BI "Governments" for Healthcare


Published on

Healthcare organizations are unique business entities that present challenges for optimally organizing governance, people, and services for next-generation BI. Learning from other industries that have adopted the concept of the business intelligence competency center (BICC), this article explores the available options and evaluates which service and organizational model appears to best fit healthcare providers and similarly complex organizations.

Published in: Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

BI "Governments" for Healthcare

  1. 1. EXCLUSIVELY FOR TDWI PREMIUM MEMBERS Volume 17 • Number 4 • 4th Quarter 2012 The leading publication for business intelligence and data warehousing professionals The Necessary Skills for Advanced Analytics 4 Hugh J. Watson BI Dashboards the Agile Way 8 Paul DeSarra Best Practices for Turning Big Data into 17 Big Insights Jorge A. Lopez Implementing Dashboards for a Large 22 Business Community Doug Calhoun and Ramesh Srinivasan Data “Government” Models for Healthcare 34 Jason Oliveira BI Q&A: Gaming Companies on the 40 Bleeding Edge of Analytics Linda L. Briggs Offloading Analytics: Creating a 43 Performance-Based Data Solution John Santaferraro BI Experts’ Perspective: Mobile Apps 49 Timothy Leonard, William McKnight, John O’Brien, and Lyndsay Wise
  2. 2. BI Training Solutions: As Close as Your Conference Room We know you can’t always send people to training, especially in today’s economy. So TDWI Onsite Education brings the training to you. The same great instructors, the same great BI/DW education as a TDWI event—brought to your own conference room at an affordable rate. It’s just that easy. Your location, our instructors, your team. Contact Yvonne Baho at 978.582.7105 or for more information.
  3. 3. 1BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 volume 17 • number 4 3 From the Editor 4 The Necessary Skills for Advanced Analytics Hugh J. Watson 8 BI Dashboards the Agile Way Paul DeSarra 17 Best Practices for Turning Big Data into Big Insights Jorge A. Lopez 22 Implementing Dashboards for a Large Business Community Doug Calhoun and Ramesh Srinivasan 34 Data “Government” Models for Healthcare Jason Oliveira 40 BI Q&A: Gaming Companies on the Bleeding Edge of Analytics Linda L. Briggs 43 Offloading Analytics: Creating a Performance-Based Data Solution John Santaferraro 49 BI Experts’ Perspective: Mobile Apps Timothy Leonard, William McKnight, John O’Brien, and Lyndsay Wise 55 Intructions for Authors 56 BI StatShots
  4. 4. 2 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 volume 17 • number 4 EDITORIAL BOARD Editorial Director James E. Powell, TDWI Managing Editor Jennifer Agee, TDWI Senior Editor Hugh J. Watson, TDWI Fellow, University of Georgia Director, TDWI Research Philip Russom, TDWI Director, TDWI Research David Stodder, TDWI Associate Editors Barry Devlin, 9sight Consulting Mark Frolick, Xavier University Troy Hiltbrand, Idaho National Laboratory Claudia Imhoff, TDWI Fellow, Intelligent Solutions, Inc. Barbara Haley Wixom, TDWI Fellow, University of Virginia Advertising Sales: Scott Geissler,, 248.658.6365. List Rentals: 1105 Media, Inc., offers numerous e-mail, postal, and telemarketing lists targeting business intelligence and data warehousing professionals, as well as other high-tech markets. For more information, please contact our list manager, Merit Direct, at 914.368.1000 or Reprints: For single article reprints (in minimum quantities of 250–500), e-prints, plaques, and posters contact: PARS International, phone: 212.221.9595, e-mail:, © Copyright 2012 by 1105 Media, Inc. All rights reserved. Reproductions in whole or in part are prohibited except by written permission. Mail requests to “Permissions Editor,” c/o Business Intelligence Journal, 1201 Monster Road SW, Suite 250, Renton, WA 98057. The information in this journal has not undergone any formal testing by 1105 Media, Inc., and is distributed without any warranty expressed or implied. Implementation or use of any information contained herein is the reader’s sole responsibility. While the information has been reviewed for accuracy, there is no guarantee that the same or similar results may be achieved in all environments. Technical inaccuracies may result from printing errors, new developments in the industry, and/or changes or enhancements to either hardware or software components. Printed in the USA. [ISSN 1547-2825] Product and company names mentioned herein may be trademarks and/or registered trademarks of their respective companies. President Rich Zbylut Director, Online Products Melissa Parrish & Marketing Senior Graphic Designer Bill Grimmer President & Neal Vitale Chief Executive Officer Senior Vice President & Richard Vitale Chief Financial Officer Executive Vice President Michael J. Valenti Vice President, Finance Christopher M. Coates & Administration Vice President, Erik A. Lindgren Information Technology & Application Development Vice President, David F. Myers Event Operations Chairman of the Board Jeffrey S. Klein Reaching the Staff Staff may be reached via e-mail, telephone, fax, or mail. E-mail: To e-mail any member of the staff, please use the following form: Renton office (weekdays, 8:30 a.m.–5:00 p.m. PT) Telephone 425.277.9126; Fax 425.687.2842 1201 Monster Road SW, Suite 250, Renton, WA 98057 Corporate office (weekdays, 8:30 a.m.–5:30 p.m. PT) Telephone 818.814.5200; Fax 818.734.1522 9201 Oakdale Avenue, Suite 101, Chatsworth, CA 91311 Business Intelligence Journal (article submission inquiries) Jennifer Agee E-mail: TDWI Premium Membership (inquiries & changes of address) E-mail: 425.226.3053 Fax: 425.687.2842
  5. 5. 3BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Speed is on everyone’s mind these days. From real-time data to on-demand reporting, BI professionals want up-to-the-minute information and they want it now. The authors in this issue of the Business Intelligence Journal understand. Agile development methodologies have long promised speedier delivery of new applica- tions or features thanks to shorter development cycles and increased user collaboration. Paul DeSarra explains how an agile approach can be leveraged to meet the highly dynamic needs of business; he uses an agile dashboard project to illustrate his ideas. Dashboards are a quick and easy way to communicate key performance indicators, and Doug Calhoun and Ramesh Srinivasan provide tips and best practices for creating a successful dashboard design. An agile approach may also be what’s needed for mobile development at a maternity clothes maker, the subject of our Experts’ Perspective. Timothy Leonard, William McKnight, John O’Brien, and Lyndsay Wise offer their advice for getting mobile BI up and running quickly. Of the three leading characteristics of big data (the so-called 3 Vs: volume, variety, and velocity), it’s the speed component that is often cited as its downfall. How can you process so much data without becoming bogged down? Jorge A. Lopez describes one approach. John Santaferraro discusses how analytics must be offloaded to separate analytics databases if big data is to provide accelerated queries, faster batch processing, and immediate access to a robust analytics environment. Senior editor Hugh J. Watson notes that studies suggest enterprises will soon face a shortage of data scientists. He explains that we will have to give business analysts and data scientists wider and more in-depth permissions and provide more training for existing staff if we’re to keep up with current trends. Healthcare organizations face a variety of tough governance challenges. Jason Oliveira explores what can be learned from other governance and services organizations that have adopted business intelligence competency centers (BICCs) and how to apply that knowl- edge to help improve healthcare’s BI disciplines. Speed can present challenges, which is why our Q&A explores how gaming companies are on the bleeding edge of analytics, using real-time information to improve gameplay (as well as up-sell or cross-sell products or services to players). How are you keeping up? We welcome your feedback and comments; please send them to From the Editor
  6. 6. 4 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 The Necessary Skills for Advanced Analytics Hugh J. Watson Analytics work requires business domain knowledge, the ability to work with data, and modeling skills. Figure 1 identifies some of the skills in each area. The importance of particular skills and the exact forms they take depend on the user and the kind of analytics involved. Let’s take a closer look. It is useful to distinguish among business users, busi- ness analysts, and data scientists. Business users access analytics-related information and use descriptive analytics tools and applications in their work—reports, OLAP, dashboards/scorecards, and data visualization. They have extensive business domain knowledge and are probably familiar with the data they are accessing and using but have a limited need for and understanding of modeling. Advanced Analytics Hugh J. Watson is a C. Herman and Mary Virginia Terry Chair of Business Administration in the Terry College of Business at the University of Georgia. He is a Fellow of TDWI and senior editor of the Business Intelligence Journal. BUSINESS DOMAIN • Goals • Strategies • Processes • Decisions • Communication • of results DATA • Access • Integration • Transformation • Preparation MODELING • Methods, techniques, • and algorithms • Tools and products • Methodologies Figure 1. Skills needed for analytics.
  7. 7. 5BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Advanced Analytics Business analysts use analytical tools and applications to understand business conditions and drive business processes. Their job is to access and analyze data and to provide information to others in the organization. Most business analysts are located in the functional areas of a business (such as marketing) and perform analytical work (such as designing marketing campaigns), or they may work in a centralized analytics team that provides analytics organizationwide. Depending on their posi- tions, business analysts work with some combination of descriptive, predictive, and prescriptive analytics. They tend to have a good balance of business domain knowledge as well as data and modeling skills. The data scientist title is taking hold even though it sounds elitist (I’ve also heard the term data ninja). Data scientists typically have advanced training in multivari- ate statistics, artificial intelligence, machine learning, mathematical programming, and simulation. They perform predictive and prescriptive analytics and often hold advanced degrees, including Ph.D.s, in fields such as econometrics, statistics, mathematics, and management science. Companies don’t need many data scientists, but they come in handy for some advanced work. Data scientists often have limited business domain knowl- edge, the ability to handle data related to performing analytics (e.g., data transformations), and strong modeling skills. They often move from project to project and are paired with business users and business analysts so that necessary domain knowledge is included on the team. Most companies have moved along the BI/analytics maturity curve and now have business users who understand and can employ descriptive analytics and business analysts who can deliver descriptive and some predictive analytics. Interest is now focusing on the organizational capability to perform predictive and prescriptive (that is, advanced) analytics to answer why things happen and propose changes that will optimize performance. This explains why enterprises are employing more data scientists. Successful advanced analytics requires a high level of business domain, data, and modeling skills, and a team of people is often required to ensure that all of the skills and perspectives are in place. As an example, consider the following experience. Southwire: Bringing the Skills Together Several years ago, I received a call from a manager at Southwire, a leading producer of building, utility, industrial power, and telecommunications cable products and copper and aluminum rods. He wanted help solving an impending problem associated with the production of copper, a key component of many of his company’s products. My experience on that project (in particular, how the problem was approached and solved) provides a good example of the skills required to be successful with advanced analytics. I learned that the there is no set “formula” for manufac- turing copper. A variety of ores and other ingredients are used depending on what is available. The current approach involved an expert who would look at what materials were on hand and decide what and how much of each ingredient should be used. It was critical that the ingredients produced copper and that the copper would be viscous enough to flow out of the smelter and refining furnace. The problem was that the expert was retiring soon and his expertise was going to be lost. A new solution approach was needed. Southwire assembled a team of business people, chemical engineers, IT, and me. We had individuals with business knowledge, subject area experts, people who were familiar with available data and systems, and members with modeling expertise. The team contained all the skills needed for advanced analytics. My role was to provide the modeling (data scientist) skills. I saw two possible modeling approaches. The first option was to create an expert/rules-based system based on the knowledge of the retiring expert. We would capture in an application the heuristics that the expert used in deciding what to put into the smelter each day. The model would be descriptive in that it would describe the expert’s thinking.
  8. 8. 6 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 The other alternative, and the one chosen, was to use linear programming. If you are familiar with linear programming, Southwire’s problem was the classic production blending application. You create an objective function (that is, an equation) that you want to minimize with the sum of the cost of the various ingredients multiplied by the quantity of each ingredient. You also write constraint equations that reflect the conditions that the solution must satisfy. The output of the analysis is the quantity of each ingredient that will minimize costs while satisfying all of the constraints. The writing of the constraint equations was fascinating to me. Remember that the solution had to produce copper and it had to be sufficiently viscous. These requirements were handled through the constraint equations and reflected what ingredients were available and the chemi- cal reactions involved. The chemical engineers’ input was critical for developing these equations. Remember when you took chemistry in high school or college and studied valences (the number of bonds a given atom has formed, or can form, with one or more other atoms)? This and other factors (such as what ingredients were available each day) were incorporated into the constraint equations. Data scientists are not a “one-trick pony” when it comes to modeling. They are familiar with multiple modeling approaches and algorithms. They are able to identify and experiment with different models until they find the one that seems most appropriate. At Southwire, a linear Advanced Analytics programming modeling approach was selected over an expert/rules-based system. Once the objective function and constraint equations were developed, it was necessary for IT and me to select an appropriate linear programming package, enter the objective function and constraint equations, test the solution, develop a user interface that operational people could easily use for entering data (such as ingredients) and interpreting the output, implement the system, and train people to use it. Assembling the Skills Enterprises have the business domain knowledge for advanced analytics. However, as illustrated at Southwire, a key to success is to make sure that people with business domain knowledge are on the analytics team. Enterprises also have the required data skills, but a few changes may be necessary to accommodate their need for advanced analytics. Data scientists (and some business analysts) may need to have fewer restrictions on the data they can access and what they can do with it. They may need access to underlying data structures as well as the ability to join, transform, and aggregate data in ways necessary for their work. They may also need the ability to enter new data into the warehouse, such as from third-party demographic data sources. A possible solution to the potential conflict over control versus flexibility is an analytical sandbox, whether it is internal to the warehouse or hosted on a separate platform. Finding the required modeling skills is a trickier proposi- tion. You can hire consultants, as Southwire did, or use a third-party analytics provider, but these options can become costly over time if your plans include extensive advanced analytics. You can probably coach some of your current business analysts. There are many conferences (such as those offered by TDWI), short courses, and university offerings that teach advanced analytics. As advanced analytics becomes better integrated into application software (for example, campaign manage- ment) and easier to use, it is likely that trained business analysts can take on tasks that have skill requirements typically associated with data scientists. A possible solution to the potential conflict over control versus flexibility is an analytical sandbox, whether it is internal to the warehouse or hosted on a separate platform.
  9. 9. 7BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 You can also hire data scientists. This isn’t a new approach; many companies have already done so and have data scientists scattered throughout their business units or in specialized groups such as analytics compe- tency centers. Studies suggest that companies are planning to hire more data scientists and will face a shortage of such resources. For example, the McKinsey Global Institute predicts a shortfall of between 140,000 and 190,000 data scientists by 2018 (Manyika, et al, 2011). Many universities are gearing up to meet the need through degree programs, concentrations, and certificates. These offerings are usually in business, engineering, or statistics and the instructional delivery varies from on campus to online. One of the first and best-known programs is the Master of Science in Analytics at North Carolina State University. SAS has been an important contributor to the program, which is offered through the Institute for Advanced Analytics and has its own facility on campus. Deloitte Consulting has partnered with the Kelly School of Business at Indiana University to offer a certificate in business analytics for Deloitte’s professionals. Just this year, Northwestern University rolled out an online Master of Science in Predictive Analytics offered through its School of Continuing Studies. Will students take advantage of these programs in large enough numbers? Advanced analytics is a tough study, and many students may not have the necessary aptitude, inclination, and drive to complete the programs, even though the career opportunities are great. Summary You have already been performing analytics under the BI umbrella. BI includes descriptive analytics, and you have probably also been performing predictive analytics. For more advanced analytics, however, you will need to “ramp up your game” a little. You have the business domain knowledge covered. For the data component, you will need to grant business analysts and data scientists wider or more in-depth permissions and you will likely need to extend and enhance your analytical platforms (such as appliances and sandboxes). For the modeling Advanced Analytics skills, you will probably need to provide training for existing staff and bring in new people with specialized analytical skills. ■ Reference Manyika, James, Michael Chui, Brad Brown, Jacques Bughin, Richard Dobbs, Charles Roxburgh, and Angela Hung Byers [2011]. Big Data: The Next Frontier for Innovation, Competition, and Productivity, McKinsey Global Institute, May. technology_and_innovation/big_data_the_next_ frontier_for_innovation
  10. 10. 8 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development BI Dashboards the Agile Way Paul DeSarra Abstract Although the concept of agile software development has been around for more than 10 years, organizations only recently began to think about how this methodology can be applied to business intelligence (BI) and analytics. BI teams are continually evolving their rapid delivery of additional value through reporting, analytics, and dashboard solutions. These teams must also discover what types of BI solutions can reinvigorate a BI deployment and produce meaningful results. One way to reinvigorate BI deployments is to take the concept of agile software development and apply it to BI initiatives such as BI dashboard solutions, which can both re-engage the business and drive actionable intelligence and confident decision making. Agile BI replaces traditional BI project methods (heavy in documentation and process) with a user-driven approach. This article discusses an approach to building BI solutions and dashboards using an agile software development methodology. Introduction Although the concept of agile software development has been around for more than a decade, it’s only been in the last few years that organizations have started to examine how this methodology can be applied to business intel- ligence and analytics. The constantly changing, highly dynamic needs of business today have increased the demands on BI environments and teams. The pressure to be more organized, turn projects around faster, and ensure user adoption at all levels is increasing. Teams need to be able to react to demands from the business and proactively develop ideas and solutions that give the business more creative ways to think about how to use data. Leveraging an agile software methodology as it applies to business intelligence is a great way to meet these constantly changing business needs. Paul DeSarra is Inergex practice director for business intelligence and data warehousing. He has 15 years of BI strategy, development, and management experience working with enterprises.
  11. 11. 9BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development In a nutshell, using an agile software development meth- odology (“agile”) instead of a traditional development methodology allows end users to experience a version of the software product sooner. Instead of adhering to a strict and intensive requirements and design phase before development begins, agile employs a series of shorter development cycles to increase user collaboration. The agile approach welcomes changes during the development process to provide a better product that delivers measur- able value quickly and efficiently. There are four guiding principles for agile software devel- opment (according to the Manifesto for Agile Software Development, These can also be applied to business intelligence development efforts. Principle #1: Value individuals and interactions over processes and tools Traditional BI development focuses on strong processes and tools to solve development challenges. As a result, many organizations end up creating silos among the busi- ness and IT teams. Each team silo focuses on individual responsibilities and objectives and, in effect, each team loses sight of the overall project goal of providing cohesive and comprehensive data-driven solutions that improve performance levels. When using an agile BI approach, all those involved in the BI initiative work together as one team with one goal and set of objectives. To accomplish this, many organiza- tions create hybrid teams and a business intelligence competency center (BICC) composed of individuals with the necessary skills to define, architect, and deliver analytic solutions. In some cases, many of these teams are organized under business units outside of IT and the program management office. Principle #2: Value working software over comprehensive documentation Traditional BI development in a big-bang approach focuses on developing detailed documentation about common metrics, terminologies, processes, governance, support, business cases, and data warehouse architectures. Organizations may create a standardized enterprise data warehouse and then fail because they were focused on the documentation and lost touch with the business and the problems they were trying to solve. This doesn’t mean we should stop creating detailed docu- mentation. BI teams can and should continue to focus on creating documentation that emphasizes the vision and scope as well as the architecture for future support. With agile BI, the focus is not on solving every BI problem at once but rather on delivering pieces of BI functionality in manageable chunks via shorter development cycles and documenting each cycle as it happens. Principle #3: Value customer collaboration over contract negotiation Using an agile BI approach does not mean giving users an unlimited budget or tolerance for changes. Instead, users can review changes discussed in the last development cycle to ensure expectations and objectives are being met throughout the project. Traditional BI development teams use functional docu- mentation to discuss what the solution will look like and how it will operate. Such an “imagine this” method often leaves users to try and visualize what they believe the solution will become. The resulting subjective expecta- tions can quickly derail a BI project. In contrast, an agile methodology reviews progress during each development cycle using prototypes so that stakeholders and business users can see how the BI solution is expected to look and Agile employs a series of shorter development cycles to increase user collaboration. It welcomes changes during development to deliver measurable value quickly and efficiently.
  12. 12. 10 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development function. Prototypes put a visual “face” to the project by showing what data is available, how it will be used, and how it will be delivered. Principle #4: Value responding to change over following a plan With an agile methodology, traditional BI projects that focus on huge project and resource plans are replaced by shorter development cycles designed to better incor- porate changes and keep the project team focused and informed. For BI projects, changes should be expected and welcomed. When users see prototypes and gain a better understanding of what analytic capabilities and information are available, they are better able to com- municate how they could use that information to make improved business decisions. Such revelations and ideas only strengthen the final product. An agile BI project still uses a plan, but its plan is short, manageable, and coupled with a prototype users can see and experience. Changes are jointly reviewed with business sponsors, users, and IT professionals at every project stage. Example: An Agile Dashboard To better understand how this methodology can be used, let’s look at a real-world example of incorporating agile BI into a BI dashboard project for an executive sales team. The vice president of sales of a large manufacturing organization asked us to help his company gain better insight into its orders, shipments, and pipeline in order to hold the sales teams more accountable. Specifically, he wanted a dashboard that he and his executive team could use to meet accountability objectives. His vision for the dashboard was solid, and our role was to take that vision and boil it down to key metrics that would drive actions. After a few meetings with the vice president of sales and the IT sponsor (in this case, the IT director), we concluded that an agile BI dashboard project was the best approach. We ensured we had the needed sponsorship from both the business and IT teams. In addition, we confirmed the organization was using a BI tool that was capable of delivering the desired solution. We advanced the project using a hybrid approach to agile development, breaking the project into three phases to quickly and efficiently develop the scope, build prototypes, conduct reviews, develop the solution, and implement it quickly. Phase 1 This was the foundational phase for our project and focused on the third agile principle (“customer collabora- tion over contract negotiation”). Phase 1 should last no more than one week and involves identifying, at a high level, the scope of the BI dashboard to ensure that the executive sponsors are engaged and the internal teams are assembled. Phase 1 is essential because it is used to narrow the scope and prioritize what can be delivered in the set time frame. SCOPE PROTOTYPE STAKEHOLDER REVIEWS BUILD RELEASE 1 ... N Figure 1. The agile process. An agile BI project still uses a plan, but its plan is short, manageable, and coupled with a prototype users can see and experience.
  13. 13. 11BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development In the first week, we worked with IT and the vice president of sales to ensure that the team had the right people with the right skills who understood the project goals. We outlined roles and responsibili- ties, opportunity and vision, and the high-level scope—all standard practices for an agile BI project. We worked with the vice president of sales along with several key business users to identify the metrics of greatest value. We worked diligently to understand what metrics were needed and how they influenced business decisions. A dashboard metric isn’t enough; we strived to enable users to respond to each metric to achieve the best business results. For example, we examined what happened after the dashboard highlighted a large gap between what the customer relationship management (CRM) application identified as a sales opportunity and the revenue actually gener- ated. We asked questions about the process of capturing these opportunities in the CRM to better understand leading factors that could influence revenue. Delving into these questions ensured that we understood the full sales engagement process. We didn’t stop there. We identified about 10 metrics for invoicing, orders, shipments, and budgets across four different dimensions—business area, Figure 2. Dashboard prototype examples.
  14. 14. 12 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development product, customer, and date attributes. In our vision, the dashboard would allow the sales teams to focus more effectively on specific sales opportunities, better track budgets, and confidently predict and forecast sales throughout the year (and know where and how to make necessary adjustments). We held two meetings with the IT team to better understand the ability of the source systems to provide the data elements required. The Phase 1 deliverables included a high-level vision and scope document that clearly set the stage for the rest of the project. By quickly defining the vision and scope as well as establishing a short time frame, we removed one barrier (long contract negotiations and timelines) so we could focus on having the right people involved and the right team defined. Phase 1 was completed in one week. Phase 2 Phase 2 is where collaboration, rapid prototyping, whiteboard sessions, and interactive brainstorming take place. This phase applies three of the agile principles (“individuals and interactions,” “customer collaboration,” and “responding to change”). Phase 2 focuses on using prototyping methods in brainstorming sessions to quickly build and show business users how their ideas and needs are reflected in the proposed solution—sometimes iteratively and on the fly. The prototyping tool may be separate from your BI tool, but it must be able to demonstrate visual elements as well as drill-up, drill- down, and interactivity. This phase requires collaboration between the sponsors, key business users, and IT. A key benefit of this phase is that users “see” the data in action and will know whether the data is being presented in a way that effectively delivers the information they need. In fact, the process often gives users new ideas for using the information to make business decisions (see Figure 2). In Phase 1 we created our vision and scope, outlined the business problem, and understood the set of metrics and dimensions necessary to reach the desired outcome. We approached Phase 2 with two goals in mind: ■■ Collaborate with the vice president of sales and the sales teams to define the “look” of the BI dashboard and the data interactions required to populate it. ■■ Work with the IT team to determine the data components and further understand what could be accomplished and delivered by the project deadline. (The overall project length was seven weeks, so we had only six weeks left.) The collaboration sessions were held with the vice president of sales, several key business users, and individuals from the IT team. The meetings started as whiteboarding sessions. Once we completed the initial design, we built a prototype with phony (but business- sensible) data and set up daily meetings to review and refine our development cycles. In each session, we identified how and why metrics were to be used and outlined the decisions that would be made using the data. We evaluated different ways to display information so it would be most useful to users. We also mocked up the drill-through detail analysis and report- ing that would be available via the easy-to-understand dashboard and made sure only a single path led to the detail at each level. The resulting dashboard prototype The prototyping tool may be separate from your BI tool, but it must be able to demonstrate visual elements as well as drill-up, drill- down, and interactivity.
  15. 15. 13BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development had four quadrants, each of which was meant to answer a specific question: ■■ How are we performing today? ■■ Are we on plan and what is our updated forecast? ■■ Where are we winning and losing? ■■ Who and what is not profitable? The mockup took the form of charts, regional maps, and dynamic and color-coded lists. It also included detailed drill-through paths and report examples to help guide users in making decisions. For example, a user could click on a troubled region on the map that identified a large revenue gap based on forecasting and get details on current activity within that region as well as open opportunities and win/loss details. All in all, we held about 10 different business sessions and kept coming back the next day with a refined prototype to generate ideas. As a result, throughout the entire process, users were engaged, excited, and willing to participate in the sessions. They also felt confident that their needs were being addressed and their ideas and feedback were incorporated. We simultaneously worked on the data components to map the vision to the actual data sources. To do so, we had to remove several roadblocks and make some tough decisions as a team (IT and business) in order to meet our deadline. As the team forged ahead, we uncovered several items that needed to be worked through as quickly as possible. ■■ A few financial metrics were not in the current ERP but would be implemented in an upgraded version, which was set to go live the following year. We worked with the business to outline the metrics and ultimately decided to put them on hold so that we could con- tinue building the rest of the dashboard. ■■ There was a need to tie in a certain product category captured in a separate data source outside the ERP. The product category was required to ensure we were capturing the full picture. This product was set to be coded in the new ERP. We decided to pull in and map this information from the separate data source and also put in place a process to map it into the new ERP when the time was right. We uncovered more than 15 potential roadblocks to the initiative, and we worked through them all with the team. We kept everyone informed and made joint IT/business decisions to move forward—accomplishing this with daily status meetings with IT and the business subject matter experts to address issues quickly and outline resolutions. Sponsors and stakeholders were also part of weekly checkpoint meetings. After we removed all our technical and business road- blocks, we completed Phase 2 and delivered the prototype dashboard, drill-through mockups, and a “Lean Require- ments” document that captured the requirements and outlined the assumptions and decisions we had made. We also built a “Lean Design” document that described the database design, data mapping, reporting designs, and ETL construct. Phase 2 was completed in four weeks. Phase 3 Phase 3 is the “build” phase and applies the second agile principle (“working software”). The foundation has been Users felt confident that their needs were being addressed and their ideas and feedback were incorporated.
  16. 16. 14 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development set, the scope has been refined to ensure rapid delivery, IT and business are fully engaged, and now the time has come to take the prototype and construct it within the BI environment. Phase 3 should take no more than a few weeks and involves building integration and ETL procedures, security, and the BI solution itself. At this point, with two weeks left, we began to build the required dashboard, drill-through reports, and supporting data layers. Building everything in dynamic prototypes made it much easier to ensure expectations were in line as development progressed. During this phase, we continued to show the results of actual develop- ment of the dashboard every two days to the business sponsor and key users. Throughout this process, changes were still submitted. We reviewed all changes and put them into one of two buckets—implement or put on hold—and made notes in our change control log. Some of the change requests that flowed through in this phase revolved around adding different relative time-period buckets for revenue and margin analysis, some minor layout changes, and three changes that were put on hold for future phases around customizing various alternate drill paths from the dashboard based on a user’s business unit and region. Phase 3 was completed in two weeks. Tips for Agile BI Success In the end, the initial phase of the dashboard was released in seven weeks. The project was a success because of the agile BI processes applied to every aspect of the project. One of the core success factors was the use of prototypes and interactive sessions. Using prototypes enabled us to keep all players involved from the beginning and provided a forum to exchange ideas, discuss issues, and actually “see” the solution as its development progressed. After reading the case study, perhaps you are now think- ing, “Can organizations really implement these types of BI solutions in seven weeks?” You may be asking, “What about data governance, load procedures, ETL, business rules, capacity planning, and security maintenance?” The reality is that you must strike a balance when using the agile software development methodology for your BI ini- tiatives. The process walks the line of ensuring that you are building a solid foundation that has longevity, speed, and strength to weather the dynamic and demanding needs of the business. The following ideas and concepts can help you implement an agile BI process. Tip #1: Start small, think big When you begin to build an agile BI solution, it doesn’t matter if you have an enterprise data warehouse coupled with a large-scale, mega-vendor BI software stack or a small data mart managed with a niche tool. The key is to focus on the immediate business need and pain, then map that to the ultimate vision. Get the stakeholders to define and work with you to build out what it will look like. Once you have the vision, determine the best approach that completes the work quickly and keeps the long-term picture in mind. If you need to take shortcuts to get the work done, that’s fine, as long as everyone approves the shortcuts and you have a process in place to close the loop at a future point. For example, if you have an ERP application and you have to group some of your sales data into a customized dimension (instead of modifying the ERP source of records) in order to deliver the BI solution, then do so, but ensure that you get approval and that everyone understands the costs and benefits. The project was a success because of the agile BI processes applied to every aspect of the project. One of the core success factors was the use of prototypes and interactive sessions.
  17. 17. 15BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development Although you are building a specific solution, you can still take steps to ensure it is repeatable, scalable, and fits into your overall data architecture. For example, in our case study, we were building a specific BI dashboard solution that was focused on shipments, orders, and pipeline processes specific to the sales functional area. However, in creating the solution, we built a data design that could scale outside of sales by building conformed dimensions and process-driven fact tables. If, for performance reasons, we had to create summary or aggregated tables to support specific business areas, we made sure these mapped back to lower-grain fact tables for data consistency and detail analysis. Tip #2: Remove the roadblocks Whether you face IT challenges or other obstacles, work systematically to overcome them. Typical roadblocks you may encounter in an agile BI project include: A narrowed scope. In some cases, it can be challenging to narrow the scope of a BI project so that a portion of the solution can be delivered in a shorter time frame. This is a slippery slope and requires the ability to prioritize and find common ground with business users and/or sponsors. If you can get the business sponsor to commit to a shorter time frame up front, it will be easier to narrow the scope. In addition, separate out the “must-haves” and the “would-like-to-haves” right away. Data gaps. In any BI project, data gaps are typical as users may not fully understand how information is collected or data anomalies are discovered. Agile BI is no differ- ent, and data profiling is a necessary step. In our case study example, we encountered data gaps that we had to eliminate or overcome by accepting risk, leaving out components, or implementing a temporary fix. Business commitment and time. Agile BI requires interaction with business stakeholders, sponsors, and users through- out the project’s life. Secure commitment up front with everyone and clearly outline the project’s benefits in terms of effective decision making. Managing expectations. There is often a gap in the expecta- tions about what it takes to deliver a BI solution and the time it actually takes. Users may believe that much more can be done in a short amount of time, which can cause extreme tension between IT and the business. Managing these expectations requires strong communication skills and an individual on the team who can effectively bridge IT and business users. This individual should understand data modeling and architecture, reporting and analytics, and dimensional concepts and be able to articulate the challenges to business managers and sponsors in a language they understand. Rogue development. Agile BI still follows a process and a method. There is still documentation and a plan; success metrics are still defined at the beginning of the project. Project management is still a core component in this process. We recommend that you still use the following tools, documentation, and processes to help guide the project: ■■ A vision and scope document is used to define initial, critical success factors and get project approval. ■■ A requirements document outlines the core busi- ness problems and key data elements, metrics, and dimensions that are needed for the BI solution. The difference from traditional BI development is that this document focuses on the smaller and shorter deliverables and keeps it “lean.” If you need to take shortcuts to get the work done, that’s fine, as long as everyone approves the shortcuts and you have a process in place to close the loop at a future point.
  18. 18. 16 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Agile Dashboard Development ■■ A design document describes the database design, data mapping, reporting designs, and ETL constructs. Again, different from a traditional BI project, this design should focus on bringing the technical team together on the architecture and for future support without getting lost in too many details. ■■ A project baseline plan for delivering a piece of functionality quickly, with the longer-term plan represented at a higher level. ■■ A change control log to track which changes are implemented and which are put on hold. ■■ An enhancement log to track enhancements that the team is unable to fit into the first release. If you have obtained the right sponsorship at the start and ensured everyone has the same vision and under- stands the project, your ability to remove roadblocks will be improved. Inevitably, however, challenges will arise, so always keep one eye on the vision and one on the scope. Tip #3: Engage the business BI professionals sometimes get so focused on the technology that even after the initial meeting with business users they may flip back to thinking mostly about the tools and technology rather than the business’s pains, needs, and objectives. In our case study, we used rapid prototyping and whiteboarding sessions to gather requirements and keep the right people involved and working in unison. We had daily brainstorming sessions to promote collaboration on the design, discuss the metrics and information needed to make business decisions, and show the BI dashboard prototype progress. From this, we built a requirements document that was focused on the key metrics and data elements, and we incorporated visuals from our prototype to ensure we had everything captured. Once we completed this phase and moved to the full development stage, keeping the key users involved continued to be highly important. During development, we still met at least twice a week to review progress and update our change control logs as we showed progress on the BI dashboard solution. The prototype became our guide to ensuring the development was on course and meeting all expectations. Summary As business becomes more dynamic and social in nature, BI environments need to be prepared to move fast and deliver value in creative ways. Intertwining BI best practices with the agile software methodology is one way to infuse speed, creativity, commitment, and value into any BI project. ■
  19. 19. 17BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 big data insights Best Practices for Turning Big Data into Big Insights Jorge A. Lopez Abstract Big data is surfacing from a variety of sources—from transaction growth and increases in machine-generated information to social media input and organic business growth. What does an enterprise need to do to maximize the benefits of this big data? In this article, we examine several best practices that can help big data make a difference. We discuss the role that extract, transform, and load (ETL) plays in transforming big data into useful data. We also discuss how it can help address the scal- ability and ease-of-use challenges of Hadoop environments. Introduction Growing data volumes are not a new problem. In fact, big data has always been an issue. Fifty years ago, “big data” was someone with a ledger recording inventory; more recently, it was a bank’s mainframe processing customer transactional data. Today, new technologies enable the creation of both machine- and user-generated data at unprecedented speeds. With the growing use of smartphones and social networks, among other technolo- gies, IDC estimates that digital data will grow to 35 zettabytes by 2020 (IDC, 2011). These new technologies have turned big data into a mainstream problem. In fact, it’s not uncommon to see small and midsize organizations with just a few hundred employees struggling to keep up with growing data volumes and shrinking batch windows, just as large enterprises do. The viability of many businesses will depend on their ability to transform all this data into competitive insights. According to McKinsey (Manyika, et al, 2011), big data presents opportunities to drive innovation, improve productivity, enhance customer satisfaction, and increase profit margins. Although many CIOs and CEOs recognize the value of big data, they have struggled Jorge A. Lopez is senior manager, data integration, for Syncsort Incorporated.
  20. 20. 18 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 big data insights (through no fault of their own) to handle this new influx of data. The problem isn’t information overload; it’s the failure to harness, prioritize, and understand the data flowing in. This is why data integration is a critical—yet often overlooked—step in the big data analytics strategy. Traditional IT approaches will not generate the results businesses expect in this era of big data. Therefore, IT organizations should look at the hype around big data as an opportunity to set a new strategy for harnessing their data to improve business outcomes. As a first step, organizations must examine their existing data strategies and ask: Are these data strategies helping us achieve the objectives of the business? Can our environment eco- nomically scale to support the requirements of big data? Can our infrastructure quickly adapt to new demands for information? To fully take advantage of new sources of information, organizations must cut through the buzz that big data creates. There are many definitions of big data, but most experts agree on three fundamental characteristics: volume, velocity, and variety. Another key aspect, often overlooked, is cost. Forrester, for instance, defines big data as “techniques and technologies that make handling data at extreme scale affordable” (Hopkins and Evelson, 2011). This touches on two critical areas that must be addressed to have a successful data management strategy: scalability and cost effectiveness. To scale data volumes 10, 50, or 100 times requires new architectural approaches to the data integration process. Doing so in a cost-effective way has been the biggest challenge to date for organizations. No matter what kind of IT environment you have or how you label your data (big or small), there are steps you can take to rearchitect and optimize your approach to data management, such as returning your attention to the data integration process in your quest for improved business insights. Not All Big Data Is Important Data Sometimes it’s easy to get caught up in the hype about big data. However, trying to process larger data volumes can significantly increase the amount of noise, hindering your ability to uncover valuable insights. It’s important to remember that not all data is created equal. Any big data strategy must include ways to efficiently and effectively process the required data while filtering out the noise. Data integration tools play a key role in filtering out the unnecessary data early in the process to make data sets more manageable and, ultimately, load only relevant data into the appropriate environment for analysis (whether that is a data warehouse, Hadoop, or an appliance). Organizations can take three approaches: 1. Define a clear data strategy that identifies the users’ data requirements. (Why do I need this data? How will it help me accomplish my business objectives?) 2. Build an efficient data model that is adequate to the demands of the business. 3. Have the right data integration tools to do the job. Ultimately, the data integration tool is the critical component; it can help materialize the strategy and execute on it to build an efficient data model. The tool must have the right capabilities as well as the scalability and performance required to work effec- tively. A key component is the ease of use that allows developers to focus on business requirements instead of worrying about performance, scalability, and cost. To scale data volumes 10, 50, or 100 times requires new architectural approaches to the data integration process. Doing so in a cost- effective way has been the biggest challenge to date for organizations.
  21. 21. 19BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 big data insights Bringing Data Transformations Back to the ETL Layer Data integration and ETL tools have historically focused on expanding functionality. For instance, ETL was origi- nally conceived as a means to extract data from multiple sources, transform it to make it consumable (by sorting, joining, and aggregating the data), and ultimately load and store it within a data warehouse. However, in today’s era of big data, this strategy neglects two critical success factors: ease of use and high performance at scale. As IT organizations confront the accelerating volume, variety, and velocity of data by applying analytics, they have been forced to turn to costly and inefficient workarounds, such as hand-coded solutions or pushing transformations into the database, to overcome their performance challenges. The costs of such scaling approaches can outweigh their benefits. The best example is staging data when joining heterogeneous data sources. This practice alone increases the complexity of data integration environments and adds millions of dollars a year in database costs just to keep the lights on. As such, an Enterprise Strategy Group survey (Gahm, et al, 2011) found “data integration complexity” cited as the number one data analytics challenge. There are new approaches that don’t require big budgets, however. To rectify this situation, we recommend bringing all data transformations back into a high-performance, in-memory ETL engine. This approach tackles four main issues: 1. Think about performance in strategic, rather than tactical, terms. This requires a proactive, not reactive, approach. Performance and scalability should be at the core of any decision throughout the entire development cycle, from inception and evaluation to development and ongoing maintenance. Organizations must attack the root of the problem with approaches that are specifically designed for performance. 2. Organizations must improve the efficiency of the data integration architecture by optimizing hardware resource utilization to minimize infrastructure costs and complexity. 3. Productivity gains can be achieved through self- optimization techniques, which means that little, if any, manual tuning of data transformations should be required. The constant tuning of databases can consume so many hours and resources that it actually hinders the business. 4. Cost savings are realized by eliminating the data staging environment, resulting in server and database maintenance cost savings; deferring large infrastruc- ture investments with the efficient use of system resources; and gaining improved developer productiv- ity because a considerable amount of time need not be spent tuning for growing data volumes, thus providing more time for strategic projects. The high-performance ETL approach should accelerate existing data integration environments where organiza- tions have already made significant investments and enhance emerging big data frameworks such as Hadoop. IT departments within several companies have initiated high-performance ETL projects to achieve a long-term, sustainable solution to their data integration challenges: ■■ A storage industry pioneer and leading producer of high-performance hard drives and solid-state drives needed to improve its assurance process and inven- tory management with faster data processing of one million data records from its manufacturing plants. Using a high-performance ETL strategy, the company has reduced data processing times from 5.5 hours to 12 minutes and has released 70 percent of its data warehousing capacity to devote to analytics. ■■ An independent provider of risk assessment and decision analytics expertise to the global healthcare industry needed to process and analyze 40–50 TB of claims data per month to uncover risk-mitigation opportunities. Through a similar approach, the healthcare analytics organization reduced processing
  22. 22. 20 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 big data insights from 2 hours to 2.5 minutes. Further business growth could also be supported by reducing turnaround time for new customers being entered into the system from 5 days to 24 hours. ■■ A mobile advertising platform company needed to quickly analyze large volumes of online activity data (such as views, clicks, and conversion rates) that was doubling every year in order to make important decisions (such as what ad space to bid on and when and where to place ads for customers). The business went from waiting one hour to obtain the information needed to adjust advertising campaigns down to 10 minutes. In addition, its two developers, who spent most of their time just maintaining the infrastructure, can now work on more valuable projects. The benefits of a proper ETL process with fast, efficient, simple, and cost-effective data integration translate into benefits across the entire organization, including opera- tional, financial, and business gains, with the ultimate benefit being quicker access to cleaner, more relevant data to drive big data insights and optimize decision making. Optimize Your Hadoop Environment In today’s world of mobile devices, social networks, and online data, organizations must massively scale data integration and analytics differently than before. Accord- ing to Forrester (2011), despite the opportunity new data presents, organizations use only a small fraction of the data available to them. A new architecture is necessary to change both performance and costs that are driving Hadoop, the open source framework for big data. Hadoop is designed to manage and process large data vol- umes. It presents several opportunities but also introduces challenges—including scalability and ease of use—that lead to siloed deployments with limited functionality, which is why Hadoop doesn’t provide significant value by itself. Organizations should not expect to rely solely on Hadoop for all their needs; other tools and platforms need to complement Hadoop to optimize the data management environment for these data sets. Hadoop gets its scalability by deploying a significant number of commodity servers. This way, the Hadoop framework can distribute the work among servers for increased performance at scale. Of course, adding commodity hardware running open source software looks like a more cost-effective proposition than adding nodes to a high-end, proprietary database appliance. However, the hardware required to cope with growing data volumes and performance service-level agreements can grow significantly. Therefore, it is not uncommon to find Hadoop deployments with a significant number of nodes. This elevates capital and operational costs because of hardware maintenance, cooling, power, and data center expenses. In addition, the required tuning involves hundreds of configurable parameters, making it difficult to achieve optimum performance. Such increased complexity is tied to ease of use, which is one of the major challenges facing nearly every organization working with Hadoop. Hadoop is not easy to develop. For instance, adding new capabilities (such as reverse sorting) and coding MapReduce jobs is typically performed manually, which requires specific skills that are expensive and difficult to find. For many organiza- tions, finding the skill set needed to manage Hadoop is the most significant barrier to Hadoop adoption. Organizations can overcome these challenges and extend Hadoop’s capabilities, maximize its value, and simplify the overall Hadoop experience by integrating the high-performance ETL approach. This approach allows for sorting and organizing the data before it is pushed into the Hadoop environment by leveraging Hadoop Distributed File System (HDFS) connectivity and by creating MapReduce jobs in a separate graphical interface rather than writing Java or Pig scripts. Data integration comes into play after analysis as well; the results of the analyzed data need to be reintegrated into other informa- tion systems. For example, comScore, a global digital information provider of online consumer behavior insights, saw its data volume increase 72 times per day within two years and deployed a Hadoop cluster to better manage the data processing.
  23. 23. 21BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 big data insights However, it is challenging to bring Hadoop into an enterprise with heterogeneous operating systems. Moreover, Hadoop lacks critical features such as real-time integration and robust high availability. Therefore, com- Score deployed a data integration strategy that groups and splits larger data files that fit more perfectly into Hadoop, which provides a higher rate of parallelism on compressed files and reduces disk costs for the Hadoop cluster. This saved 75 TB of data storage per month and slashed processing time from 48 hours to just 6 hours, so comScore can now process twice the data each month (compared to a year before), allowing the company to provide its customers data insights faster. Summary Today’s enterprises must make sense of the increasing volume, velocity, and variety of data while maintaining cost and operational efficiencies. Your business intelligence strategy must focus on optimizing the data integration process so it is fast, efficient, simple, and cost effective. This means ensuring you have all the right data at your fingertips by managing the volume and new sources of data, coupled with scalability as data requirements evolve. Quicker access to cleaner, more relevant data is what drives big data insights and what will truly lead your enterprise to faster and more profitable decisions. ■ References Gahm, Jennifer, Bill Lundell, and John McKnight [2011]. “The Impact of Big Data on Data Analytics,” Enterprise Strategy Group, research report, September. report-the-impact-of-big-data-on-data-analytics/ Hopkins, Brian, and Boris Evelson [2011]. “Expand Your Digital Horizon With Big Data,” Forrester, research report, September. +Your+Digital+Horizon+With+Big+Data/fulltext/-/E- RES60751?objectid=RES60751 IDC [2011]. “The 2011 Digital Universe Study: Extracting Value from Chaos,” digital iView report, June. microsites/emc-digital-universe-2011/index.htm Manyika, James, Michael Chui, Brad Brown, Jacques Bughin, Richard Dobbs, Charles Roxburgh, and Angela Hung Byers [2011]. Big Data: The Next Frontier for Innovation, Competition, and Productivity, McKinsey Global Institute, May. technology_and_innovation/big_data_the_next_ frontier_for_innovation
  24. 24. 22 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 practical dashboard development Implementing Dashboards for a Large Business Community A Practical Guide for the Dashboard Development Process Doug Calhoun and Ramesh Srinivasan Abstract Dashboards are becoming more prevalent as business intelli- gence tools, and the reason is obvious: well-designed, accurate dashboards can quickly communicate important business indicators and trends and provide actionable information. However, creating and implementing a successful dashboard involves a great amount of work. It often requires implementing tight controls while allowing the flexibility needed to test and learn with the business. This article outlines tips for how to integrate these seemingly divergent processes as well as how to ensure the data accuracy, ease of use, and optimal performance that make a truly successful dashboard. Introduction The use of dashboards as a primary business intelligence tool is expanding quickly. When supporting a business unit fueled by data, how does an application team build dashboards that will provide great business value and be sustainable? There are many methods for doing this, as we will explain in this article. However, there are also certain fundamen- tal principles that may seem obvious, but can be difficult to implement: ■■ Engage business users, not just at the beginning and end of a project, but throughout the entire process. Make business users your partners. Doug Calhoun is systems analyst, claims technology—data and information delivery at Allstate Insurance Company. Ramesh Srinivasan is manager, claims technology—data and information delivery at Allstate Insurance Company.
  25. 25. 23BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 practical dashboard development ■■ Involve the entire application team throughout your project’s life. A factory-like approach of handing off tasks from phase to phase will not work well. ■■ Although design updates may require an iterative approach with business users, the number of compo- nents needed should drive the team to define phases and key deliverables early in your project to keep it on track. ■■ Sophisticated user interfaces are great, but in the end, it’s really about the data. Ensure that everyone is in agreement about how to define the data from a business point of view, and create a plan for how to validate it. ■■ Ease of use is critical. Make sure your business part- ners get hands-on opportunities as often as possible. ■■ Design your technology based on the number and types of users. Performance and capacity should be considered when designing and building dashboards, much as they are with more traditional transactional systems. This article is not intended to serve as a guide to visual design. That topic has already been studied extensively and successfully. We will discuss best practices for the process of creating a successful design. In addition, the word dashboard is used here as a general term for data visualization tools showing at-a-glance trends and other indicators. It is not meant to signify the timing or refresh rate of the data, and could be used interchangeably with “scorecard” depending on how a business unit chooses to define it. In the business intelligence world, “dashboard” has become the most common term, so it will be used here with assumed broader connotations. Another concern is process methodology. Many compa- nies primarily employ a waterfall life cycle, which can be a difficult fit for a business intelligence implementation. However, a purely agile methodology for dashboards can also lead to trouble, as there are complexities with development and testing that require a certain level of more traditional phase-gating. Essentially, the dashboard needs to be treated both as an application (with all the functional testing required) as well as a mechanism for providing data, including iterative testing and prototype updates. A certain level of flexibility in your development process may be required to achieve a happy medium and ensure a successful rollout. Depending on the size of your company, you may also need to leverage the assistance of other technology groups to implement. Where appropriate, involve groups such as your business intelligence community of practice or center of expertise; data, solution, and/or BI architecture; database administrators; all associated infrastructure/ server administrators; change and release coordinators; and any other applicable groups you believe should be enlisted. Do this early. All of this may require “innovating your process,” which might sound like a contradiction in terms to process methodologists but may have practical application to your work. The best practices below will guide you in the direction that best fits your project’s needs. Starting the Project If you are embarking on a dashboard project for the first time, there are several rules of thumb you should follow at the project’s outset. First, as with any project, you will need to define team roles and lay the groundwork for how the project life cycle will work. At the same time, you will need to Sophisticated user interfaces are great, but in the end, it’s really about the data. Ensure that everyone is in agreement about how to define the data from a business point of view.
  26. 26. 24 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 identify and engage all stakeholders and ensure both groups agree on expected outcomes. It is unlikely that you will be working on a dashboard project without a business case behind it, but getting a request from the business and truly engaging users as partners are two very different things. Although it can be easy to take orders and make assumptions, keeping the key business partners involved throughout the entire project life cycle, and beyond, is absolutely essential. Finally, you will need some vital information before you begin. Some questions are obvious from a technical point of view. What are the data sources? Will data be stored separately? What tools will be used? What environment(s) must be built? Other questions are just as vital, but may not be so obvious. For example, is the project feasible, especially as an initial effort? We recommend you limit the scope of an initial dashboard to a simple, straightforward first effort that has high business value. This way, a quick win is more possible, success can be attained early, and business trust will be earned as you “learn the ropes.” You will also need to be sure that the project is appropri- ate for a dashboard or other visualizations. For example, if the business primarily wants to track how hundreds of their individual workers are performing, a dashboard is likely not the right vehicle. However, if they want to track how their offices are performing over a period of time, using standard, well-known measures within the company, then a dashboard may be the best option. (You can still consider getting to the individuals’ detail, which we’ll discuss shortly.) The main lesson here, and throughout the early phases of your project, is to ask questions and keep on asking them! If something does not make sense or seems impossible, work with business users until you reach a mutually satisfactory agreement. Once the project looks possible, list all your assump- tions—whether business related, technical, or process/ project based. You’ll need this list to build an order-of- magnitude estimate, define the technical space you will be working within, and help business users understand their role during the project (and how crucial it is). Having everything in order even before detailed require- ments are determined will give both you and business users confidence. After all, before you start involving them in detailed requirements meetings, they’re going to want some idea about when to expect a finished product. Finally, as you devise this plan, treat the dashboard as a full-blown application. Although the dashboard is built in the business intelligence space, it has both the complexity of a dynamic user interface (with the myriad possibilities of errors on click events), as well as the need for absolutely exact, gold-standard data. Both the data and the functionality will need to be tested thoroughly, as if you were developing a transactional application. If you release the slickest, most attractive dashboard your business has ever seen but the data is wrong or a button doesn’t work, user confidence will quickly erode. Your dashboard may be pretty—pretty meaningless. Consider the metrics and aggregations needed and what types of structures will be required to support your project. Depending on your company’s standards, you might be using denormalized tables, dimensional tables in a warehouse (or a combination of these), an integra- tion of detailed and aggregated data, OLAP cubes, or many other possible sources and targets. As with any BI solution, you need to choose the appropriate data model. Limit the scope of an initial dashboard to a simple, straightforward first effort that has high business value. practical dashboard development
  27. 27. 25BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 The point here is that performance is paramount for user adoption. Document and agree upon functional requirements and data definitions while offering the flexibility of iterative testing and tweaking that a business intelligence solution should provide. It is critical to lock down the logic behind the displayed metrics early in the project. If that changes or is vague to everyone, there is little chance you’ll deliver a successful dashboard. Gathering Requirements Involving business users in your work is crucial—and most clearly needed—early in a project, especially during requirements gathering and scoping. You may need to remind yourself to keep your business partners actively involved, because it’s vital to your project’s success! Multiple meetings will certainly be necessary, but make sure to keep users actively engaged via various methods, including whiteboarding at first, dashboard prototyping later, and sharing early data results. This will not only help hone the requirements, but also allow business users to feel they are truly partnering on the project. This will ensure that they know and trust what they will be getting. In addition, the entire development team should be involved from inception through implementation to ensure nothing gets lost in translation through the work. See Figure 1 for a gauge of both business and technical involvement through a general project life cycle (regard- less of the specific methodology used). The following best practices can help you avoid pitfalls during requirements gathering, even when the relation- ship with the business is good. Know your user. It is possible that your business partner may represent only one part of the larger group using the dashboard, or may be assigned to a project and may not be an ultimate end user at all. Some users may have different business needs from your primary business partner. Make sure that you define all the groups of users who will have access to the dashboard, and ensure all of their voices are heard. This is not as easy as it sounds, but is worth the effort. Scope Datarequirements Dataanddashboarddesign BuildandUIupdate Usertesting Implementandchangenav Architects/BI CoE/DBA Testing team Developers Analysts (technical) Business champion/SME Business sponsor Figure 1. Both business users and technical staff should be involved throughout a project’s life cycle. practical dashboard development Dashboard Implementation Effort by Role and Phase
  28. 28. 26 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Define a use case for every component you build. There is no point in creating a dashboard component unless there is a direct use for it that can be easily defined and documented. Documenting the requirements is crucial to ensure business users get what they have asked for and so developers and testers have a clear guide about what they must build. You want to ensure that the use cases, and the data shown, will stay meaningful over time for each component; it is not a good idea to introduce new or rarely used metrics with a dashboard solution. Finally, require sign-off for all use cases, business requirements, and scope documentation you create. The scope should be limited to the business metrics and granularity of the data at this stage; visualization requirements can be developed later. Know your data sources and plan your approach. You must understand both where the data initially resides and, if you use an extract-transform-load (ETL) or similar process, where it will eventually reside. If storing the data, you will need to know how it should be stored, how long the data will need to be available for access, and how often it needs to be refreshed. Especially if using ETL, three-quarters of your work will be spent on the analysis, load build and testing, and validation of the data. Even without ETL, our experience is that the majority of the time should be spent working with the data rather than building the front end. Given the visual nature of dash- boards, it is easy to assume that the bulk of your work is spent building attractive, user-friendly interfaces. This is simply not the case with successful implementations, especially when so many easy-to-implement dashboard tool suites are available. Include only trusted, known metrics whenever possible. Exceptions may arise, but if metrics are well known, the exceptions will be much easier to validate. The sources of the data must also be trusted, and business users should be included in selecting data sources. Know your refresh rate. Will the dashboard be loaded monthly, weekly, daily, hourly, or a combination of these frequencies? The fundamental dashboard design approach will depend on your answer to this question. Use cases will drive your design. Make sure you have thorough discussions about what is really needed versus what would be nice to have, because the more often the dashboard will be refreshed, the more support (and cost) it will require after rollout. Identify all filters and selections. The earlier in the project’s life you can do this, the better. This information has a major influence on your dashboard design and will affect decisions about performance and capacity. If a large combination of multi-select filters can be selected for one component, there will be a multitude of data combina- tions to validate and possibly many thousands of rows to be stored. Technologists can be tempted to impress their business partners, but be careful not to promise something that is not scalable or sustainable. Understand what levels of aggregation and detail are required. An early requirements exercise should involve the filters and dimensions that will be used as well as how they should be aggregated. Time periods are a common dimension as are office or geographical territories. On the flip side, sophisticated business users will inevitably want to know the details behind what is driving their trend or that one outlier metric. Not having a method of either drilling down to (or otherwise easily accessing) the detail behind the aggregation will frustrate users after the post-implementation honeymoon period has ended. Determining aggregation/detail needs should be part of the discussions during requirements gathering, but remember to balance your requirements with develop- ment difficulty and desired timelines. If detailed data is provided, it should be accessed directly via the dashboard, Define all the groups of users who will have access to the dashboard, and ensure all of their voices are heard. practical dashboard development
  29. 29. 27BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 whether through sub-reports or drill-down capabilities in the components themselves, depending on your tool set. Identify how much history you need. Some graphical trends will require year-over-year comparisons. Beyond that, it may be worth considering how long any data that no longer appears on the dashboard should be retained. If it does need to be retained for compliance or other purposes, an archival strategy should be considered, or possibly a view built on top of the base data. The more the dashboard can be limited to querying only the data it needs to display, the better it will perform. Define the data testing and validation process. It is never too early to address how you will ensure data quality through a validation process. Defining specific responsi- bilities and expectations, and what methods will be used for validation, should happen even before design. This will also ensure that business users will be ready when they are asked to begin testing. The validity of the data is the most critical factor in the dashboard’s success and continued use. Integrate business users. There are several ways to involve business users in requirements gathering and refinement besides letting them dictate while you take notes. These options include: ■■ Prototype early and often. Prototyping can start with simple whiteboard exercises, and many dashboard tools now lend themselves to quick prototyping so business users can see and play with something similar to the final product deliverable. This hands-on method is excellent for rooting out requirements gaps, although it should not replace formal documentation. ■■ Use real data wherever possible when prototyping to give business users a better context. It also helps you to identify and correct data issues early. ■■ Integrate developers. Requirements gathering should not be done solely by analysts. If there are separate individuals responsible for coding, they must be involved at this stage so they truly understand the value and meaning of what they will build. ■■ Set expectations for production support. Agree upon a process for communication of user questions or any defects users discover. Depending on the user, this can be done many ways, although users at the executive level will likely prefer a direct communication path with the team’s manager(s). Additional suggestions appear in the post-implementation section later in this article. ■■ Define milestone deliverables. Regardless of the software development methodology you use, defining milestone deliverables is critical for instilling and retaining business confidence in the project. It is also necessary to ensure the development team is progressing as expected. Milestone due dates should be communicated early and deadlines met. If a deadline is at risk of being missed, share this information (as well as the reasons for the problem and the recommended course of action) with business users so new dates and dead- lines can be agreed upon or so the team can remove items from the project scope or adjust resource levels and assignments. An early requirements exercise should involve the filters and dimensions that will be used as well as how they should be aggregated. practical dashboard development
  30. 30. 28 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 Required deliverables from the business requirements- gathering phase may include: ■■ Scope lockdown, with documentation of what is in scope and out of scope. ■■ Final prototype with business sign-off. (Note: This remains a working prototype, and all team members must understand and agree that the design may change later in the project if practical.) The highest- level sponsor of the project should be part of this sign-off, as well as further sign-offs of the actual product prior to rollout. ■■ Detailed requirements definitions, including images from the prototype. Such documents can tie the business definitions of the metrics to the way they will be displayed. Such a connection will bring clarity both to the business client and to the developers/analysts building the solution. ■■ Technical conceptual design. This high-level docu- ment defines all data sources and targets, what delivery mechanisms are being used, and the general business use case(s) for the dashboard. Designing and Building the Dashboard: Soup to Nuts When dashboard design has begun, all layers should be considered in relation to one other. For example, if the dashboard will be connected to aggregated tables designed for performance, those tables, the way they are loaded (or otherwise populated), and any performance and capacity concerns should be considered. This is just as important as designing the dashboard functionality. In general, the dashboard design should: ■■ Ensure a single, consistent view of the data. This can apply to the visual look and feel as well as how often the components on a screen are refreshed. The user should not have to think about how to interpret the dashboard; the data presentation should be clear and intuitive. ■■ Keep everything in one place. If detailed data or supple- mental reports are needed, use the dashboard like a portal or ensure a centralized interface keeps the data logically consolidated. Also, make sure the same data source is used for both detailed and aggregated data on the dashboard. Keep in mind, however, that business users may expect that a snapshot of the dashboard will not change. For example, a monthly metric could possibly vary slightly in the source data, but re-querying every time for the dashboard view with different results could erode confidence and even skew expected trends. Have a conversation with business users early on to discuss such scenarios and determine whether storing point- in-time dashboard snapshots will be required. ■■ Understand the usage scenario. Knowing the size of the user base, as well as the types of users and when they will be accessing the dashboard, can drive design. You should understand the usage volumetrics early in your project and plan accordingly. You must also ensure that any maintenance windows do not conflict with peak-time use. Environment sizing, capacity, and performance will all be critical to ensure a stable tool. ■■ Address multiple environments for development. If your environment has the necessary capacity, build develop- ment, test, and production environments. It’s worth it. Defining milestone deliverables is critical for instilling and retaining business confidence in the project. practical dashboard development
  31. 31. 29BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 ■■ Plan to validate data accuracy as early as possible, and ensure your design and project plan allow this. To avoid rework, it is crucial to make every effort to get the data perfect and acquire sign-off in a lower database environment during user testing. This will ensure that the data acquisition process is free of bugs. At the same time, ensure that you validate using data from the production source system(s), because the data will be well defined and likely have an independent method of validation. ■■ Roll out with historical data available. Plan on migrating all validated data to production tables along with the rest of the code. Implementing a dashboard with predefined history and trends will ensure a great first impression and enhance user confidence. In addition to these areas of focus, consider several design best practices for both database/data quality and dashboard interfaces. Database-Level Best Practices Ideally, your dashboard will be running in a stable database environment. This environment may be man- aged by your team or may be the responsibility of another area of your company. Either way, your dashboard is meant to provide data for quick and meaningful analysis, so treating the data and the tables in which it resides is critical. Some best practices include: ■■ Using ETL or other data acquisition methods to regularly write to a highly aggregated, denormalized table. This will ensure optimal performance, as dashboard click events need to be fast. A good goal is to ensure that no click event takes more than three seconds to return data to the dashboard. ■■ Use predefined and well-maintained dimensional tables wherever possible. This ensures consistency and eliminates redundant data structures. ■■ Store the data using IDs, and reference static code or dimensional tables wherever possible. This way, if a business rule changes, only one table must be modi- fied, and no data has been persisted to a table that is now outdated. ■■ Design and model the data so the front end can dynamically handle any business changes at the source level. This eliminates the need to update the code every time business users make a change, and maintenance costs will be much lower. The develop- ment team will then be able to work on exciting new projects rather than just keeping the lights on. ■■ Detailed data should be kept separate and not reloaded anywhere, if possible. However, it should be available in the same database so the aggregate and related detail can easily coexist. ■■ Unless absolutely necessary, do not store calculated values or any data that is prone to business rule changes. If persisted data becomes incorrect, it can be a huge effort to re-state it. Calculated fields can be done quickly using front-end queries or formulas (if designed properly). ■■ Create a data archival strategy based on business needs for data retention and how much history the dashboard needs to show. ■■ Ensure that any queries from the dashboard to the tables are well-tuned and that they will continue to run quickly over time. ■■ Likewise, ensure that any middle-tier environment used for running the dashboard queries is highly stable and can take advantage of any caching opportunities to enhance performance. Dashboard-Level Best Practices Spending a great deal of time on getting the dashboard data modeled, stored, automated, and correct will, of practical dashboard development
  32. 32. 30 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 course, all be for naught if the dashboard front end is not intuitive, does not perform, or otherwise does not have high availability. To address this, take these steps throughout the life cycle: ■■ Check the dashboard usability by bringing in end users who were not involved in the initial project. Observe how quickly and easily they can meet their objectives, and remove all bias as you watch. You will need to plan for their participation well in advance, and this work should be done early in your testing (make sure to have production data at this point) so there is time to react to their input. ■■ Within the dashboard code, implement dynamic server configuration so all dashboard components can automatically reference the proper environment for the database, middle tier, and front end itself. This reduces reliance on hard-coded server names and can prevent deployments from accidentally pointing to the wrong location. ■■ Users will want to use Excel regardless of how well-designed your dashboard is. Make sure an Excel export option is available for all the data shown on the dashboard and any included reports. ■■ For every dashboard component, include a label referencing the specific data source as well as the data refresh date. This simple step resolves confusion and will greatly reduce the number of support questions you receive post-rollout. ■■ Do everything possible to avoid hard-coding filters, axes, or any other front-end components that change based on predictably changing business. The data and the front end both need to be flexible and dynamic enough to display information based on a changing business. The dashboard should not have to display invalid or stale data for a time period while the devel- opment team scrambles to implement a production fix. That would inevitably lead to a drop in user adoption and reduced confidence in the dashboard’s validity. ■■ Test plans should include scripts for testing the overall dashboard load time as well as specific load times for all click events. This will afford the time needed to tweak code for optimal performance. ■■ Near the end of testing, simulate a performance load test whether you have automated tools to do this or you have to do it manually with multiple users. The purpose is to ensure no part of the underlying infrastructure has an issue with load. ■■ Test boundary conditions to avoid unforeseen defects later in the project’s life. For example, what happens when a multi-year trend goes into a new year? Will the x-axis continue to sort properly? Define all conditions like this and find a way to test each one. Running the Project (and Subsequent Projects) Considering the myriad of complexities involved in implementing a dashboard, from ensuring correct data is available when expected, to designing a usable and innovative front end, to working with the business through multiple and complex requirements, costs can be high and timelines can easily be missed if the project is not carefully managed. The following procedures will help ensure a successful dashboard release, all in the context of the best practices explained so far: Create and use an estimating model. The model should cover all aspects of a dashboard release (from data to user interface), all the technical roles and resources that will be involved, and be sufficiently detailed to break down time in hours by both phase and resource type. A model Do not store calculated values or any data that is prone to business rule changes. practical dashboard development
  33. 33. 31BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 that can be defined by selecting answers to requirements- based questions will be the easiest for your analysts to use, such as: How many metrics and components will be displayed? How many data sources will be used? Does data for validation exist? The model should be refined after each large project by entering the answers to these questions and determining how closely the model’s hours match those actually spent. Data validation is your top priority. Plan and allocate the time with your business partners and understand what data sources you will use for validation. If there is no indepen- dent source, you and your business users must reach an agreement about how validation will be performed. Share real data as soon as it becomes available and the team has reasonable confidence in its accuracy. There is no reason to wait to share data, regardless of how early in the process this occurs, because the earlier data defects are identified and resolved, the more smoothly the subsequent processes will go. As we’ve mentioned, we recommend you implement your project with historical data loaded. If this is planned, ensure that business users are aware and secure their pledge to spend adequate time comparing and validating the historical data. Define phases of work and identify key deliverables for each. Regardless of the development methodology your depart- ment uses, you must align milestones to specific dates to ensure the project does not get out of control and to keep business users confident in your progress. Depending on your business client and their expectations, you may need to blend agile and waterfall methods. Although this will not satisfy ardent practitioners of the methodologies, a blended approach can allow for the iterative testing and discovery that this type of work requires while ensuring adherence to a strict timeline, which a release of this complexity also requires. Implementations are complex, so make a detailed plan. The manager or lead of the project should define all the steps needed, assign dates and responsible parties, and build a T-minus document/project scorecard. These tasks should be completed during the initial stages of the work, soon after any intake approval and/or slotting, and the document should be reviewed with the entire team at least once a week to ensure the project is consistently on track. Escalate all identified issues and risks early and often. If your department already has a process for bringing issues and risks into the open and to the attention of those who can mitigate them, use it. Otherwise, create your own process for the project. Enlist all stakeholders and technology leaders for support, and do this proactively. Review, review, review. Plan multiple design and code reviews, and assume at least a draft and final review will be needed for each major piece of work. Devote ample time to design review, because waiting until the dashboard is built may make recovery impossible if a fundamental design flaw has gone unnoticed. Formalize a method for tracking and implementing all changes identified during reviews. Keep the development team engaged. For example, if the development team includes offshore resources, record key meetings using Webinar technology. This can serve as both knowledge transfer and training material later. Make sure everyone knows about the recording and ensure that no legal or compliance issues will arise. Even though your work may be completed in phases, dashboards can rarely be efficiently delivered if a “factory” Depending on your business client and their expectations, you may need to blend agile and waterfall methods. practical dashboard development
  34. 34. 32 BUSINESS INTELLIGENCE Journal • vol. 17, No. 4 approach is used (in which requirements are passed to designers, and designs passed to builders, without every- one being involved). When a developer is far removed from business users working on a dashboard project, this can lead to project failure. Implement a formal user-acceptance testing process. Once the development team has completed all internal testing of data and functionality, plan time (we recommend two to three weeks) to allow the business team to complete their tests. Testing should include as much data valida- tion as possible. We recommend you formally kick off the testing phase with business users and employ a docu- mented process for submitting defects and questions to the development team. Make this easy for your business partners. They should focus on testing, not on how to submit their test results. Require sponsor/stakeholder approval before rollout. This will give your dashboard legitimacy to the ultimate end users and is invaluable for those early weeks when adoption may hang in the balance. This approval should include a presentation during which the sponsor can view and provide feedback about the dashboard, with sufficient time allotted to make adjustments. As mentioned, we recommend you conduct sponsor reviews of the dashboard throughout the project, including during prototype design. Post-Implementation (You’re Never Really Done) After the dashboard is implemented, team members are often tempted to relax. There may also be new projects demanding focus. Do not become distracted or complacent, because there are certain post-implementation steps that will ensure both that the few critical months after rollout go smoothly and that the development team does not become bogged down by production support or answer- ing business questions. First, build post-implementation work into the initial plan. Sustainability and support should be factors in scope and technical design. For larger rollouts, consider best practices for the sponsor- ing business group and the technology team to handle presentations. This way, both business and technical questions can be answered accurately, all key partners are included, and accountability is shared. Post-rollout sponsorship and change navigation coordination are crucial. The business unit will likely be responsible for communications and training, but the technology team can and should influence this. If possible, ensure you have a method to collect usage metrics. If you can identify usage by user ID, that is even better because delineation between business and technol- ogy usage can be made and groups can be identified for training if usage is lower than expected. The development team can suggest and implement innovative ways to communicate with users: ■■ Add a scrolling marquee to the dashboard or use some other technique for instantly communicating impor- tant messages. This component should be database driven, and the technical support team should have write access into a table separate from the dashboard’s main tables. This way, announcements such as planned downtime or key data load dates can be easily delivered to all users. ■■ Add an e-mail button that goes directly to the dashboard support team. This may not be a popular choice for all technology teams, but dashboards are Create an internal process for ticket and defect handling, and implement bug fixes in small, bundled releases. practical dashboard development