2011 Inductees

Tripper Dickson

Company: Norfolk Southern
Category: Optimization

As a necessary part of shipping intermodal freight, Norfolk Southern’s (NS) customers reposition their empty containers around the country in order to move them to markets where the freight is reloaded for shipment. Customers reposition these empties without regard to NS cost structure.  In order to capture the loaded freight business, NS was shipping such empty containers at special low rates and was incurring significant losses on this aspect of the business.

To address this problem, Tripper created an optimization model that maximized NS’s margins by optimally routing the empty containers in such a way that clients’ requirements are satisfied at no increase in the cost to clients. He started with NS’s three largest clients: each week the clients send in their empty distribution plan for the following week – terminals they plan on using and volumes by lane.  Within five minutes (utilizing VBA macros) NS sends them an optimized distribution plan with suggested lane changes.  Their planning and cost is unchanged and NS margin is maximized.  By influencing their forecasted plan, NS was able to immediately capture these gains going forward.

The model was implemented for the last 4 weeks of Tripper’s internship and immediate margin improvements of $65,000 were recognized.  He then trained NS employees to run the model and it will continue to be used on a weekly basis to influence the distribution of virtually all empty containers moving on Norfolk Southern railways. It is estimated that the total annualized margin improvement from implementing the Tripper’s model will exceed $3 million.

Paul Rhynard

Company: McKinsey & Co
Category: Optimization with binary variables

In periods of excess market capacity, Paul’s client did not properly utilize its six production facilities: all facilities continued to run at slower rates regardless of the financial impact and without considering alternate asset optimization strategies.

Paul built a Solver model that determined the optimal operating rate for each facility under various market conditions.  Given the fixed and variable cost structures of the different facilities, some were better off running at full capacity while others should be slowed or shut down, minimizing costs and maximizing profitability.  The model optimized the production decisions (modeled continuous variables) and the facility closure decisions (modeled as binary variables) utilizing the facility costs, supply, and demand inputs out to 2020.

The model resulted in significant benefits to the client: given the flexibility of the inputs (forecasts, costs, shutdown costs, pricing, etc.) and the speed at which it calculates (under 1 second), the management considered multiple market scenarios and set on a robust course of action. Best of all, the model enables the operations and finance departments of a $2 billion business unit to flex their operations to save millions (potentially hundreds of millions) of dollars a year in excess supply situations.  It also helped to guide future capital expenditure decisions as it brought to light how important the supply/demand balance really is to the organization.

Ryan McCollum

Company: Accenture
Category: Time series forecasting, regression, pivot tables

 Ryan worked for Accenture in their Supply Chain Management Consulting practice where his client was a major cable provider in the US. Ryan’s task was to improve client’s forecast accuracy for cable boxes: low forecast accuracy has been driving excess equipment purchases and unnecessary inventory levels.

Ryan analyzed the client’s historical demand data, to understand correlations between the demand for different cable box types and technologies and ultimately built a model for the client to use when creating their forecasts.  His analysis included taking the large amount of data, using pivot tables to sort the data into meaningful and useable data segments, regression analysis to understand correlations between product types, and ultimately the development of time-series forecasting models using ARIMA methodologies.

Keeping a portion of the data in the hold-out validation sample, Ryan established that his model improved forecast accuracy from the client’s historical 20-50% up to about 80%. Such improved forecast accuracy is estimated to result in some $30M reduction in excess inventory among other cost savings such as reduced storage and transportation costs.

Whit Walker

Company: CNN.com
Category: data reporting, pivot tables, optimization

Whit worked for CNN.com’s Internal Strategy Group, focusing on Product Strategy & Partnerships. This group did not have an accurate picture of the value exchange between CNN.com and 18 strategic partners that provide content to the “highest margin” sections that are essential to CNN Digital’s long term growth.

These content partnerships were initially structured in an ad hoc manner that focused on driving site traffic and sharing content, but didn’t initially consider the financial value derived from each. Whit created a quantitative way to analyze the performance of these partnerships, treating each as though it was a separate company.

Whit used the available site statistics to establish a final relative value measurement that identified which partnerships were highest or lowest performing and also identified which ones needed restructuring.

After quantifying the value exchange between partners, Whit developed a package of five “what-if” analysis models to help CNN further understand partner value creation mechanics, restructure deals, negotiate favorable deal terms and inform staffing decisions based upon the restructured agreements.

Quoting CNN’s Digital Partnerships Manager, Chris Finan: “Whit has changed our approach to partnerships: taking us from a reactive stance to a more proactive and strategic approach… [His] model will help us determine the success and revenue implications for each of our partners across CNN.com.”

Katherine Gordon Murphy

Company: UPS
Category: Hypothesis testing, Pivot Tables, Monte Carlo simulations

Kate interned at UPS where she performed a strategic assessment of the international dangerous goods (IDG) shipping market. To date there is no central tracking of the movement of dangerous goods and UPS wanted to know the size, potential and its share in the IDG market. To estimate the market size, Kate used a global trade database and worked with experts to estimate what percentile range of each commodity was hazardous. She then ran a Monte Carlo simulation which produced risk profiles (histograms) of the complete IDG market size.

After analyzing the external environment of international dangerous goods shipping, Kate looked at the role it plays internally for UPS. UPS views IDG as a strategic product that 1) leads to incremental non-niche shipping and 2) attracts clients that are more profitable than “average” UPS clients. Interestingly, these hypotheses had never been tested. Kate analyzed UPS’ practices related to IDG and discovered that a “natural experiment” occurred in the data between 2008 and 2010: some of the clients that were pre-approved for IDG shipping by UPS never used the service. Kate used pivot tables and hypothesis testing to compare the differences in shipping volumes and profitability between the pre-approved clients who shipped IDG and those who did not. While the results must remain confidential, Kate’s work led to very interesting conversations about the future of the IDG product offering and plans for its expansion.

Ilya Teslenko

Company: Navy Federal Credit Union
Category: Time Series Analysis, Linear regression, Optimization

 Over the summer Ilya worked in the Lending Department to improve the forecasting models of losses in consumer portfolio. The current approach was to first forecast the outstanding balance of the portfolio, and then use another model to forecast the loss as a percentage of the balance. This forecasted loss percentage was applied to forecasted portfolio balance to get the absolute loss number.

Ilya first attempted to create a single model that would forecast the absolute loss, but that attempt was not successful. He therefore kept the general structure with two models unchanged.

To forecast the outstanding balance he built separate models for the three main segment of the portfolio, Consumer loans, NAVCheck and Leasing, using time-series ARIMA models with auto-correlation, seasonality and trends. Aggregating the prediction of his models yielded a noticeable improvement in the quality of the forecast for outstanding balance. Interestingly, however, when Ilya applied the existing loss percentage model forecasts to his new outstanding balance prediction, the resulting forecast of the absolute loss became less accurate. As Ilya discovered, this was because the first existing model over-forecasted the balance, while the second under-estimated the loss percentage, so that when taken together they produced a forecast that was “just right.”

Ilya therefore also redid the model that forecasted the loss percentage. The existing model for losses was weighting the past 3 years of historical data equally. Ilya used Solver to find optimal weights minimizing the sum of squared errors; as expected Solver gave more weight to the recent data. These weights were used for the modified model for forecasting the loss percentage.

Ilya’s new model for outstanding balance combined with modified model for the loss percentage proved to produce a much better forecast in terms of squared errors than the current model in place. Ilya’s model was launched in test-run as a candidate to replace the current one; its performance will be evaluated at the end of the year.

2010 Inductees

Brandon Smit

Company: Amazon.com
Categories: Linear Regression, Optimization

Over the summer, Brandon Smit worked for the Shoes In-Stock Management team at Amazon.com on two projects: the Site Merchandise Forecasting Model and the Liquidation Model.

In the first project, he analyzed sales data for approximately 10,000 shoes and built four linear regression models to predict 20-week demand at the style-color-size-width level, one for each of the four demand segments. When tested on a holdout sample, his models outperformed those historically used by Amazon.com in terms of forecasting accuracy (as measured by Mean Absolute Percentage Error, or MAPE). Brandon’s models are currently used to inform procurement decisions.

In the second project, Brandon analyzed the economics of a liquidation decision. He built a model to determine at what point an item should be withdrawn from sale and sold to liquidators to recoup a fraction of the procurement cost. The model used a VBA macro that performed a Solver optimization across thousands of unique products to calculate the breakeven period demand and number of units to liquidate.

Brandon’s recommendations led to the liquidation of nearly 100,000 shoes; that freed up about 5% of warehouse capacity and provided capital to invest in products for the upcoming holiday season.

Charles (Chip) Hogge

Company: Deloitte
Categories: Pivot Tables, Optimization

 Chip Hogge worked with the domestic expatriate operations of a leading health insurance provider. The client did not have an accurate picture of its organizational processes or how resources were allocated across 15 business units and 10 global locations. In order to transition to a more competitive cost structure without sacrificing service levels, the client needed to determine ”who is where” and ”who does what” to facilitate an organizational and business process redesign. The original employee-level data were stored in an Excel file and tagged with cost center information. Chip used iterative lookup functions to assign process categorizations and outsourcing priority to each employee. Several pivot tables were then used to organize the data and provide different views of the overall organization, including by line of business and process, location and employee type. Chip then developed an optimization model that minimized compensation spending based on numerous constraints established through interviews with Business Unit leaders across the organization. The client used this information and recommendations to initiate vendor discussions and develop a reorganization plan consistent with its strategic priorities. These initiatives had an immediate impact on profitability by reducing compensation spending by 30% and total SG&A spending by 10%.

Stefan Talman

Company: Delta Airlines
Category: Regression 

Stefan Talman worked in Delta Airlines’ Revenue Management department. Because its existing inventory demand forecasting system was based on historical data, it couldn't capture/plan for the demand upswings for new large events. For example, because the Super Bowl is in Honolulu (airport code HOU) this year, demand forecasting for flights to HOU based solely on historical data would be clearly inaccurate. Stefan used multiple regression to describe the change in revenue for the days before and after an event (accounting for day-of-week and seasonal effects) using a set of independent variables (event size, market share, airport size, event type, etc.) for some 150 events in various market types (e.g., hub and spoke) and event types.  The final model was based on event size adjusted by market share and airport size (both in seats/week). Stefan’s model was adopted by the inventory management team and will continue to be refined as new large events occur.

Bryan Lieber

Company: Target
Categories: Pivot Tables

Although Target had high-level chain-wide visibility into the future profitability of its Pharmacy business, it did not have a good understanding of how the Pharmacy P&L varied across each of its nearly 1,600 stores. Using pivot tables, Bryan Lieber analyzed historical prescription ("script") sales data by store to understand what store characteristics (store size, age, region/geography, etc.) had the biggest impact on sales growth. Ultimately, he developed a sales-maturity curve based on age for each of Target's four regions. This allowed him to create five years of annual sales projections for each of Target's approximately 1600 pharmacies. Bryan combined these projections with store-specific financial data and long-term industry trends to develop a model that projects annual profitability for each of the next five years on a per-store basis. He then segmented the store population into four classifications based on their profitability and growth levels: Stars, Cash Cows, Questions, and Dogs/Sub-Optimals. This segmentation enabled much more granular reporting about how each of these various segments contributed to chain profitability, particularly highlighting the large gap between the best- and worst-performing stores. Finally, Bryan reviewed the time-of-day purchasing patterns for the "Dog/Sub-Optimal" stores to identify candidates for the firm’s hour reduction initiative and estimated the annual EBIT improvement from this initiative.

Going forward, Target will use the Pharmacy Segmentation model in three ways. First, segments can be incorporated into monthly reporting to provide additional visibility into Pharmacy profitability. Second, the model will be used to refine the target list of stores for hour reduction programs and calculate potential savings. Third, the Pharmacy team will use the segmentation to be more selective when rolling out new strategic initiatives. Rather than deploying new strategies to all stores in a given geography, with the help of Bryan’s model, they can target the type(s) of stores in the areas that best match their objectives.

Geoffrey Coffield

Company: BCG
Categories: Data Reporting, Pivot Tables

Geoff Coffield worked with the manufacturing IT group of a large pharmaceutical company on two projects. In the first project, he was tasked with figuring out how and where money was being spent on contract employees and identifying opportunities for savings. From the company’s disorganized data, he created large pivot tables and charts using translation matrices. From the results, he was able to determine how, where, and by whom money was being spent on contract employees and to a certain extent what the employees were doing. Based on that analysis, he identified an addressable opportunity of about $18 million in annual spending and developed an approach to realize these savings. This realization process is currently under way and on track.

In his second project, Geoff supported a global reorganization affecting thousands of stakeholders around the world. He built a multidimensional employee survey to track progress across the different operating groups and geographies. He then created a comprehensive dashboard that allowed the client to analyze data across relevant dimensions and identify opportunities for improvement. In particular, because the client’s manager who will own this moving forward is not experienced with Excel, Geoff had to automate nearly every aspect of the process to the point that the manager need only copy and paste data into the dashboard template. Since then, the survey and the dashboard have been used twice to help identify a number of opportunities the company is currently pursuing.

2009 Inductees

Ryan Rayborn

Company: Coca-Cola
Category: Regression and Pivot Tables

The forecasting system Ryan Rayborn constructed to predict sales of Coke products in Target stores outperformed their legacy system and is currently being subjected to side-by-side testing prior to roll out. More

Benjamin Potter

Company: Geico
Category: Regression

Ben Potter used regression analysis to make inferences about the extent to which lower prices led to higher online sales for Geico relative to Progressive. More

Josh Burke

 Company: Campbell
Category: Pivot Tables

To quickly and accurately quantify the benefits to Campbell Soup of implementing his suggestion to classify correctly that portion of trade liabilities that are non-current, Josh Burke applied a pivot table to data on the prior year’s 60,000 separate trade promotions. More

Kevin Graney

 Company: Natural Resource Partners
Category: Time Series

 Natural Resource Partners (one of the country’s largest owners of US coal properties) used to conduct simple best case/best guess/worst case analyses of projects until Kevin Graney came along and showed them how to use Crystal Ball to simulate a complete time series of future coal prices. The new, more sophisticated, approach allows NRP to better quantify the value of projects –especially those involving contingent contracts. More

Sihartha Sinha

 Company: Merkle
Category: Optimization

To decide how best to process 150+ million pieces of mail each day, Sid Sinha built Merkle a Solver optimization model. More

2008 Inductees

Roberto Brillembourg

Company: Kraft Foods
Category: Regression

Roberto Brillembourg used regression analysis to help Kraft Foods measure the effectiveness of its TASSIMO coffee continuity program in increasing consumption of TASSIMO coffee. More

Shankar Raja

Company: BCG
Category: Pivot Tables

During his summer internship with BCG, Shankar Raja segmented the US healthcare market using Pivot Tables and developed a comprehensive, market based, bottom-up revenue forecast to inform the post-merger, strategic planning of a client firm. More

Anand Rao

Company: Bain
Category:Pivot Tables

During his internship with Bain, Anand Rao used pivot tables to create first-ever organizational charts based on the reporting relationships and cost center assignments contained in a client firm’s 2500-record employee database. More

Christopher Richins

Company: Bain & Company
Category: Regression

To help a client firm price its transportation offering, Christopher Richins used regression analysis during his internship with Bain & Company  to build a model forecasting refrigerated trucking rates for shipping produce across the U.S.  More

Will Teichman

Company: Target
Category: Regression

In the process of creating Target’s first inventory of water consumption across all facilities, Will Teichman used regression analysis to identify the primary drivers of water usage for the purposes of prioritizing future water-saving initiatives. More

Anand Veeraraghavan

Company:
Category: Optimization

In order to select the best options for growth across five business units of an Indonesian energy company, Anand Veeraraghavan used a stochastic optimization model (built using Crystal Ball’s Optquest) that accounted for 25 key uncertainties and linkages across the business units.  More

2007 Inductees

Sarah Glass

Company:Merck/Schering-Plough
Category:Pivot Tables and Regression

Sarah Glass used Pivot Tables at Merck to consolidate patient-level prescription data for the $2B cholesterol drug ZETIA into 30+ sources of business as the basis for her regression-driven, bottom-up sales forecast. More

Alex Holsenbeck

Company: Embarq
Category: Regression

Alex Holsenbeck used regression to forecast stock-option exercise rates as a key component of his Crystal Ball model to forecast cash proceeds, taxes, and share dilution resulting from EMBARQ’s outstanding options. More

Bobby Jordan

Company: PRTM
Category: Pivot Tables

Bobby Jordan used pivot-table generated daily performance metrics to both guide and validate performance improvements his consulting team implemented for a distribution center shipping more than $120M annually. More

Kevin Kosefeski

Company: Danaher
Category: Optimization

Kevin Kosefeski used Solver to identify the optimum production schedule for a Danaher factory in China making hand-held digital test strip readers. More  

Gautam Sukumar

Company: Dell
Category: Pivot Tables

Because Gautam Sukumar used pivot tables to summarize the profitability of over 500K orders filled by Dell, the company now has an accurate understanding of configuration profitability.” More

Saul Yeaton

Company: Target
Category: Inference

Saul Yeaton used data analysis to convince TARGET that the 75% higher cosmetic sales in stores with in-store beauty advisors was due, for the most part, to how the advisors were assigned to stores. More

2006 Inductees

Christopher Langbein

Company: ConocoPhillips
Category: Regression

Chris Langbein used regression to forecast monthly working capital
needs for ConocoPhillips. More

Michael Lorence

Company: L-3 Vertex Aerospace
Category: Regression

Mike Lorence showed L-3 Vertex Aerospace how to use logistic regression and decision analysis to decide which projects to pursue. More

Douglas Polen

Company: Titan Virginia Ready-Mix
Category: Regression

Douglas Polen used regression analysis to determine the true source of profitability for Titan Virginia Ready-Mix’s business. More

Brian McCahill

Company: Danaher
Category: Optimization

Brian McCahill built an optimization Excel model the Danaher Corporation used to create a manufacturing strategy for a new $100MM OEM program. More

Matthew Kirby

Company: Home Depot
Category: Pivot Tables

 Matt Kirby used a Pivot Table to help Home Depot save millions of dollars. More

Rick Ramsey

Company: Gilbarco
Category: Time Series

 Rick Ramsey built a time series forecasting model to help Gilbarco forecast monthly sales of its main product, gas dispensers. More