Affordable Analytics for Small and Medium Business Cost Modeling with Predictive Analytics improves forecast capability of a midsized manufacturing company SimaFore 2011 White Paper Series Table of Contents Affordable Analytics for Small and Medium Business ........................................... 1 Table of Contents.................................................................................................. 2 Executive Summary .............................................................................................. 3 1. Business Analytics for SMBs: Background ..................................................... 4 2. Objectives of Cost Modeling ........................................................................... 5 3. Using Predictive Analytics for Cost Modeling ................................................. 6 3.1 Business Objective: Enable rapid data-driven cost forecasting................... 6 3.2 Data Sources: Mostly public domain ........................................................... 7 3.3 Analytics Process: A 4-step approach......................................................... 7 3.4 Model Deployment: An App-based approach.............................................. 9 3.5 About Cloud computing and deployment .................................................. 10 4. Conclusions: ROI from using predictive analytics......................................... 11 References ......................................................................................................... 13 About SimaFore .................................................................................................. 13 Contact SimaFore ............................................................................................... 13 About the Authors ............................................................................................... 13 Business Analytics for SMM © 2011 SimaFore LLC Page 2 www.simafore.com SimaFore 2011 White Paper Series Executive Summary In recent years, a trend has emerged – predicated on the availability of large amounts of data - wherein data analysis and statistical methods are being used in nearly every business function. It is being applied from customer acquisition and retention, to risk management, to demand forecasting. There is a new allencompassing name for this discipline – business analytics. As before the pioneers have been large companies. Thus, business analytics is an oft-used and well-worn term in companies like Bank of America, Progressive Insurance, Amazon and Google, for example. Small and mid-sized businesses (SMBs) seem to be largely under-represented in this new wave. However, this trend is likely to change in the near future. Two major drivers for this are the availability of affordable technology via Open Source and affordable computing via the Cloud. In this report, we will demonstrate how an SMB can implement and deploy predictive analytics for a common business application - Cost Modeling and Forecasting. We will discuss how L&L Products, a mid-sized custom chemical compounding company specializing in structural composites is leveraging analytics to improve their cost forecasting. The primary questions addressed include: 1. How can an SMB affordably develop objective and reliable models that can be used for cost forecasting? • This report describes a systematic process for doing so 2. How can a SMB leverage open source technology and cloud computing to address similar business problems? • We will demonstrate that through our process, this task is easy enough to be applicable to any business problem that an SMB may have to deal with. 3. What is the return on investment from using predictive analytics for cost modeling? • Predictive analytics can reduce the time to develop accurate cost estimates from several days to a few hours, as L&L is aiming to do Using this report SMBs can use this report to understand how cost modeling may be implemented in their own business. They can get informed on of how to leverage open source technology and cloud computing to help deploy such solutions. Additionally, they can extrapolate and identify other functional areas that can benefit immediately from the type of affordable analytics described here. Business Analytics for SMM © 2011 SimaFore LLC Page 3 www.simafore.com SimaFore 2011 White Paper Series 1. Business Analytics for SMBs: Background Business Analytics is “the extensive use of data, statistical and quantitative analysis, explanatory and predictive models, and fact-based management to drive decisions and actions.” [1] While this is a comprehensive definition of business analytics, a much better appreciation can be had by looking at some actual cases where analytics has made a difference. Some relevant examples in the context of this report pertain to small and medium businesses, specifically: • A family owned Midwestern dairy farm [2] uses analytics to Root out used milk-bottle return fraud at dairy stores Quickly identifies and corrects manufacturing glitches by mining customer complaint data Gain valuable insights on what factors would improve customer satisfaction and retention • A family owned vacation home rental business in the South [3] uses analytics to Track more than 15 competitors including large hotel chains to make the most efficient resources allocation Manage more than 1100 vendors for a variety of service needs ranging from lawn maintenance to pool and spa service. • A specialty clothing retailer [4] uses analytics to answer key marketing questions such as When has a prospective buyer looked over the items in the product catalog? How many times has each item in the catalog been viewed? What is the ratio of product views to sales? High capacity and process integrated data collection, sophisticated analysis tools and expertise were three key advantages that large businesses traditionally have over their smaller competitors. So how can a SMB adopt some of these game changing technologies? Today, SMBs have easy access to one of these three key ingredients – data. Sophisticated analytics tools are also becoming very affordable, thanks to open source [5]. With the maturing of cloud computing, deploying these solutions is also becoming less of a challenge. A central missing element for SMBs today is the expertise to build and manage these models. In the next several sections we address how to fill in this piece of the puzzle and answer the question raised above using one example application: cost modeling and forecasting with predictive analytics. Business Analytics for SMM © 2011 SimaFore LLC Page 4 www.simafore.com SimaFore 2011 White Paper Series 2. Objectives of Cost Modeling Cost Modeling is a common activity for any business. Cost modeling is a prerequisite for cost forecasting. However it can be a very challenging activity. What is Cost Modeling? Why should an SMB worry about cost modeling? Product development and manufacturing requires transforming raw materials into finished goods. This requires obtaining the raw materials, processing them to produce the desired goods, storing and distributing the finished goods. Obviously all of these activities involve cost from the business' point of view. But these costs are not constant and fluctuate on a daily, monthly or quarterly basis. Let us use the example of a small and medium manufacturing (SMM) business to demonstrate this. The typical customer for such a company is a large original equipment manufacturer (OEM) such as an automotive or aerospace company. OEMs however do not allow an SMM to vary the price of the finished good in tune with the market prices of raw materials or processing costs. In order to produce a part profitably, and yet compete successfully in the market, an SMM must have solid estimates of input factors' (material, labor and transportation) cost variability and how these will impact the ultimate cost of the finished part. Cost Modeling allows one to capture all these costs, relate them to the final cost of the product and helps develop accurate forecast ranges for the input factors' costs. Two major challenges in cost modeling The first challenge is the information required for building an effective cost model: it typically resides with multiple stake holders in the business as well as outside the business. The challenge is to successfully integrate these disparate sources of information, capture them in a "living" database to build effective models. The second challenge is the presence of uncertainty in the information. This uncertainty could come from both the supply side and demand side. When production volumes must swing between 100,000 to 200,000 parts, it adds additional life-cycle costs. How do you account for these variabilities and still come up with a cost estimate that allows managers to make a decision about production? Under such situations, one needs to focus more on cost distributions than single point cost estimates. But when optimal purchasing prices and volumes are of interest, tracking and forecasting these strategic point values is essential Business Analytics for SMM © 2011 SimaFore LLC Page 5 www.simafore.com SimaFore 2011 White Paper Series 3. Using Predictive Analytics for Cost Modeling 3.1 Business Objective: Enable rapid data-driven cost forecasting The key to successfully applying analytics is to correctly describe the business problem and establish meaningful objectives for the analysis. In this application, cost modeling was employed by L&L’s Purchasing team to develop estimates of the overall cost of a given product to the company so that the Sales and Marketing teams could develop winning and profitable bids. Without predictive analytics however, these cost models were “built” by visually comparing several dozen input factors which could potentially affect overall costs of a product and combining this with past experience or “gut-feel” to develop heuristics for cost forecasting. Cost Model before using Predictive Analytics Figure 1. Cost Modeling included tracking 40+ input factors such as indices from Bureau of Labor Statistics, Petrochemical prices, Shipping indices, related commodity prices, Producer Price Index, etc. The work was tedious and subjective. Predictive analytics was employed to address two main business problems: 1. Build a reliable model that connected the factor inputs to product costs 2. Verify if this model then could accurately forecast product costs Business Analytics for SMM © 2011 SimaFore LLC Page 6 www.simafore.com SimaFore 2011 White Paper Series The objective was to provide a robust cost model with quantitative basis that eliminated cumbersome and error-prone visual tracking of cost indices and replace it with an automatic process and more objective data driven decision making, with contributions from fewer, more clearly related factors. 3.2 Data Sources: Mostly public domain From a technical perspective, the objective can be translated as follows: Build a model that connects factor inputs to product costs Verify the accuracy of this model on a test sample of data Employ this model to make predictions about product cost when key factor inputs vary. The actual data consisted of 42 time sensitive independent variables or “predictors” and 7 dependent variables or “responses”. This data was available on a monthly basis for the previous 48 months. Data used to build the Predictive Analytics model Figure 2. Cost Modeling data set. Nearly 80% of the data was public domain information. The rest of the data came from purchased sources and in-house manufacturing information. 3.3 Analytics Process: A 4-step approach The sequence of steps followed a process starting with establishing and verifying the validity of the model is a well-known 4-step process which includes Data preparation, Variable Reduction, Model setup and Model Validation. The schematic in figure 3 shows the different steps within this process. There are typically several competing algorithms which can effectively perform the task of modeling. In such a situation, inputs from the end-users of the model, the Purchasing team in this case, becomes very important in the final model Business Analytics for SMM © 2011 SimaFore LLC Page 7 www.simafore.com SimaFore 2011 White Paper Series selection. The choice depends upon two factors: can the model structure be easily understood and is it providing value. In this case, the choice of the algorithm was dictated by the first factor. 4-Step Analytics process of building the cost model Figure 3. Based on input from the end-users, a Forward Stepwise Regression model was chosen for its easy interpretability. This 4-step approach is fairly standard and can be implemented for any business issue that can be addressed with predictive analytics. However, this is also the part of the analytics implementation where expertise is most needed and can consume significant amount of time and resources that SMBs seldom have. SimaFore consultants are in the unique position of being industry veterans who straddle several industry verticals and the analytics horizontal. It is SimaFore’s goal to make analytics accessible and affordable to the SMB sector. To this end, SimaFore’s analytics portal, visTASC, is aimed at developing awareness for analytics consumers (SMBs) and helping analytics users (analysts) build competency. Business Analytics for SMM © 2011 SimaFore LLC Page 8 www.simafore.com SimaFore 2011 White Paper Series 3.4 Model Deployment: An App-based approach For companies which do not have the analytics tools and expertise, and cannot afford to spend the resources to acquire them, SimaFore offers a very practical alternative. SimaFore will implement the 4-step process described above on a consulting model where our experts will work closely with the SMB teams. The objectives of this part of the solution are: to crystallize the business problem, identify, collect, clean, and organize the pertinent data build and validate the model. SimaFore experts will develop the model using algorithms provided by open source tools, such as RapidMiner, R and KNIME, which will result in a substantial cost save for the SMB. The modeling technology chosen is completely transparent to the business. Once the model is developed and accepted by the SMB as a valid representation of the business problem, the solution can be hosted on a public cloud by SimaFore. The SMB will have access to the model via a basic dashboard or a Custom App as illustrated in figure 4. The app is specific to a business problem that is addressed by the predictive analytics model, such as the one described in this report. Screenshot of a customized App dashboard Figure 4. Dashboard of a predictive analytics model App that business users can use to simulate what-if scenarios and generate predictions for parameters of interest. The model resides in a cloud server and is updated/validated regularly. Business Analytics for SMM © 2011 SimaFore LLC Page 9 www.simafore.com SimaFore 2011 White Paper Series 3.5 About Cloud computing and deployment Innovation, it is said, happens when two distinct but related technologies merge. This merger generates value that is greater than the individual value each would separately bring. This has been the case since the gasoline internal combustion engine merged with petroleum refining leading to the democratization of personal transport, all the way to the more recent peer-to-peer sharing of documents with high density memory chips, leading to the mp3 player. Gartner technology consultants, who develop the technology hype cycle, have in 2011 positioned predictive analytics at the “mature” stage of its growth. However cloud computing is still evolving and going through its “peak of expectations” phase. But it is quite possible that with the merger of predictive analytics with cloud computing, the resultant technology will catapult itself to the “productivity” phase much earlier than cloud computing alone. What is Cloud Computing? Cloud computing refers to any activity where the computational resource (memory, disk space and/or CPU time) is shared by several users and the physical hardware itself is located miles (or several continents) away. By allowing multiple users or “tenants” share resources, the provider can extract more value from a fixed investment, as not all users may need all the resources at the same time. This will in turn allow the cost of the resource to be spread across multiple payers resulting in significant reduction in cost for each tenant. Merging basic analytics models with cloud computing The potential applications of the model described in this report are ideally found in businesses which are beginning to discover predictive analytics. The computational resource demand for the analytics model described in this report is minimal. However for other more complex models, the demand – mostly memory and CPU, can be quite substantial. James Taylor of Decision Management Solutions [7] recommends that the ideal cloud computing solutions for such applications tend to be pre-packaged. This means that the model is pre-built and solution is readily deployed via a dashboard. Whatever minimal computation is needed takes place in the cloud. This is precisely the scenario described in this report. App Dashboard Business Analytics for SMM © 2011 SimaFore LLC Cloud Server Page 10 www.simafore.com SimaFore 2011 White Paper Series 4. Conclusions: ROI from using predictive analytics As was shown in Figure 1, the process of building price estimates for complex parts was very cumbersome. Data from several disparate sources had to be first assembled. Then detailed charts had to be generated for all the 40-odd variables. The final portion of cost estimation involved digesting this information, relying on past experience and estimating a composite number for the ultimate cost drivers. The process was tedious and prone to subjective assessments. In our case, it took L&L purchasing team typically eight to twelve hours to develop a detailed cost model. Using predictive analytics for this process results in two main advantages: 1 Reduction of 40+ parameters to a core set of 4 manageable variables more easily accumulated and forecast 2 Rapid evaluation of the impact of changes to these parameters on the final cost drivers This has resulted in a significant reduction of cost forecast development times along with increases in forecast interpretability. Additionally, since the model has been tested and validated, there is very little room for subjectivity in decision making. It can take a few minutes to run a report for a new cost forecast and going from data to decision making will be in the order of hours rather than weeks. Cost Modeling effort - Traditional versus Analytics based Cost Modeling Effort Effort (hrs) 12 10 8 6 4 2 0 Traditional Predictive Analytics Figure 5. Advantages of using analytics include not only time saved in developing a new cost forecast, but also reducing the subjectivity in decision making process. Business Analytics for SMM © 2011 SimaFore LLC Page 11 www.simafore.com SimaFore 2011 White Paper Series The National Center for Manufacturing Sciences (NCMS) estimates that within the industry, of the 300,000 manufacturing companies, 95% are considered to be “small and medium” which account for two thirds of the total employment in the sector [6]. Advanced analytics tools for engineering and design (such as modeling and simulation) are effectively used by the larger companies (e.g. Ford, Boeing, Caterpillar, etc.) to reduce time to market and improve product quality. 97% of the companies which use advanced analysis tools say that they cannot function without these tools. However the smaller and medium sized companies – the so-called “missing middle” have very little or no access to these productivity tools: whether they are for engineering modeling and simulation (M&S) or business optimization, and are being left behind in the competitive race. Business analytics tools are essential for commercial leadership. In a survey conducted by SimaFore [8] among 100+ manufacturing companies in the summer of 2011, it was seen that Cost estimation and purchasing activities, followed by Sales and Marketing were the most receptive functions for analytics tools among the SMMs. There is clearly a need to provide accessible analytics of the type described in this report. This can be a potential game changer for competitiveness of the manufacturing industry. The key takeaways from this report can be summarized as follows: • • • Using a combination of well-established data analytics process, open source technology and cloud computing, we can build affordable analytics tools customized for specific business needs Employing predictive analytics for common business problems, such as cost estimation and forecasting can significantly reduce the time to go from data to decision making while simultaneously improving adoption and accuracy Adopting the data-driven approach that predictive analytics enables will reduce the subjectivity in decision making Business Analytics for SMM © 2011 SimaFore LLC Page 12 www.simafore.com SimaFore 2011 White Paper Series References 1. http://practicalanalytics.wordpress.com/2011/08/13/analytics-as-a-serviceunderstanding-how-amazon-com-is-changing-the-rules/ 2. Business Analytics: An inside perspective – Analytics in action at Oberweis Dairy. Insights from a webinar in the SAS Applying Business Analytics Series, February 2011. 3. http://www.informationweek.com/news/smb/hardware_software/22940262 0?pgno=1 4. http://www.noblivity.com/blog/2011/03/redefining-small-business-analytics/ 5. http://www.informationweek.com/news/smb/hardware_software/22940309 1?pgno=1 6. http://www.ncms.org/index.php/multimedia/podcasts/ 7. Predictive Analytics in the Cloud, James Taylor, CEO, Decision Management Solutions, 2011 8. http://www.simafore.com/download-whitepaper-smb-analyticssurvey?utm_campaign=costmdlwp&utm_source=whtppr About SimaFore SimaFore was established with the express objective of helping small and medium businesses find the right analytics tools for the right problems in order to convert data into information assets. Our team of experts come from different backgrounds but the analytical techniques each have applied in their experience have several common threads: using data to understand cause and effects, building models to simulate systems or processes, and using a combination of analytics tools for forecasting. Our team shares a common vision - to help businesses remove the complexity involved in applying and deploying analytics. To make analytics accessible and affordable to anyone who has data. We invite SMBs to sign up for a preview of our specialized analytics portal here. Contact SimaFore 330 E Liberty Lower Level Ann Arbor, MI 48104, USA 248-686-2100 http://www.simafore.com info@simafore.com About the Authors Bala (BR) Deshpande has 15 years of experience in the biomedical, automotive and consulting areas. He holds an MS in Mechanical Engineering (University of Utah), an MBA (University of Michigan) and a Ph.D. in Bioengineering (Carnegie Mellon University). Before founding SimaFore, he was involved in engineering consulting, software publishing and automotive product development for more than 12 years. Follow him on Twitter at http://twitter.com/#!/complexman Connect with him on LinkedIn: http://www.linkedin.com/pub/br-deshpande/6/302/861 Steven Reagan has 18 years of experience in the methods and practice of computational methods. He holds a Ph.D. in Mechanical Engineering (The University of Virginia). He is currently the manager for computational modeling at L&L Products, before which he was involved in crashworthiness research at Ford Motor Company. Some of his research and modeling interests include use of high performance computing for the numerical modeling of nontraditional problems/processes, statistical design exploration, nonlinear transient modeling, and optimization. Connect with him on LinkedIn: http://www.linkedin.com/home?trk=hb_tab_home_top Business Analytics for SMM © 2011 SimaFore LLC Page 13 www.simafore.com