Executive Edge: How savvy execs make the most of data analytics
By PATRICK TAYLOR
When it comes to analytics, organizations can use data insights on a strategic as well as an operational level. The 2011 film “Moneyball,” based on the best-selling book by Michael Lewis, tells the story of how Major League Baseball’s Oakland Athletics, on a limited budget, compiled an outstanding team of players using deep data analysis to drive their team-building strategy. The team then furthered their use of data to make day-to-day playing decisions based on analytic insights. This combination of strategic and operational analysis led the A’s to an outstanding performance – making the MLB playoffs and nearly beating the Yankees – with one of the lowest payrolls in baseball.
Many businesses face the daunting challenge of building analytics programs within their organizations, yet become so wrapped up with the system and technology that they fail to realize the full value of the insights. Some kick-start the process focused entirely on the strategic value of the data. Others implement analytics on the operational side, using it to flag exceptions or identify anomalies in the way processes are followed, but never use the data to reveal massive, game-changing findings. The most successful businesses do both, just like the Oakland A’s. Whether knee-deep in a big data implementation or just starting to explore the options, companies should consider some tips, pitfalls and best practices for getting the maximum value from their data. A good way to start is to make data analytics decisions with eyes wide open about what is truly required for set-up, which tools are most effective for the organization, and how to maximize always-limited resources.
Note to Self: Data is Not Perfect
Anyone who has ever worked with data understands that no data set is ever “clean.” The situation becomes even more complicated when organizations are pulling data from multiple production applications. A few examples highlight the enormous, unavoidable challenges associated with data inconsistencies.
Consider an international company looking to identify fraud in offices worldwide. The company may start with a database of countries with the highest risk of corruption, and then evaluate transactions for those countries. In different production applications, countries may be noted in multiple different ways depending on the system, the purpose for which the information was captured, and the individual who entered the data. For example, South Korea may be entered as a standard two-letter abbreviation such as “KR” in one system, and specified in various other standard text formats such as “South Korea,” “Korea, South” or “Republic of Korea.”
Similar issues exist for person names. Taking the United States only as a simple case, generally names are straightforward with a first and last name such as “John Smith.” However, sometimes middle names are captured such as “John James Smith” or the names are entered in an alternate format such as “Smith, John.” In a simple text comparison, “John Smith,” “John James Smith” and “Smith, John” do not match; however, they could be the same person. It gets more complex internationally where people may use up to five or six name components. To accurately identify activities associated with a particular person, the analytics tool must be flexible and intelligent enough to allow for various name formats.
There are many possible solutions, such as normalizing names to remove special characters and standardize formats; breaking the names down into components and matching on various combinations of the name components (tokenizing); and cross referencing known alternate spellings into standardized names such as ISO country names. The important thing is to ensure that the analytics solution being used is capable of effectively handling variances. A brittle solution that only accommodates a single naming convention will likely have issues.
Cloud-based Analysis vs. Analytics
The benefits of moving to the cloud are widely recognized. Scalability, accessibility and expandable horsepower and storage provide resources precisely when and where they are needed. As a result, many companies are turning to cloud-based analytics: analytics tools available in the cloud. While cloud-based analytics solutions present all of the familiar benefits associated with the cloud, they still require the same data scientist prowess needed to power in-house analytics solutions. Statistical knowledge, business understanding and analytical savvy are all required to use cloud-based analytics programs to effectively bridge the gap between business questions and meaningful data insights.
Cloud-based analysis is a new breed of solutions that encompass much of the “heavy lifting” when it comes to analysis. With cloud-based analysis, domain expertise is resident in the solution. Rather than exclusively serving as an analytics tool in the cloud, cloud-based analysis also offers pre-configured analytic queries to apply to the data sets found in a given industry. Companies can upload their data, which is then analyzed using a series of vetted, tried and true statistical analyses and algorithms that instantly reveal actionable insights for that particular industry. These cloud-based analysis solutions are best suited for operational analysis where best practices and industry norms are the most relevant.
For example, in travel and expense management, companies rely on categorization of expenses to help classify and report for trending purposes, but also to prepare tax filings, which may include different deduction rates depending on the expenditure category. Employees may either inadvertently or purposefully misclassify expenses. Cloud-based analysis can analyze T&E expenses for miscategorization, frequent offenders and merchants associated with multiple misclassified expenses. With this insight, the company can investigate to determine if there is fraudulent activity taking place, if certain inappropriate merchants (i.e., dating services expensed as “meals”) should be blocked, or if a process or policy change needs to be implemented to guard against problems.
Cloud-based analysis puts available data to work immediately, asking key questions and delivering business-critical insights on day one. Minimal ramp-up time is required and enterprises can start seeing trends immediately. These new solutions enable companies to see immediate benefit from analytics and also avoid the lead-time and resources required to progress through the learning curve of which questions to ask, which queries to configure and how to deliver meaningful reports. Companies should consider if there are areas where cloud-based analysis can deliver immediate operational value, allowing analytics gurus to focus on “deep dive” strategic issues.
Account for Nuances of the Business
|Anyone who has ever worked with data understands that no data set is ever “clean.”|
While some expenses may be a red flag for most any business (i.e., dating services) beyond the most obvious examples, determining what kinds of transactions represent a possible risk for a particular company is a critical first step to ensuring the analytic reports delivered are valuable.
For every industry and every business, there are differences in what qualifies as “typical” or “atypical.” For example, a large invoice to a plumbing vendor may represent a red flag to a pharmaceutical company, but be quite typical for a construction company. Likewise, a $500 dinner expense at the Ritz in New York may not be uncommon for a company with all East Coast clients, but the same dinner expense in Robert’s Restaurant (part of the Scores strip club) may be a red flag. The type of business, number of daily transactions and specific situations combine to make each company different. Managers typically understand these exceptions and anomalies, but they may not come to mind when initiating an analytics program.
There are a couple of ways to capture and integrate this information. One is to start off with a questionnaire prior to implementing an analytics solution. A survey may queue managers to think of anomalies about their business. The following questions may encourage managers to think along the right lines:
• What types of vendors indicate possible risk for your business?
• Is there a typical size/number of transactions per week/month that are typical of your business?
• What policies/guidelines are in place that you typically find employees skirting to avoid hassle or make transactions easier? (For example, breaking expenses in half to avoid expense limits that require a long pre-approval process)
With the answers to these questions in mind, the analyst can gain a better understanding of what to be looking for, and perhaps more importantly, what not to be looking for in results.
Sometimes an even easier way to get to this information is for the analyst to deliver the first set of reports and then collect feedback in real time. Managers typically don’t understand statistical calculations, but they do understand well-delivered results and have a keen eye for identifying when something is amiss. Based on reactions to initial reports, the analyst can adjust the queries/algorithms to take into account the newly shared insights. For example, a retail analyst may identify that 75 percent more cash refunds for product were issued at register No. 4 than at any other register. This is a potential red flag for fraudulent returns perpetrated by the cash register operator. However, in looking at the report, the manager may know that to keep customers moving quickly through check-out, refunds are directed to the customer service desk (home to register No. 4), where these transactions are handled whenever possible to prevent delaying other customers. This operational policy needs to
be taken into account in the analysis so that the “normal” volume for register No. 4 refunds is appropriately adjusted.
By spending some time up-front and in the first few cycles of analysis to account for nuances in the business, analysts can set up much more valuable reports and avoid time and energy spent on mislabeled red flags.
Look Beyond Operational Analysis
Leveraging analytics for operational analysis is a great place to start due to the quick ROI and powerful insights yielded in a short time. However, as in the example of the Oakland A’s, the savviest organizations should use analytics for both operational and strategic insights. Once organizations become comfortable with operational analysis to deliver insights for better day-to-day decision-making, it is easy to fall into a pattern of contentment. However, once cloud-based analysis gets rolling, it should leave in-house talent with the bandwidth to explore strategic-level queries that could lead to the next “ah-ha” discovery that will reshape the business.
If cloud-based solutions can be leveraged for some day-to-day analysis, then analysts with true domain expertise can focus their energies on coming up with the next big discovery. Companies often know the questions they would like to have answered. Big, game-changer questions like: How can we know which past customers of one product are the most likely customers of a new product? Or, which new markets are the most potentially lucrative?
Data analytics hold the answers to these questions, but it often requires some lead time and many interim answers before arriving at the ultimate answer. It can take months or even years to investigate these questions. Therefore, companies should begin applying their analytic manpower to those big questions as soon and as efficiently as possible.
Strategic-level insights may also be conducted at different times of year and at different intervals than continuous monitoring. For example, at the end of the year, managers may be making strategic sourcing plans and may wish to identify vendors that cause the most problems over the last year, requiring a different kind of analysis of the data with comparison against a different baseline. Likewise, larger trends may require a comparison of a full year’s results over those of the last several years to identify operational challenges or sales trends.
Analysts also need to focus on delivering information in a consumable format that is understandable and usable by their “customers.” Sets of data in tables may not be as understandable to the typical business user as a chart or graph. A chart may also be enhanced with accompanying explanatory text. Delivering the information in a way that is too challenging may leave critical insights unaddressed.
It is when continuous monitoring is combined with strategic data analysis that companies fully realize the value of analytics. The Oakland A’s went beyond conventional baseball statistics like batting averages and stolen bases to perform much deeper, more rigorous statistical analysis to understand and select players. Then they sought to outperform their opponents at each and every game by using more tactical insights such as having batters take more pitches to tire the opposing pitcher. The combination led to astounding success. In the same way, savvy executives can outperform their competitors with a combination of strategic and continuous analytics.
Patrick Taylor is CEO and founder of Oversight Systems, a provider of business analytic software.He is a member of INFORMS.