When big data goes lean: Article|McKinsey Quarterly

25 Feb

The combination of advanced analytics and lean management could be worth tens of billions of dollars in higher earnings for large manufacturers. A few leading companies are showing the way.

February 2014 | byRajat Dhawan, Kunwar Singh, and Ashish Tuteja

The application of larger data sets, faster computational power, and more advanced analytic techniques is spurring progress on a range of lean-management priorities. Sophisticated modeling can help to identify waste, for example, thus empowering workers and opening up new frontiers where lean problem solving can support continuous improvement. Powerful data-driven analytics also can help to solve previously unsolvable (and even unknown) problems that undermine efficiency in complex manufacturing environments: hidden bottlenecks, operational rigidities, and areas of excessive variability. Similarly, the power of data to support improvement efforts in related areas, such as quality and production planning, is growing as companies get better at storing, sharing, integrating, and understanding their data more quickly and easily.

Pioneers in the application of advanced-analytics approaches, some borrowed from risk management and finance, are emerging in industries such as chemicals, electronics, mining and metals, and pharmaceuticals. Many are lean veterans: these companies cut their teeth during the 1990s (when sagging prices hit a range of basic-materials companies hard) and more recently doubled down in response to rising raw-materials prices. The benefits they’re enjoying—an extra two to three percentage points of margin, on top of earlier productivity gains (from conventional lean methods) that often reached 10 to 15 percent—suggest that more big data applications will be finding their way into the lean tool kits of large manufacturers. Indeed, our work suggests that, taken together, the new uses of proven analytical tools could be worth tens of billions of dollars in EBITDA (earnings before interest, taxes, depreciation, and amortization) for manufacturers in the automobile, chemical, consumer-product, and pharmaceutical industries, among others (exhibit).


New uses of proven analytical tools will serve manufacturers across a range of industries.

Nonetheless, to get the most from data-fueled lean production, companies have to adjust their traditional approach to kaizen (the philosophy of continuous improvement). In our experience, many find it useful to set up special data-optimization labs or cells within their existing operations units. This approach typically requires forming a small team of econometrics specialists, operations-research experts, and statisticians familiar with the appropriate tools. By connecting these analytics experts with their frontline colleagues, companies can begin to identify opportunities for improvement projects that will both increase performance and help operators learn to apply their lean problem-solving skills in new ways.

For example, a pharmaceutical company wanted to get to the root causes of variability in an important production process. Operators suspected that some 50 variables were involved but couldn’t determine the relationships among them to improve overall efficiency. Working closely with data specialists, the operators used neural networks (a machine-learning technique) to model the potential combinations and effects of the variables. Ultimately, it determined that five of them mattered most. Once the primary drivers were clear, the operators focused their efforts on optimizing the relevant parameters and then managing them as part of routine plant operations. This helped the company to improve yields by 30 percent.

Similarly, a leading steel producer used advanced analytics to identify and capture margin-improvement opportunities worth more than $200 million a year across its production value chain. This result is noteworthy because the company already had a 15-year history of deploying lean approaches and had recently won an award for quality and process excellence. The steelmaker began with a Monte Carlo simulation, widely used in biology, computational physics, engineering, finance, and insurance to model ranges of possible outcomes and their probabilities. Manufacturing companies can adapt these methods to model their own uncertainties by running thousands of simulations using historical plant data to identify the probabilities of breakdowns, as well as variations in cycle times and in the availability of multiple pieces of equipment across parts of a production process.

The steelmaker focused on what it thought was the principal bottleneck in an important process, where previous continuous-improvement efforts had already helped raise output by 10 percent. When statisticians analyzed the historical data, however, they recognized that the process suffered from multiple bottlenecks, which shifted under different conditions. The part of the process that the operators traditionally focused on had a 60 percent probability of causing problems, but two other parts could also cripple output, though they were somewhat less likely to do so.

With this new understanding, the company conducted structured problem-solving exercises to find newer, more economical ways of making improvements. Given the statistical distribution of the bottlenecks, it proved more efficient to start with a few low-cost maintenance and reliability measures. This approach helped improve the availability of three key pieces of equipment, resulting in a 20 percent throughput increase that translated into more than $50 million in EBITDA improvements.

(Monte Carlo simulation holds promise in other areas, too. A mining company, for instance, used it to challenge a project’s capital assumptions, in part by deploying historical data on various disruptions—for example, rainfall patterns—to model the effect of floods and other natural events on the company’s mines. This effort helped it to optimize handling and storage capacity across its whole network of facilities, thus lowering the related capital expenditures by 20 percent.)

A second analytical tool the steelmaker employed was value-in-use modeling, long a fixture in procurement applications, where it helps to optimize the purchasing of raw materials. The steelmaker used these techniques to see how different blends of metallurgical coal1 might affect the economics of its production activities. The team investigating the problem started with about 40 variables describing the specifications (such as ash content and impurities affecting production) of different types of coal. Later it added fuel consumption, productivity, and transport costs. This approach helped operators to identify and prioritize a series of plantwide kaizen activities that lowered the company’s raw-materials costs by 4 to 6 percent. Moreover, procurement managers integrated the model’s findings into their routines—for example, by monitoring and adjusting coal blends on a quarterly basis; previously, they might have done so only once or twice a year, because of the complexity involved.

As the steelmaker’s example suggests, the key to applying advanced analytics in lean-production environments is to view data through the lens of continuous improvement and not as an isolated series of one-offs. The ability to solve previously unsolvable problems and make better operational decisions in real time is a powerful combination. More powerful still is using these advantages to encourage and empower frontline decision making. By pushing data-related issues lower in the organization, the steelmaker is encouraging a strong culture of continuous improvement. It is also identifying new areas to apply its growing proficiency in advanced analytics. One area is production planning, where the operations group is working with internal marketing and sales, as well as external suppliers, to improve the accuracy of sales forecasts and make production more efficient.

The steelmaker’s story shows that senior executives must take an active role. In our experience, the information and data required for many big data initiatives already exist in silos around companies—in shop-floor production logs, maintenance registers, real-time equipment-performance data, and even vendor performance-guarantee sheets. In some cases, data may come from outside partners or databases. Determining what to look for, where to get it, and how to use it across a dispersed manufacturing network requires executive know-how and support.

Original article source: click here

About the authors

Rajat Dhawan is a director in McKinsey’s Delhi office, where Kunwar Singh is an associate principal;Ashish Tuteja is an associate principal in the Mumbai office.

The authors wish to thank Abhishek Anand, Rajat Gupta, Snehanshu Mahto, Dev Ramchandani, Aman Sethi, Saurabh Srivastava, and Abhishek Tikmani for their contributions to the analysis underpinning this article.

Leave a comment

Posted by on February 25, 2014 in Investments


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: