Business intelligence provides companies with valuable historical information, keeping many organizations competitive during tough times.

Purchasing departments, for example, use BI to monitor, choose and negotiate with suppliers. Customer service departments use it to identify problems that can be fixed. And airlines use BI to monitor the status and performance of their fleets and personnel.

Though BI provides many advantages, it is limited in its ability to predict, forecast and make inferences on unknown facts and relationships – for instance, predicting customer behavior, the probability of fraud, or suggesting the next best offer during an online transaction. For these reasons, most companies are enhancing their BI practices to include predictive analytics and data mining. This combines the best of strategic reporting and basic forecasting with additional operational intelligence and decision-making functions. By developing the capability to move from insight to action, leading businesses are combining historical and predictive analysis to determine what immediate actions to take.

In today’s challenging economic environment, “good enough” is no longer satisfactory. In fraud analysis, for example, knowing what happened yesterday and stopping the same thing from happening in the future is only step one. By using advanced analytics, organizations now have the capability to identify fraud before they write a check, refund money or settle a claim.  

This seems simple enough, but even companies with sound enterprise data management practices, processes and infrastructures that were built for traditional BI reporting are not always positioned to effectively address the complex requirements and unpredictable workloads of operational analytics. In fact, existing practices can hinder the process. The good news is that much of the work that establishes enterprise-class BI – creating a single version of the truth, rigorous systems governance, and an investment in sophisticated data integration and data quality – can serve as a sound foundation for advanced analytics. However, the question often raised is how to use existing production platforms and maintain predictable performance and governance without limiting the ability of analysts to explore, transform, condition and develop data and models in an ad hoc fashion.

No matter how you slice it, advanced analytics is different. The discipline requires a demand-driven, forward-looking, flexible, exploratory process. Any attempt to limit these dimensions will inhibit the effectiveness of the analyst.

To address this challenge, today’s best-in-class enterprise data warehouse technologies allow practitioners to establish virtual or physical analytic “sandboxes.” These sandboxes comprise dedicated disk space and processing resources that allow analysts the freedom to do what they need. Regardless of data size or model complexity, adding sandbox capabilities to the corporate reference architecture gives IT the ability to dynamically provision part of the EDW for an individual line of business or analyst. This can be accomplished while still maintaining optimum, balanced and predictable performance for the rest of the enterprise.

Business requirements, not technology limitations, should determine how to set up a sandbox. Examples include analytics that are exploratory, cyclical or unpredictable in nature; have only a few power users with a narrow focus; or use models that are extremely resource-intensive. Whether implementing sandboxes within the production system or in a separate analytics development environment, the data used to feed them must be fresh and accurate, the systems supporting them must provide linear performance, and analysts must be able to explore at a granular level with minimal data duplication and movement. Most importantly, the sandbox must enable analysts to write and edit enterprise-wide data (and incorporate outside data if necessary) to create accurate, representative data sets.

Consider the example of a supermarket chain. For marketing analysts to optimize customer profitability, focus on customer retention or create the best sales promotional calendar, they need to take advantage of query-based data mining and exception analysis as well as advanced statistical modeling. Absence of thorough analysis can lead to inconsistent conclusions and incorrect operational decisions. Data must be granular and cross a number of subject areas: inventory, financial, customer, regional demographics, etc. Simply sending all customers a thank-you letter or offering all customers a special discount to remain loyal isn’t an effective use of money.

By leveraging each customer’s spending history and trends, statistical models can identify those customers who are most at risk. Customers who are not at risk can be targeted in a different way that is more appropriate for them. Such analysis allows delivery of more relevant offers to each customer, which will increase both immediate return on investment and long-term customer satisfaction.

Another reason for creating an analytical sandbox is to avoid advanced analytics in a vacuum. Many organizations deploy expensive analytics data marts based on silos of data focused on an individual department or line of business. Yet despite the investments, some companies can’t answer simple questions, such as “Which of our most valuable customers are likely to churn?” or “Which vendors provide the best price and assortment but do not drive profitability across their category?” That’s because they’re stuck working with limited or out-of-date information from the EDW. Let’s look at the problems this causes.

Casinos are very interested in identifying their best customers and rewarding them. One casino company decided it needed a 360-degree view of its customers’ spending and behavioral patterns across all properties and outlets. Achieving this prior to implementing an EDW was costly and time-consuming. Analysts spent more time gathering and moving the data from different properties than modeling it. The company consolidated the information in one EDW and gave access to its analysts to model the data in a separate sandbox. The result: The gaming company drove loyalty and wallet share up from 35 to 43 percent. In addition, the company can now identify gamers with the attributes to be particularly profitable customers and reward them the minute they walk into any of the company’s properties.

When analysis is done in an isolated fashion it also leads to the types of mistakes that drive customers away. A particular hotel chain uses marketing automation to identify customers likely to respond to promotion offers. The marketing campaigns succeed in driving incremental booking calls to the reservation desk. However, the marketing department doesn’t coordinate its data with the conventions division. If the promotional offers coincide with a weekend when several of the properties are booked with conventions, the reservations agents must then “walk” customers to competitive properties.

It’s possible the convention information would have been available to the marketing department if the company was using an EDW. But without the complete and up-to-date view of all subject areas associated with the models, a sound analytic conclusion suggested an incorrect business action. Unfortunately, in this case the insight was not synchronized with the appropriate operational action.

Business requirements, not technological limitations, should drive analytic architecture and infrastructure design. Having immediate access to the right data, in the right format, is no longer a goal – it’s an imperative. Development of an EDW to drive a best-in-class BI strategy is a big step in the right direction. Combining it with real-time operational business analytics is a huge leap in delivering game-changing differentiation, innovation and performance.

Look for part 2 of this series in next week’s InfoManagement Direct newsletter. The authors will discuss how rapid analytics saves money and improves customer satisfaction.

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access