External Reporting to Regulatory Bodies and Shareholders

External reporting doesn’t lead to competitive advantage under “business as usual” conditions. It’s not normally an advantage to report more quickly and accurately beyond a certain level. Cisco Systems, for example, has touted over the years its ability to close the books instantaneously and to report its results almost immediately at the end of a financial period. But we often wondered why the company bothered; the SEC doesn’t demand instantaneous closes, and it’s not clear what else a company could do with financial results intended for external consumption. But it is a different story with the information used by managers to make better strategic, investment, and operational decisions.

Reporting and scorecards, both financial and operational, are some of the most common applications of business intelligence and decision support. They’re obviously important to managing any business, and they’re increasingly important (with the advent of Sarbanes-Oxley legislation, for example) to keeping senior executives out of jail. While organizations do not compete on the basis of their reporting and scorecards, having systems in place to monitor progress against key operating metrics and monitor progress against plan is critical to strategy execution.

New regulatory data requirements can become opportunity to create a rich data source to improve performance. For example, pharmaceutical companies routinely interact with healthcare professionals both as customers and as consultants, researchers, and marketing speakers. Due to the potential for a conflict of interest, more than forty countries have enacted “anti-kickback” regulations that require transparency into these relationships. As a result, companies must aggregate all physician spend data into a single location so that it can be reported appropriately to the correct jurisdiction. According to Everest Group, “The industry is viewing the increased regulatory supervision as a burden. However, the regulatory requirements are putting in place a foundation for data requirements that can be used to drive pharma and life sciences analytics. Organizations can extract additional value in a challenging market, and create competitive differentiation by utilizing this opportunity.”¹ Medical device manufacturer Zimmer Biomet, for example, uses MediSpend’s data and analytic dashboards to monitor compliance with global compliance laws. But it also leverages the data to create new business models that drive device and drug development.

As we argued in chapter 1, reporting activities have typically not made extensive use of analytics, but there are exceptions. One is the prediction of future performance.

Public companies have to make regular predictions of future performance for investors and analysts. The consequences of poor predictions can be dramatic; investors may severely punish companies that fail to “meet their numbers.” Most companies make straightforward extrapolations of results from earlier periods to later ones. Yet in certain industries with a high degree of change and uncertainty, extrapolations can be problematic and yield poor predictions.

The information technology industry is one of those uncertain industries. Products and customer desires change rapidly, and disproportionate sales volumes take place at the ends of reporting periods. Hewlett-Packard found that it was very difficult to make accurate predictions of revenues in such an environment, and in one quarter of 2001 had a “nearly disastrous” revenue growth forecasting error of 12 percent.² Hewlett-Packard executives decided to put some of their quantitative researchers in HP Labs onto the problem of creating more accurate revenue forecasts.

The researchers used a Bayesian inference approach (see definition in the box earlier in this chapter) to predict monthly and quarterly revenues from data thus far in the period. After several adjustments, the algorithm yielded significantly better revenue predictions than the more straightforward approach used previously. Hewlett-Packard incorporated the new algorithm into its performance-reporting dashboard, and the company’s (then) CFO, Bob Wayman, noted, “It is reassuring to have a solid projection algorithm. It’s crisper and cleaner, it has rigor and methodology as opposed to my own algorithm.”³

Better prediction of future performance from an operational standpoint also allows companies to take action sooner. Using “near-real-time” operational data, managers can quickly identify emerging trends, make predictions, and take prompt action. For example, during the last recession, Dell executives used predictive models to recognize months before their competitors how thoroughly sales were going to soften. They took preemptive action on prices and products, which resulted in better (or at least, less awful) financial performance during the downturn. More importantly, as the recession ended, they were able to adjust again to boost their market share.

Enterprise Performance Management and Scorecards

Another exception to financial reporting applications involving analytics is when companies try to explain financial performance from nonfinancial factors. A successful enterprise performance management initiative requires companies not just to predict performance accurately but also to come to grips with some broader questions: Which activities have the greatest impact on business performance? How do we know whether we are executing against our strategy? An organization needs quantifiable insight into the operational factors that contribute to business results and a way of measuring progress against them.

Over the past decade or so, the biggest advance in management reporting has been the shift from IT delivering structured reports or fairly static scorecards to giving managers unprecedented access to enterprise data. Information workers and managers are able to access and explore financial and operational information from across the organization using business intelligence applications that incorporate alerts, management dashboards, dynamic filters, and data visualization tools. As analytical tools incorporate artificial intelligence technologies like natural language generation, they are able to help managers understand and interpret the data, too. For example, Credit Suisse integrated Quill (an advanced natural language generation platform from Narrative Science) into HOLT, its investment data analytics platform. Credit Suisse analyzes both proprietary and market data to produce investment research reports that assess company expectations, upside and risk to help analysts, bankers, and investors make long-term investment decisions. Credit Suisse has achieved full analyst coverage on all five thousand companies profiled within its platform, increasing the number of companies that Credit Suisse analyzes by more than 300 percent. Because investment analysts no longer need to write summary reports themselves, they are able to focus more on their clients and on conducting more sophisticated analyses.

These performance management systems report not only on financial performance but also on such nonfinancial domains as customer relationships, employee learning and innovation, and operations. These have been a great step forward in understanding performance. But too many adopters of balanced scorecards still aren’t very balanced, focusing primarily on financial reporting. It’s great to add nonfinancial metrics to a scorecard, but if investors and regulators don’t really care about them, it’s natural to emphasize the financial numbers.

Another problem with most dashboards is that even when companies do include both financial and nonfinancial measures, they don’t relate them to each other. Management professors David Larcker and Chris Ittner studied several companies that had adopted balanced scorecards.⁴ None of the companies they examined had causal models that related nonfinancial to financial performance.

Nonfinancial or intangible resources (such as human knowledge capital, brand, and R&D capability) are growing in importance to both company business performance and external perceptions of company value.⁵ Even the most favorable studies only show an explanatory power of approximately 50 percent for the effect of financial measures such as earnings per share, net income, economic profit, or return on invested capital on a company’s market value.⁶ In some industries, EPS accounts for less than 5 percent of value. We believe that a management team that manages all its sources of value—tangible and intangible, current and future—has a significant advantage over those who do not.

Some companies are working to develop a holistic understanding of both financial and nonfinancial value drivers. A few companies at the frontier are seeking to manage both their current and future shareholder value.⁷ These organizations are exploring how to infuse their scorecards with data from Wall Street analysts and future-value analytics to gain better insight into the implications of decisions on shareholder value. Companies that develop such a capability might be well on the way to competitive advantage.

Reporting and scorecards are most likely to lead to competitive advantage when the business environment is changing dramatically. In those circumstances, it’s particularly important to monitor new forms of performance. New measures need to be developed, new models of performance need to be created, and new management behaviors need to emerge. The speed and effectiveness of organizational transformation in such periods can certainly be or lead to a competitive advantage. For example, a property and casualty insurance company we worked with needed to transform itself. It was a poor financial performer, with losses over the last four years of well over a billion dollars. It paid out $1.40 in claims and expenses for every premium dollar it took in. The company had monitored its financial results, but it didn’t have a handle on what was creating the poor performance.

As part of a major corporate transformation, the company focused on three key processes: producer (agent) relationships, profitable underwriting, and claims execution. In addition to redesigning those processes, the company created new measures of them and collected them in a balanced scorecard. The scorecard served as a means to rapidly communicate the performance of and changes in the processes, and the success of the change initiative overall, to the management team. The scorecard assessed the company’s ability to deliver on such objectives as:

• Selecting profitable new markets to enter
• Attracting and selecting the right customers
• Driving pricing in accordance with risk
• Reducing the severity of claims

Reward systems for employees and managers at all levels of the company were changed to tie to performance on these objectives. Some automated analytics were embedded into underwriting systems to speed the process and improve the quality of pricing decisions.

The company used these process changes and reporting approaches to dramatically turn itself around. It began to make substantial amounts of profit and was eventually acquired by another insurance firm for $3.5 billion—up from perhaps zero value a few years earlier.

Cost Management

Although some might question whether cost management can lead to a competitive advantage, there are a still a few examples of organizations that have made effective analysis and management of costs a strategic capability.

In health care (at least in the United States), few hospitals even know their costs of treating patients, and are thus rarely able to control them. One exception is University of Utah Healthcare, which includes four academic medical centers and a variety of clinics and centers in the Salt Lake City area. UUHC embarked on a multiyear effort to accurately measure its costs. The project, called Value Driven Outcomes (VDO), used an activity-based costing approach. VDO fully integrates all available data regarding a patient encounter to provide a complete and comprehensive view of the care of patients. UUHC started by developing accurate cost accounting data using the actual acquisition costs of each item used in each patient encounter. Then comprehensive utilization methods were employed to assign all the costs to each patient encounter. Once cost information had been gathered, UUHC also developed key quality metrics and integrated the quality data with the cost and clinical data, so that a comprehensive analysis of cost-effectiveness could be used.

The hospital was then able to apply statistical methods to answer obvious—but previously impossible-to-answer—questions such as: What is the cost of poor quality? Where is there meaningful variation in cost or quality between physicians? Where is patient care over-utilizing labs or imaging where there is no additional quality or value as a result?

UUHC also engaged its physicians in the development and use of the VDO data and analytics tools. They developed self-service descriptive analytics tools with physician guidance, including continual feedback to ensure that what was being developed would meet their needs and answer their questions. The results are impressive: while costs at other academic medical centers in the area have risen by 2.9 on average annually over the past few years, UUHC’s have declined by 0.5 percent a year.

Energy costs are another expensive resource for many organizations. Microsoft’s five-hundred-acre corporate campus in Redmond, Washington, contains 14.9 million square feet of office space and labs. Seeking to save energy and reduce utility and maintenance costs, the company considered investing over $60 million to upgrade its incompatible networks of equipment and over thirty thousand sensors. Darrell Smith, director of Facilities and Energy, wanted to avoid a disruptive and expensive construction and replacement project. Additionally, the Microsoft team was concerned about displacing employees and the resulting loss in productivity. So instead, Smith and his team developed an “analytical blanket” to connect and weave together the diverse systems used to manage the campus buildings. The analytical layer enabled the team to string together data from thousands of sensors in the buildings, as well as in equipment such as HVAC, fans, and lighting systems. Soon they were accumulating billions of data points every week. Microsoft describes the benefits this way:

For example, in one garage, exhaust fans had been running continuously for a year (resulting in $66,000 of wasted energy). Within moments of coming online, the smart buildings solution detected the mistake and the problem was corrected. The software also informed engineers about a pressurization issue in a chilled water system in another building. It took less than five minutes to fix the problem, which saved about $12,000 each year.

The system also prioritizes problems to be fixed. Algorithms can balance out the cost of a fix in terms of the money and energy wasted with other factors such as the impact fixing the problem will have on employees who work in that building. So a lower-cost problem in a research lab with critical operations may be given higher priority than a higher-cost fix that directly affects few. According to Smith, almost half of the issues the system identifies can be corrected in under a minute.

Engineers appreciate being able to spend their time anticipating and preventing issues and solving problems instead of gathering data and reacting to emergencies. “I used to spend 70 percent of my time gathering and compiling data and only about 30 percent of my time doing engineering,” notes facilities engineer Jonathan Grove. “Our smart buildings work serves up data for me in easily consumable formats, so now I get to spend 95 percent of my time doing engineering, which is great.”

Microsoft is forecasting energy savings of 6–10 percent per year. And as each new algorithm comes online, the company finds new opportunities to save energy. For example, a “new algorithm for detecting when the air in a given building is being overcooled,” can save fifteen minutes of air conditioning for each instance detected. The team seems confident that even greater energy and cost savings are on the horizon.

Risk Management

Identifying and minimizing risk is a priority for every organization. Risk managers have been using data and analytical tools to do their jobs for many years. Recent technical advances and new types of data are reinvigorating the use of predictive analytics to mitigate risk. While the specific risks vary by organization, two of the most active areas have been in fraud detection and cybersecurity.

A growing challenge for financial services firms is to protect their customers and themselves against fraudulent claims or purchases. The Coalition Against Insurance Fraud conservatively estimates that fraud costs insurers $80 billion a year.⁹ Companies increasingly rely on sophisticated fraud detection algorithms to recognize fraudulent credit card purchases in real time. Credit card companies like Visa and American Express have been using advanced analytics methods for years to identify stolen credit cards and fraudulent purchases in real time. By combining point of sale, authorization, and geolocation information with past purchasing behavior, these companies are getting better at pinpointing suspicious behavior and alerting cardholders of potential problems.

Insurers are fighting back too, using predictive analytics to identify where they need to focus their attention. According to Bill Dibble, senior vice president of claims operations, Infinity Property & Casualty:

In another example, Infinity analyzed its old adjusters’ reports to develop a new algorithm that resulted in $12 million in subrogation recovery.¹¹

IT groups historically have been better at creating predictive models for other parts of the business, but they have been slower to embrace analytics to manage their own function. That is changing as cybersecurity challenges proliferate and become more daunting. The number of threats to large organizations is growing rapidly, as is the number of hackers who create them and the number of systems at risk from cyberattacks. Data breaches are increasing, according to one report, by 85 percent a year, and in 2016, half a billion personal records were stolen or lost.¹²

And with the proliferation of connected devices, there is no doubt that the challenge of protecting an organization’s data can only grow. Cybersecurity processes within companies are too often reactive to hacks and breaches; investigation and actions are taken only after (sometimes long after) a problem has occurred. The technology most commonly used to address cyberattacks employs “threat signatures” based on patterns of previous attacks. Of course, these approaches are of little value in preventing new types of attacks. Predictive analytical methods originally developed to detect and prevent credit card fraud—a form of anomaly detection—are now being applied to behaviors in cybersecurity attacks.¹³ Some cognitive technologies, including deep learning, can also identify anomalies in transaction patterns. These approaches can identify emerging anomalies much faster than using threat signatures, and may be able to identify threats much earlier and prevent substantial hacks and data losses before they occur. Given the sensitivity of cybersecurity issues, humans will still be necessary to confirm and investigate threats, particularly when they are internal. But the amount of investigative labor can be substantially reduced through analytics. Organizations in both public and private sectors today are using analytics and—to a lesser degree—cognitive technologies and automation to improve their cybersecurity programs. It’s unclear when these technical capabilities will be fully mature, but there should be no doubts about their necessity and the likelihood of their ultimate adoption.¹⁴

Manufacturing

Real analytical competitors in manufacturing, then, are those that go beyond just manufacturing. There are a few great examples of this approach. One is at a small steel manufacturer in the United States, and it illustrates that analytical competition applies both to smaller firms and to the manufacture of commodity goods. Rocky Mountain Steel Mills, a steel rail-, rod-, and bar-making division of Oregon Steel, faced a critical decision about manufacturing capacity in early 2005. It had shut down its seamless pipe mill in 2003 because of price pressures, but the primary customers for seamless pipe were oil drillers, and by 2005 oil prices had risen enough to make Rob Simon, the company’s vice president and general manager, consider reopening the pipe mill. However, he found the rules of thumb and standard costing/volume analyses previously used for such decisions to be too simplistic for a fast-changing mix of demand, pricing, production constraints, and industry capacity.

Simon decided to turn to a more analytical approach, and Rocky Mountain installed an analytical software tool called Profit InSight. He began to work with monthly analyses to determine whether the plant should be reopened. Potential customers and other managers thought that the high prices for pipe clearly justified the capital outlay that would be needed to reopen, but Simon’s analyses suggested that increased revenues for pipe would be offset by lower production of rod and bar, and would not be profitable. Only when pipe prices continued to rise throughout 2005 did Simon decide to reopen the plant in December of that year. Even when production started, his models suggested holding off on taking orders because prices were predicted to rise. Indeed, by January 2006 they had risen by 20 percent over the previous quarter.

Rocky Mountain estimates that in addition to the higher prices it received, it averted a $34 million loss it would have faced from production constraints had it reopened the plant earlier in 2005. The success with the new pipe mill was also a key factor in a substantial rise in Oregon Steel’s stock price. Profit InSight is now used as a weekly strategic planning tool, and Rocky Mountain Steel Mills has completely abandoned its previous “rhetorical” sales planning and forecasting approach for the new analytical methods. They are measurably superior, but Simon still had to demand that everyone listen to what the analyses show and quit second-guessing using the old approaches.

Engineers involved in product design have been using computer-aided design (CAD) for decades. But recent advances using parametric modeling permit more flexible, individually customized designs that also cut manufacturing time. Firewire Surfboards, a manufacturer of high-performance, premium surfboards wanted to give customers more freedom to customize their board without sacrificing board performance or overwhelming their design and manufacturing processes. Firewire boards are constructed through proprietary methods and a combination of high-tech materials previously not offered by other commercial surfboard manufacturers. But Firewire’s CEO Mark Price knew that this wasn’t enough to be a market leader. Hardcore surfers want customized boards—made for them specifically to suit their personal style and local wave conditions. But the complexity of Firewire’s production process made customization through its computer-aided design system extremely labor-intensive. Firewire worked with ShapeLogic and Siemens’ NX 3D CAD software to develop advanced parametric models of its stock and added specific rules that control how the customer’s changes affect the board’s curves to ensure peak performance.

Online customers begin by selecting one of Firewire Surfboards’ standard models and then altering the design to fit their needs. Once the customer orders a custom board, a precise solid model is sent directly to Firewire’s factory, where it is used to run the machines that manufacture the board. The model makes it possible to quickly machine custom boards to about 97 percent of their net shape, which minimizes the required finishing processes, manufacturing time, and costs.¹⁷ As more powerful and sophisticated (yet consumer-friendly) CAD solutions become more accessible, we expect customer customization of bicycles, sports equipment, automobiles and even medical devices will become commonplace.

For internet-based businesses, operations means churning out the basic service for which customers visit a website. Successful online businesses use analytics to test virtually all aspects of their sites before implementing them broadly. For example, because the primary reason customers visit Google is to use its search capabilities, the company has a very extensive program of testing and analytics with regard to its search engine. Google employs a wide range of operational and customer data and analytics to improve search attributes, including relevance, timeliness, and the user experience. The company developed many of its own proprietary measures of search relevance. Most of the metrics are gathered in an automated fashion, such as the percentage of foreign results, how deeply users go in the retrieved items list, the percentage of users who go to each page of the search result, and the measures of search latency or timeliness. But Google also collects human judgments on the search process, and even observes individual users (at Google headquarters and in users’ homes) as they use the site for individual queries and entire sessions. One technique employed is eye tracking, from which “heat maps” are created showing which areas of a page receive the most attention.

Google is heavily committed to experimentation before making any change in its search site. As Google’s Bill Brougher put it:

Google’s methods for analytical operations are as rigorous as any firm’s, and the nature of the business makes a large amount of data available for analysis.

Another key aspect of manufacturing analytics is to ensure that the right products are being manufactured. We’ll refer to it as the configuration problem—making sure that the products offered to the market are those that the market wants. The configuration problem, like the ones described earlier at Rocky Mountain Steel and Firewire Surfboards, is cross-functional; it takes place at the intersection of sales and manufacturing, and usually also involves the company’s supply chain, financial, and even human resource processes. To compete on the basis of configuration is thus by definition an enterprise-wide activity. Configuration is highly analytical. It involves predictive modeling of what customers want to buy, as well as complex (usually rule-based) analysis of what components go with what other components into what finished products.

What companies compete on the basis of configuration? Some high-technology companies, such as Dell, are known for their configurable products. Wireless telecommunications companies may have many different service plans; some have developed automated analytical applications to find the best one for each customer. They also tailor their services to each corporate account. Automobile companies need to compete on configuration, though US and European manufacturers have traditionally done it poorly. Since manufacturing a car from scratch to a customer’s specification is viewed as taking too long (at least outside of Japan, where it is commonly done), car companies have to forecast the types of vehicles and options customers will want, manufacture them, and send them to dealers. Far too often, the mixes of models and option packages have not been what customers want, so cars have to be substantially discounted to be sold during promotions or at the end of a model year. The mismatch between consumer desires and available product has been one of the biggest problems facing Ford and General Motors.

Both companies are trying to do something about configuration, but Ford is probably the more aggressive of the two. The company has shifted its focus from producing whatever the factory could produce and worrying later about selling it, to trying to closely match supply and demand. As part of this initiative, Ford is using configuration software to maintain rules about options and components, which will reduce the number of production mistakes and more closely match dealer orders to production schedules. Ford’s Smart Inventory Management System (SIMS) optimizes inventory for nearly three thousand Ford and Lincoln dealerships in North America. SIMS uses advanced analytics to produce dealer-specific vehicle order recommendations to ensure that dealers have the right number and mix of inventory to accommodate customer preferences and demand. As a result, annual revenue has increased and Ford dealerships are confidently making smart and cost-effective inventory ordering decisions.¹⁹ Ford has not yet entirely mastered the art of competing on configuration, but it is clearly making strides in that direction.

Quality

Analytics can also be applied to assess manufactured quality. Honda, for example, has long been known for the quality of its automobiles and other products. The company certainly has analytical individuals in its manufacturing quality department. However, it goes well beyond that function in identifying potential quality problems. Honda instituted an analytical “early warning” program to identify major potential quality issues from warranty service records. These records are sent to Honda by dealers, and they include both categorized quality problems and free text. Other text comes from transcripts of calls by mechanics to experts in various domains at headquarters and from customer calls to call centers. Honda’s primary concern was that any serious problems identified by dealers or customers would be noticed at headquarters and addressed quickly. So Honda analysts set up a system to mine the textual data coming from these different sources. Words appearing for the first time (particularly those suggesting major problems, such as fire) and words appearing more than predicted were flagged for human analysts to look at. Honda won’t go into details about any specific problems it’s nipped in the bud, but says that the program has been very successful.

Toshiba Semiconductor Company is another business that has made extensive use of analytics—in particular, visual representation of statistical analysis—in manufacturing quality. The initial applications focused on advanced analysis for new product and technology development, but they expanded quickly into other areas such as sales, marketing, development, production, and quality assurance. The semiconductor business unit’s executives are strong advocates of analytics, and have led the company by the concept for over fifteen years. Toshiba’s overall approach is encompassed by a broader initiative entitled Management Innovation Policy and Activity.

The visual analytics approach was first used by engineers in several semiconductor fabrication plants (fabs) for yield analysis—a key problem in the industry. According to Shigeru Komatsu, the company’s chief knowledge officer (it is rare for analytics to be addressed in such a role, but they are at Toshiba Semiconductor), “We have worked on standardizing performance indicators, we have built shared databases, we have made efforts to share analysis cases and results, and we have implemented analytic software such as Minitab and [TIBCO] Spotfire DecisionSite in order to increase our efficiency in analytics.”²⁰ Toshiba Semiconductor continues to invest in yield analytics, most recently by incorporating artificial intelligence into analytical models used to determine the root cause of defects and further boost production quality.²¹

Safety

Safety was not the earliest area to apply data and analytics, but it’s growing rapidly today. It turns out that certain types of accidents are—at least to some degree—predictable. They are a function of the people, equipment, and company settings involved. If you have data on past safety incidents and the attributes associated with them, it’s not a huge stretch to predict when they will happen in the future, and intervene to try to prevent them.

Safety analytics is one of the specialties of the boutique analytics consulting firm First Analytics, which Tom helped to found in 2009. A manager at a large railroad read the original “Competing on Analytics” article, and contacted First Analytics CEO Mike Thompson in 2010. They began discussing how the company could improve its safety using analytics.

The railroad manager explained that safety was a top priority for the company and that it had improved considerably on this front, but it got harder to keep improving. He said the company had already used some data to identify likely risks, but there was a lot more that could be explored.

The railroad and First Analytics began with a proof of concept project to take existing data on the company’s train operators and score how likely they were to be at risk. The available data included location, job position, weather conditions, work schedule, absenteeism, scores on rules tests and training exercises, previous rules violations, and more. The data eventually came from about twenty different databases across the company. The railroad had previously used a risk scoring system based on reasonable logic, but the new one based on analytics improved upon it dramatically.

Since the proof of concept had worked well, the railroad worked with First Analytics to expand the analysis to other safety areas; for example, an analysis was created to identify and prioritize the most at-risk railroad grade crossings. The personnel safety analytics were extended beyond those operating the trains to the personnel maintaining the tens of thousands of miles of track. The overall result of these efforts was a dramatic improvement in safety. The company’s vice president of safety recently discussed the improvements with customers:

Some companies in other transportation industries have adopted similar approaches to safety analytics. This is facilitated by the increasing ease of capturing driving data. Schneider National, Inc., for example, a large trucking firm, captures driver behaviors such as speed, acceleration and deceleration, and driving times. A predictive safety algorithm alerts supervisors that drivers are at risk for accidents even before they have had one.

Other industries that have taken aggressive approaches to safety analytics include mining, energy, and manufacturing. As sensors become more pervasive, we’re likely to see many more companies and industries adopt these approaches.

Logistics

When you invest in operational analytics at the level of hundreds of millions of dollars—and more importantly, deliver value at multiples of that sum—it’s safe to assume that you are competing on analytics. UPS has a long history of measurement and improvement through industrial engineering. Today, UPS captures information on the 18.3 million packages and documents, on average, that it delivers daily, and it receives 69.4 million tracking requests a day. The company has a deep, long-standing commitment to using analytics for decision making.

Jack Levis, the company’s senior director of process management, who also leads the Operations Research and Advanced Analytics groups, leads the ORION project. ORION, an acronym that stands for On-Road Integrated Optimization and Navigation, may be the largest commercial analytics project ever undertaken.²³ ORION is a prescriptive analytics logistical model for UPS’s fifty-five thousand drivers in the United States (the international rollouts will come soon). Before drivers start their routes, ORION analyzes the packages to be delivered that day and determines an optimal routing for the “package cars.” Handheld computers tell the drivers where to go next.

UPS is a big company, and the benefits of ORION are commensurately large. Jack Levis likes to say “Big savings come from attention to detail.” Shortening each driver’s route by one mile daily translates into $50 million to the UPS bottom line annually. Every minute saved per driver daily is worth $14.6 million, and preventing one minute of idle time daily saves $515,000. As a result, what seem like small incremental improvements can yield big savings. For example, savings in driver productivity and fuel economy by driving more efficient routes add up to about $400 million a year. By cutting routes just a fraction of a mile here and there, UPS is driving 100 million fewer miles, with a resulting reduction in carbon emissions of 100,000 metric tons a year. You don’t see those levels of benefit from an analytics project very often, and these have been confirmed through intensive measurement and reported to Wall Street analysts.

ORION required a sustained commitment over more than a decade on the part of UPS management. The system took more than a decade from inception to fully roll out, and more than $250 million of investment. So the company clearly went all in on this project. What took so long? First, the sheer volume of data and the complexity of the math needed to optimize routes are truly staggering. Consider that for a single driver making 120 deliveries, the number of potential routes is 120 factorial, which is a sum greater than the age of the earth—in seconds.²⁴ Now imagine computing the optimal route for every driver, every day.

The optimization algorithm itself was difficult to develop, but that aspect of the work paled in comparison to the other challenges. UPS had to develop its own maps to ensure that the drivers would be directed to the right place every time, and to the correct location for package drop-off or pickup. No commercially available maps could do that for the 250 million different locations to which UPS delivers. Next, telematics sensors were installed in more than forty-six thousand company trucks, which track metrics including speed, direction, braking, and drivetrain performance.

But it was the change management challenges that were the most complex to address. Imagine communicating and inculcating a new way of performing a core daily task to fifty-five thousand skilled union workers. UPS devoted six days of training and support to each driver. Most of the drivers wanted to know how the system worked before they would give up their time-honored routes, and so considerable effort was devoted to turning the “black box” algorithm into a “glass box.” To their credit, most of the drivers were enthusiastic about the new approach once they had experienced it.

ORION’s benefits to date are only the beginning for UPS. There is, of course, the global rollout of these tools. And to maintain a level of simplicity for drivers, ORION only optimizes routes at the beginning of the workday. More sophisticated programs down the road will reoptimize during the day, taking factors such as traffic patterns and weather into account. UPS plans to continue to extend ORION’s capabilities. Levis explains that future enhancements will also make other decisions to improve customer service without sacrificing efficiency. “If a customer calls with a request while drivers are on road, ORION will look at all drivers and determine the best choice. ORION will then put it in their handheld computer and adjust their route accordingly.”

There aren’t many companies that have made bets on analytics to this degree. But with the ORION initiative, UPS has truly transformed itself from a trucking company that uses technology to a “technology company with trucks.”

Managing the logistics of a global supply chain is another fertile area for investing in analytics. Caterpillar’s Global Supply Network Division (GSND) is responsible for overseeing an interdependent supply network with over eleven thousand suppliers and 150 facilities. But for thirty years, Caterpillar’s supplier performance appeared stagnant, with no sustainable improvements. GSND personnel were hampered by outdated manual processes, data unavailability, and network complexity. Users became accustomed to making decisions on assumptions and incomplete data. “We were trying to manage a supply network in the dark on spreadsheets.”²⁵

To solve this problem, in 2012 Caterpillar began to develop the Assurance of Supply Center (ASC). The ASC is a data analytics and visualization platform that captures and transforms supply network data from dozens of systems into a suite of powerful business tools that is used to drive both every day decisions and strategic network design decisions. Three years later, the transformation has been remarkable. CEO Doug Oberhelman told shareholders, “[ASC] simplifies the supply network—a network that involves thousands of suppliers shipping more than a million parts and components every year. Now Caterpillar can see orders from production to delivery—by facility, business unit and cost.”²⁶

Using mobile devices, users are able to see data on inventory, transportation, suppliers, performance, network footprint, defects, and tariffs. A shipment occurs every ninety seconds, producing millions of new data points every day that are analyzed, visualized to support decisions to over 10,000 users. The system incorporates a library of over 100 million data points feeding 45,000 predictive and prescriptive models, including data on 640,000 part numbers from over 7,000 suppliers, shipping to 127 facilities across the globe. All this data enables them to answer important questions such as:

• Where is my inventory?
• Are we ready for an upturn?
• Why is supplier X performing poorly?
• What is the network impact of a major natural disaster?
• Factoring in projected demand, consumption, and performance, what should our parts inventory be for all part numbers next year?
• How could I change my network to make it more cost-effective and profitable?

Caterpillar shares its data and insights with suppliers and works with them to improve their performance. In just three years, the results have been dramatic: in 2012, 67 percent of schedules were shipped on time; now it is 93 percent. Quality defects have been reduced by 50 percent. The organization now has the ability to keep the network moving even when disaster strikes. Users are able to resolve network performance issues in minutes rather than months. And ASC has become a continuing source of competitive advantage for Caterpillar as it continues to introduce more innovative analytic solutions to its business.