic results and reason that games which were popular in the past will continue to be so in the future. Therein lies a missed opportunity. With mountains of invaluable customer data available, a growing number of casinos around the world are turning to advanced analytics to assist with slot floor planning—and it’s proving to be a winning bet.

Case study: SaskGaming’s slot floor optimization
SaskGaming has enjoyed positive revenue growth since opening Casino Regina in 1996 and Casino Moose Jaw in 2002. However, like others in the industry, its initial period of double digit growth eventually plateaued as it reached a more mature stage in its market cycle. Customer demand for slot machine play in particular seemed to be saturated.

In the fall of 2011, this thought had been on the mind of Elliott Daradich, SaskGaming’s Director of Slots for nearly 17 years. On board since the casino’s inception, Elliott witnessed the development of the business into an increasingly dynamic environment. With so many changes taking place at once, he wondered: How can one plan the right mix of gaming choices, denominations, and machine placements to optimize customer interest?

To answer this, SaskGaming paired with SAS Analytics to discuss options to assess and refine SaskGaming’s data needs to improve its longterm slot business planning process. Project heads from each organization adopted a multi-phased team approach that began in summer 2012 with a detailed test, or ‘proof of concept’.

SaskGaming determined its current business needs were beyond what had been envisioned when its slot databases were originally created. Therefore, the first step in the process was assessing and cleaning the available data to enable a detailed categorization of relevant information to gain insights into slot performance to date.

The data needed to be reviewed and revised for consistency to allow the history of similar games to be tracked. This was a key challenge of the initiative because results of analysis can only be as good as the information that gets analyzed.

“It took a lot longer than we anticipated; in the end there were five iterations,” recalls David Koch, Analytics Specialist with SaskGaming.

Next, the team used this information to provide a best-case predictive forecast into how each game would perform in the year to come. In the process, it became possible to begin collecting insight into leading predictors of guest preference that would optimize profitability while supporting the integrity of fair and random play.

“Our databases lacked details about the attributes of individual games and machines,” explains Daradich, adding, “Now that we’ve seen what can be learned from this kind of information, we plan to redevelop and augment our data capture with an eye toward future analysis capabilities.”

Leveraging categorization and each game vendor’s market research, it became possible to isolate a surrogate to help it analyze options for new game purchases. Moving forward, the technology will allow SaskGaming to predict the potential impact of changes on slot performance based on ‘what if ’ scenarios. This provides much greater forecasting power than the traditional approach to decision making, which was limited to reports based on one variable, looking exclusively at historical data.

Finally, advanced optimization was utilized to determine the best approach to future business, considering factors such as physical space and budget. This information will help SaskGaming optimize its slot purchase options, including the analysis of which machines to replace and when to replace them, while also ensuring player experience is not hampered through down-time.

“In the end, the solutions offered by the test case promise a new perspective,” says Daradich. “Better yet, we were there to take part and see it happen. Every Friday as we met as a team, we considered the results and next steps together. It gave me the opportunity to become comfortable with, and have confidence in, the outcome.”

“To me, what’s different about this approach is no one is trying to isolate the one magic variable that matters above all else to customers and will make the business thrive for years to come,” adds Koch. “Rather, we’re simply identifying those games which customers find appealing even if we don’t know why. This solution has helped us to immediately improve our understanding of customer preferences. As our databases become richer with new game and machine attributes, we’ll also have stronger predictors of long-term gaming trends performance, allowing us to make better decisions in the future.

Power in SaskGaming’s hands
Competition in the entertainment market can be fierce. SaskGaming recognizes that with more competition for discretionary entertainment spending by customers, it needs to offer guests an entertainment experience that exceeds their expectations. As such, being able to make empirically sound decisions about which slots will best appeal to customers is more important than ever. The analytics solution enables users to move seamlessly through the entire process, from data collection to forecasting, prescriptive optimization, and reporting.

“To have the power to reasonably forecast future results and do those ‘what if ’ scenarios at our convenience lets us make decisions about the timing and nature of machine replacements so as to achieve the most desirable business outcomes,” said Daradich.

An example for all
Using analytics, SaskGaming now has a new perspective on its games data, allowing the casino to use analytics to offer the right games, in the right locations to attract loyal and valuable customers. SaskGaming looks forward to considering the potential for slot floor analytics to improve its analysis capacity in other areas of its operations.

With flexible analytics solutions that can be implemented by any department at a casino from food and beverage to entertainment, it is easy for casinos to take a ‘sky’s the limit’ approach using analytics to transform the business planning process.

Originally published in Canadian Gaming Business, Spring 2013

David Loshin, President of Knowledge Integrity, Inc.

The concept of the “Chief Information Officer” (CIO) title has been well-established for many years. And although the role the CIO plays has slowly evolved in alignment with (and sometimes in reaction to) changes in the world of technology, dramatic shifts in the perception of the creation, use, and employment of information have somewhat skewed the direction that the CIO role has taken.

In the past, the main focus of information management was subsidiary to the execution of “business as usual,” typically framed within the development framework for applications that implement operational or transaction business processes. Here, the data acquired, created, modified, and used was solely intended to guarantee the proper completion of the process.

This allowed siloed business functions to develop the same or similar data models, interfaces, and functionality. In this context, the CIO’s main focus was system infrastructure – ensuring that the system (including processing engines, storage, and networking) was configured to meet business needs.

Trends and drivers for information infrastructure
However, organizations began to recognize that data sets (previously presumed to be byproducts of the operational environments) actually held significant value. When those data sets were collected and combined for reporting or analysis, their repurposing introduced new demands – and uncover new constraints – in the information infrastructure. There are a number of trends that are worth noting:

• Technology adaptation: Innovative technologies can disrupt the presumed information infrastructure needs, such as the explosive use of smart handheld mobile devices, which both generate and consume information. Some industries are particularly sensitive to technical changes, such as the energy industry’s adoption of new smart meters that generate exponentially more data than before.
• Big data and big data analytics: This trend only confirms the need, as more organizations seek to absorb larger volumes of data sets from varied sources and of varied structure.
• Integrated predictive analytics: The time gap for exploiting information is rapidly closing as organizations focus on competitiveness. Many organizations are tightly coupling their analytics engines to their operational systems to inform decision-making in real time.
• Management of auditable compliance: Whether one examines the result of the recent financial credit crisis, deregulation of industries, new laws enacted governing health care reform, or numerous other legislative initiatives, the implication is that demonstrating compliance with regulations requires access to historical data.
• Data governance: Increased reuse and repurposing of information, coupled with the expanded scope of information management, has highlighted the gaps in which the absence of defined and enforced data policies can impede the business. Operational data governance requires retooling the environment to enable inspection, monitoring, and reporting of data policy compliance.

Any one of these trends would imply the need for sound information management practices. However, the nexus of all the trends creates the impetus to institute the proper information management policies and infrastructure to capture, filter, and analyze data and turn it into knowledge that drives positive business results.

In the next post, we will examine the objectives for information management within a typical organization.

Panel at the SAS Financial Services Executive Summit

Jill Dyché, vice president of best practices at SAS, led the roundtable discussion, asking senior-level managers about ways to recruit and retain a talent base of data scientists, statisticians and other roles. The consensus? It’s not easy, but panelists said that it’s critical to demonstrate the challenges of the role and the type of business impact they can have.

During the hiring process, Halina Karachuk, vice president of innovation, research and analytics at AXA Equitable, stressed transparency about the role and the impact the candidate can have. “I have to be honest about the level of resources,” Karachuk said, explaning that her team had to stay “scrappy” when going up against bigger firms with more capabilities. “I have to find people who enjoy the scrappiness.”

Another tactic is to nurture and reinforce creativity within the analytics profession. Jessica Dunn, senior vice president of business intelligence and the analytics center of excellence at Bank of America, encourages team members to find new solutions to old problems.

“[Creativity] is a big thing on my team,” Dunn said. “We say, ‘Don’t think about what’s done today. Think about what’s possible.’ The key people on my team are the ones that can think outside of the box.”

John Brocklebank, a 30-year SAS veteran, discussed how the software company locates the resources that can solve complex business problems. As the vice president of SAS Solutions OnDemand, Brocklebank now directs a team that solves “Nobel Prize problems” with SAS technology. The challenging work keeps employees focused and nimble while building their skillset.

“We cover all industries,” Brocklebank said, noting that SAS Solutions OnDemand encompassed everything from clinical trials to fraud detection. “If people get tired of working on one project, they can move to another area. And you can often transpose the skills learned in one area to another.”

Not surprisingly, the participants covered the emerging role of the data scientist. The panelists agreed that there are great opportunities for data scientists, particularly for those that understand the intersection of business and IT. For managers, the requirement is to synthesize the many types of analytics capabilities into a strong working unit.

“At the end of the day, you have to get a group together with a variety of strengths and weaknesses,” Karachuk said. “It’s my job to figure those out and play to the strengths of the team.”