MILITARY READINESS ENSURING READINESS WITH ANALYTIC INSIGHT Autumn Kosinski Principal Kosinkski_Autumn@bah.com Steven Mills Principal Mills_Steven@bah.com
ENSURING READINESS WITH ANALYTIC INSIGHT THE CHALLENGE: UNDERSTANDING THE TRUE STATE OF READINESS After more than a decade of operating on a wartime footing, the military services are struggling to meet readiness requirements in many critical areas. Increasingly tight budgets, combined with operating reserves that are stretched to the limit, leave little room for error in readiness planning. Numbers and capabilities must be precisely calculated and shortfalls anticipated, so that mission resources can be allocated in the most effective manner to meet mission objectives. Unfortunately, current decision-support systems do not provide the deep level of insight in enough time to address today s demanding readiness challenges. The factors used by military leaders to determine readiness include qualitative and quantitative assessments ranging from commanders reports to measures of personnel, training, equipment usage, maintenance cycles, supply chains, and other related factors. Readiness reporting systems across the Department of Defense (DoD) collect current readiness inputs and near-term projections from military units with the goal of generating objective, accurate, and timely qualitative and quantitative readiness measures. The systems contain a profusion of readiness data, but neither the systems nor the data provide decision makers responsible for assessing composite readiness with a sufficient understanding of the complex interrelationships among readiness variables. For example, how will an anticipated change in a readiness input really impact readiness at the unit level and, equally important, how will it impact readiness outside of the unit? Having a more sophisticated and accurate understanding of readiness impacts and trade-offs will enable leaders to make more realistic readiness decisions, and do it in timely and cost-effective manner. A NEW PERSPECTIVE: CAPITALIZING ON THE FULL VALUE OF READINESS DATA Military leaders recognize the limitations of readiness reporting systems in addressing today s readiness challenges. They also recognize that emerging analytics capabilities can provide sharper insight into the actual meaning and implications of readiness data, including how changes in micro-level readiness ripple through the entire readiness enterprise. Many even anticipate that advanced analytics will eventually enable predictive insight into future readiness. That is, there is a growing expectation that data science will help us build analytical tools that go beyond current readiness reporting to forecast how the decisions and choices made today will impact future readiness. The conventional approach to creating more objective, data-driven readiness systems has called for collecting more data and different types of data. Although it seems reasonable to believe that more data will generate more insight, this approach actually marches up the wrong path. The problem isn t lack of data or even lack of the right data. For the most part, the services are already collecting the data needed for more robust analytic assessments of readiness. The problem is a lack of understanding of how the data are interrelated. Data scientists would say the current data sets can give us a descriptive understanding of readiness but not a diagnostic understanding and certainly not a predictive view on the issue. A brief look at weather forecasting will help to illuminate these concepts. Meteorologists use a variety of instruments to observe and describe the weather. For example, their instruments may show that the temperature is 86 degrees, the humidity is 61 percent, and the wind is blowing east to west at 8 miles per hour. This is descriptive analytics, as shown in the chart below. 1
The problem isn t lack of data or even lack of the right data. For the most part, the services are already collecting the data needed for more robust analytic assessments of readiness. The problem is that we lack a complete understanding of how the data are interrelated. Descriptive analytics tells us what happened (or what s happening now). In addition, meteorologists know that there is a cause-and-effect relationship among the different weather elements. Air temperature, humidity, atmospheric pressure, and other elements interact to create thunderstorms, snowstorms, and other weather events. This is diagnostic analytics, which enables meteorologists to gain insight into why specific weather events occurred (or why they are occurring now). Diagnostic analytics is an important building block and necessary step on the path toward predictive analytics. To forecast weather for hours and even days in advance, meteorologists have amassed an enormous amount of diagnostic data that provides them with the foresight into how the weather elements are interrelated that is, how they interact with each other over time. Meteorologists can also bring into play other interrelated variables, such as elevation, ocean temperatures, and air streams to build even more reliable forecasts. Weather forecasts are not 100 percent accurate, and the reliability of forecasts declines as we look further in the future. Nevertheless, the diagnostic and predictive capabilities of weather forecasting offer valuable insight and improve decision making for everything from picnic planning to rocket launches. Military readiness systems currently contain the necessary data for descriptive analytics, such as equipment levels, maintenance schedules, spare parts inventories, supply chains, procurement pipelines, personnel levels, training, and other readiness-related factors. The goal now is to create diagnostic analytics that can model the extremely complex relationships among all of these variables across all the services worldwide. That is, how do the different data elements interact to cause or create actual readiness for the full range of potential missions? And not just at the unit level, but across the entire military enterprise worldwide. Establishing the measurable relationships among readiness variables presents a complicated but not impossible task. Military commanders already have an intuitive understanding of the cause-and-effect relationships among the variables for example, whether training in one field translates into competence in another field; whether helicopters operating in Afghanistan s rugged terrain will be worn down at faster than normal rates; or how a given readiness level of a Close Air Support squadron impacts the mission of the troops it supports. Commanders are often asked to make such readiness assessments based on their knowledge, experience, and judgment regarding the meaning of the descriptive data. However, no commander or group of commanders has the ability to consistently and rapidly make complex set of calculations and precise assessments of readiness quickly and accurately across the entire spectrum of readiness data now available to decision makers. In order to create diagnostic readiness models with the needed power and precision that can provide decision-quality information in a timely manner, we must incorporate the 2
ANALYTIC BUILDING BLOCKS commanders knowledge and expertise about causeand-effect into the readiness data and systems, thus enabling advanced analytics to make those calculations and assessments. OUR APPROACH: BUILDING CAUSE AND EFFECT INTO READINESS MODELS The starting point for building a diagnostic readiness model is not the data but the questions that need to be answered. For example, the data can tell us what experience and training a unit has, but military leaders need to know: Is it the right experience and right training for the mission at hand? The data can tell us that a unit is ready to deploy, but the question is: How will the unit s decreased readiness impact the readiness of other forces and, eventually, of the mission itself? It s not that the data are unimportant. This is, after all, data-driven decision making. However, the key point here is that as we move from descriptive to diagnostic analytics, the emphasis shifts from the readiness inputs to the readiness outputs: What are the data telling us about our true state of readiness? Productive diagnostic readiness models (and, eventually, predictive models) rely on a multi-disciplinary approach that includes both data scientists and domain experts. Data science itself consists of a variety of disciplines, including mathematics, probability and statistics, information science, data mining, data warehousing, advanced and predictive analytics, and machine learning. The domain experts are those with military mission and readiness experience, people who understand the factors that determine readiness and how those factors interact to create different levels of readiness. The domain experts also include the user communities that participate in the readiness process and use the readiness systems. They understand military s readiness reporting data, processes, and systems. 3
Toward this end, the team of data scientists and domain experts will focus on three interrelated activities as they build the diagnostic readiness model: Compiling the data and ensuring its reliability. Data scientists tackle and resolve such questions as: Where does the data originate? Is it standardized? Is it accurate and up to date? Can it be readily accessed, stored, and combined with other data? Domain experts are essential to helping data scientists address these questions, because they understand the data that is being collected, where it is located, and what it purports to measure. Understanding relationships (cause and effect) among data. Domain experts take the lead in this step, because they have the operational insight that enables them to understand the context and implications of the data, as well as the important cause-and-effect relationships among the data elements. Mapping the relationships among data. Data scientists with mathematical and statistical expertise will measure and map the data relationships and help create models that express those relationships. The algorithms and equations expressing these judgments and defining the relationships among variables can be programmed into the analytic models, as guided by the expertise and judgment of experienced commanders. Over time, the models themselves will also provide feedback to help refine the algorithms that is, refine the mathematical expression of the relationships to improve readiness forecasts. This approach combines art and science. Data, on its own, is relatively meaningless. Domain experts help the data scientist understand its meaning in relation to mission readiness. Working together, the data scientists and domain experts draw on both their imaginations and experience to see how different types of data might be combined to create meaningful insights. They may also posit hidden relationships that can be tested and refined. Intuition, as well as knowledge and experience, guide the development of analytic models. The multidisciplinary team builds on progressive tiers of data collection, refinement, and analytics, combining rigorous scientific methodologies with deep domain knowledge and less tangible but equally important skills, such as creativity, inquisitiveness, and collaboration. The immediate goal is to gain greater insight into readiness at the unit level, and then to see how the various measures of readiness interact to create overall service readiness to meet specific missions. In addition, the creation of a diagnostic readiness model puts in place the foundation for predictive readiness models. The creation of these models will have a number of positive effects, such as helping the services to: Address current issues related to data collection, siloed systems, etc., thus improving the access to and reliability of the data Improve the questions asked of data Identify and measure the right readiness activities and variables Incorporate lessons learned from commanders Increase understanding of the relationships among readiness variables Apply machine learning to improve the algorithms and sharpen the insights Articulate a defensible case for the resources necessary to maintain a given level of readiness or grow readiness All of these factors will work together to enhance the reliability of the diagnostic and predictive analytics models. With a diagnostic readiness model in place, military leaders can significantly improve insight into three areas of readiness: 4
1. Actual readiness at the unit level (based on current readiness data). 2. Implied readiness of the larger entity (resulting from the improved unit assessment). 3. Future readiness of the larger entity (based on the long-term impact of current readiness factors). The level of precision or confidence in the model s calculations will decline as we move down each of these three areas. That is, we will have more confidence in the model s assessment of current readiness at the unit level than in its predictions of future readiness. This is true of weather forecasts as well. Weather forecasts become less precise as they peer further into the future. Nevertheless, our diagnostic and predictive knowledge of weather is used to guide a wide variety of economic, military, transportation, and recreational activities, as well as our own personal activities. We infuse our readiness systems with equally powerful analytic capabilities and trenchant insight to guide decision makers and strengthen the mission readiness of U.S. forces. BOOZ ALLEN HAMILTON: YOUR ESSENTIAL PARTNER IN DATA-DRIVEN READINESS Booz Allen Hamilton is an industry leader in the diagnostic approach, including developing diagnostic and predictive assessments. Booz Allen is uniquely positioned to bring the skillsets together that will enable Data-Driven Readiness. Our single profit and loss (P&L) center allows for rapid and seamless integration of diverse skillsets, and as a result, we operate with an enterprise mindset bringing together technical experts in analytics and data science, with operational experts and decades of experience in readiness related issues and challenges. Readiness Innovation Leader. Booz Allen has a long history of working in the readiness arena our projects have focused on innovative improvements, technical systems and operational enhancements. The Booz Allen Readiness Center of Excellence Mission Readiness Capability focuses on formalizing best practices and applying innovation to deliver readiness solutions with increased speed and schedule efficiency. Analytics and Data Science Expertise. An integrated approach to readiness calls for a multi-dimensional team, with the highest level of technical expertise. We offer highly qualified personnel possessing deep experience with government & industry who are providing the thought leadership and innovation that drives the leading edge in Readiness. Booz Allen s Data Science and Readiness Analysis teams were established in 2010 as a natural extension of our business intelligence and cloud infrastructure development work. Seeing the need for a new approach to distill value from our clients data, we built a multidisciplinary team of computer scientists, mathematicians and domain experts to tackle the collection of problems and opportunities for leveraging data. Their collaborative effort immediately produced new insights and analysis paths, solidifying the validity of the approach. Since that time, our Advanced Analytics and Data Science team has grown to 500 staff supporting dozens of clients across a variety of domains. This breadth of experience provides a unique perspective on the conceptual models, tradecraft, processes, and culture of data science. 5
ABOUT OUR AUTHORS Autumn Kosinski, Principal Autumn Kosinski is a Booz Allen Hamilton Principal and leader of Booz Allen s Readiness Center of Excellence. A recognized expert in strategic, operational, and tactical readiness, she has more than 15 years of experience in military operations, and readiness systems development and life cycle process support across the DoD. Steven Mills, Principal Steven Mills is a Principal at Booz Allen Hamilton with expertise in data science, operations research, and modeling & simulation. His expertise in analytics and data science helps government and military organizations make more accurate and cost-eff ective readiness decisions through data-driven readiness. Mills team has created The Field Guide to Data Science, designed to help organizations of all types and missions understand how to make use of data as a resource, Explore Data Science, an interactive web-based data science training, and the Data Science Bowl, the premier data science for social good competition. He leads the Firm s Data Science for Social Good agenda and the Cloud Analytics & Data Science community of practice. He currently serves as Booz Allen s Director of Machine Intelligence. Mills holds a B.S. from Frostburg State University in Wildlife and Fisheries Management and an M.S. in Operations Research and Forest Resource Management from Penn State University. 7
About Booz Allen Booz Allen Hamilton has been at the forefront of strategy and technology for more than 100 years. Today, the fi rm provides management and technology consulting and engineering services to leading Fortune 500 corporations, governments, and not-for-profi ts across the globe. Booz Allen partners with public and private sector clients to solve their most diffi cult challenges through a combination of consulting, analytics, mission operations, technology, systems delivery, cybersecurity, engineering, and innovation expertise. With international headquarters in McLean, Virginia, the fi rm employs more than 22,600 people globally and had revenue of $5.41 billion for the 12 months ended March 31, 2016. To learn more, visit BoozAllen.com. (NYSE: BAH) 2017 Booz Allen Hamilton Inc. analytics viewpoint 03242017 BOOZALLEN.COM