Supply Chain Analytics

Violette Berge

Energy Analyst & Optimization Consultant
Artelys Canada

Europe: Benefits of Integrating Regions and Energies to the Security of Supply

The European Climate Foundation (ECF) is an organization whose aim is “to help Europe foster the development of a low-carbon society”. Last year, ECF has assigned Artelys and its partners to analyze least cost-lowest risks infrastructure investments to meet gas security of supply in Europe. This is particularly relevant, in a context with a substantial risk of supply disruption and, in turn, major investment projects being contemplated.
To address this issue, Artelys has used advanced hourly supply-demand equilibrium models and capacity expansion algorithms for gas and electricity grids for all European countries. In addition to evaluating the resilience of Europe to gas supply disruptions, Artelys has analyzed what are the benefits of coordinating gas and power grids in those places where gas security of supply concerns do occur.


Violette Berge is an energy analyst and optimization consultant at Artelys Canada in Montreal. Within her work at Artelys she took part in several studies about electricity and gas network models, including the development of a gas-electricity model used for resource adequacy analyses and a tool for optimizing hydro power generation for a North American IPP. She was also involved in the design of an algorithm based on statistical analysis that simulates forecasting errors on wind and solar power generation. She graduated from ENSTA ParisTech Engineering School and the Parisian Operations Research Degree

Teresa Bianchi-Aguiar


Quantifying Promotional Effects for Inventory Management in Fresh Products

In Portugal, more than 40% of grocery sales come from promotional items, and this number is expected to increase with the proliferation of the online market and the easy access to information on discounts and promotions. Therefore, promotional reality is no longer residual, repositioning itself to the forefront of operational success.
This talk will address the following topics:
– How to identify and quantify the key-factors of a promotion that have impact on sales?
– How to use past promotional data to generate commercial insights to help negotiate future promotions?
– How to generate promotional sales forecasts?
– How cannibalization can affect sales of non-promotional products?
– How to manage replenishment during promotions, taking into consideration the fresh reality, where the excess stock of a promotion can be translated into high shrinkage rates?
We have worked with a European food retailer that sells more than 70k SKUs in over 200 stores. The company has a promotional intensive strategy combined with a strong loyalty program that already covers over 70% of the country’s households.
This project’s objective was to develop an approach to sales forecasting and inventory management for promotions, focusing on fresh categories – fruits and vegetables, meat and poultry, fish, and charcuterie. Its motivation was twofold. Firstly, there was no analytical background to choose future promotions and promotional sales forecasts were generated by manually finding mirror campaigns, in a non-systematized and non-monitored process. Secondly, replenishment decisions did not consider the uncertainty in the promotional forecast neither the perishability of the products.
We have developed an integrated sales forecasting and inventory system. Promotional sales history is analyzed with an econometric model that uses a non-linear multiple regression to relate all promotional factors and to compute each factors’ elasticity (for instance, level of discount, type of communication and discount). Elasticities are the basis for generating commercial insights and for promotional forecasts. The forecasts of non-promotional products are also adjusted taking into consideration the effect of cannibalization, detected using econometric models at the SKUxSKU level. The inventory system is based on the principle that there are different periods during each campaign that should be managed differently and dynamically, for instance in terms of service level. It also takes into consideration the uncertainty of the forecast and the perishability of the products.
Besides the strong analytical emphasis, the success of our approach was reinforced with proper processes and training, namely a close validation of forecasts using visual analysis of historic promotions and mirror campaigns; forecast revalidation based on initial sales and daily monitoring of key KPIs, such as sales, stockouts and sales coverage.
To validate our approach, we conducted a pilot test. Promotional forecasts with the econometric model were more accurate and unbiased, leading to earnings greater than 8p.p. in accuracy. We were also able to reduce stockouts and increase sales across all categories with reduced stock levels (up to 13% sales increase and a 6p.p. stockout reduce in one of the categories).


Teresa Bianchi-Aguiar is Manager at LTPlabs, a management consulting company that applies advanced analytical methods, and a researcher at the Center for Industrial Engineering and Management of INESC TEC. She holds a PhD in Industrial Engineering and Management by the University of Porto and was a Visiting Graduate Student at University of São Paulo and the UCLA Anderson School of Management. Her main activity is combining advanced analytics with business expertise to solve complex real-world problems, especially within the retail industry. She has large experience in shelf space planning, demand forecasting and inventory management. She was a finalist from the Daniel H. Wagner Prize in 2016.

Bharath Gangula


The Race to Autonomous Driving: Leveraging Maximum Difference Scaling and Glm to Prioritize What In-vehicle Technologies Consumers Want

With ever evolving autonomous in-vehicles technologies, it is increasingly difficult for manufactures to prioritize what new features are most useful to the consumers. To help manufactures better position their product development strategies, we conducted a study of over 20,000 consumers in 17 major markets. One key outcome of this research was to determine consumers’ prioritization of various new in-vehicle features. Using Maximum Differential Analysis we determined consumers’ relative ranking of 32 new in vehicle technologies. Using Sawtooth’s best-worst scaling, respondents were presented with a gamified method of choice where they were asked to select which of five features was the most and the least useful technology. Each of the 32 features was presented two times in a random order. Consumers’ choices were then aggregated to yield an overall utility score for each of the 32 technologies, which were then rescaled to 1,000. To determine whether different consumers view the usefulness of these technologies differently, we applied a generalized linear model (GLM) where the 32 utility scores served as the dependent variable and country, age, gender, income, location, type of driver (based on daily driving distance) brand owned (premium or non-premium) and their interactions served as independent variables. Results show that safety related items are perceived as most useful across all markets. Especially, safety technologies which automatically recognize presence of objects on the road and avoid collision, inform driver or automatically block him from dangerous driving situation, and, automatically take steps in case of emergency are examples of the most preferred technologies across all markets. Our findings also indicate that interest is these technologies is not just for the young buyers. In fact, across all markets, older consumers, those living in rural areas considered safety features more useful. Preference scores across various consumer segments, almost universally, show that safety related features are the most important. A more detailed level of analysis based on multi-variate tests (MANOVA) shows that older generations, females and those living in rural areas prefer safety features more than their respective counterparts. Similarly connectivity and service enabling technologies though less preferred overall, are more preferred by younger generations, urban population, mid-high income population and premium brand owners than their respective counterparts. These results can help manufacturers to prioritize resources towards technologies that matter to the consumer and better position the benefits of these technologies in their marketing message (e.g.: autonomous as safety) to win the trust and confidence of the consumer. These optimal product and marketing strategies help automakers survive in this rapidly evolving, hyper-competitive world of wafer-thin margins and win the trust of 21st century consumer.


Bharath Gangula, PhD, is a subject matter specialist and leads multiple research initiatives in automotive and manufacturing sectors in Deloitte Services LP’s Center for Industry Insights. Recent research efforts driven by him are on topics ranging from consumer choices and attitudes in automotive sector, strategies and capabilities driving high performing manufacturers, global manufacturing competitiveness, and autonomous technologies in automotive industry.

Wei Tan

Director, Advanced Planning Systems
Seagate Technology

Seagate’s Global Demand and Supply Balancing Model

Seagate has embarked an end-to-end supply chain transformation for a couple of years. In this presentation we’ll share with you the key successes that we’ve achieved in the area of modeling and planning the entire hard drive supply chain that spans from customer demands at hubs, the distribution network, manufacturing footprint, to the network of suppliers. You will hear a few key lessons that we’ve learned from implementing a global advanced planning solutions of this magnitude. You will hear the unique challenges that a hard drive manufacturing company faces. You will also have an opportunity to learn the sophisticated demand and supply balancing model that we’ve implemented. The success of the program enables Seagate Technology for rapid and accurate response to changing customer and market conditions and providing end-to-end visibility on demand (customer and dependent), supply (finished goods and components), and key constraints across the entire supply chain that had never been available in the past. The hard drive supply chain is unique and complex. On one hand the customers want to have their commits preserved from week to week regardless of supply chain fluctuation and uncertainty. On the other hand the manufacturing processes involve long lead time, multi-level bill-of-material, yield and waterfall, and alternate choices of component supplies in conjunction with many other constraints. Those and other complexities make the off-the-shelf supply chain planning solutions not suitable for hard drive supply chain. In this talk you will have a close look at the solution and model that we’ve implemented. The demand and supply balancing model is based on attribute matching algorithm that considers all possible alternate bill-of-materials, supply options, and material and capacity constraints with the objective to meet the customer demand on time. Specifically we’ll share with you the below key topics: 1. Complexity of the hard drive supply chain; 2. The demand and supply balancing model that we implemented; 3. Demand and supply commits preservation; 4. Weekly planning process; 5. Key lessons learnt; and 6. Key successes achieved. We’ve successfully applied our solution to one of the key business segments – notebook drives. We have realized many short term and long term benefits including: a. Overall inventory savings in the range of 5-15%; b. Significantly improved planning lead-time; and c. Enhanced visibility that offers previously unimaginable granularity, timeliness and driving for actions in both demand and supply functions. We are continuing on our journey to roll out this critical solution across the entire enterprise.


Wei Tan is a Director of Advanced Planning Systems within supply chain organization of Seagate Technology. He is a long time Operations Research practitioner and has over 20 years of industry experience in supply chain and distribution, focusing on development and implementation of analytical models and processes in supply chain planning, statistical demand forecasting, manufacturing and distribution. He has worked in many industries including automotive, CPG, and high-tech. He received a PhD in Industrial Engineering and MSc in Computer Science from the University of Southern California.

Xin Yuan

Dun & Bradstreet

Tier N- Supply Chain Discovery and Risk Assessment

No organization operates in a vacuum. Understanding the business network and relationships is critical to managing risks and identifying opportunities. In this presentation, I’d like to showcase the Tier N model. This analytic solution provides a systematic way of assessing the business risk in totality incorporating the risk of the company itself and the risk of other businesses in its network which includes suppliers and customers. The solution has two major components: one is Tier N business relationship network discovery, and the second – risk assessment in totality with network effect. Network discovery identifies business relationships with buyers and sellers and links businesses together into a network. It not only links the entities that a company does business with directly (which is referred to as a Tier 1 business relationship) but also the entities that its Tier 1 businesses do business with directly (which is referred to as a Tier 2 business relationship in relation to the original subject), and so on and so forth until a Tier N business relationship is discovered. The discovery process is recursive mapping using data on buyer-seller relationships. The result is a Tier N business relationship network database that includes relationship chain as a unique path from the Tier N business to the subject business. The second component, which is risk assessment in totality with network effect, evaluates the risk of a company based on the risk of all the businesses in its network. Company’s risk level is determined by the risk of each business in the network, how critical it is to the operation of the analyzed enterprise, the distance to its Tier N business, and other factors that can describe the properties of the business relationship network. The model can be applied to a variety of business use cases. For example, this solution can be used to strategically mitigate supply chain risks; evaluate the risk of supply chain disruption for an insurance contract; monitor business connections to ensure their compliance with regulatory requirements (such as avoiding dealing with entities involved in conflict mineral mining, human trafficking or entities listed on OFAC, EPLS, etc.); and identify potential counterfeit exposure. This analytic solution can also be used to assist product marketing within a business network as well as for FDA inspections. Last but not least, investment banks have started to analyze business relationships to make decisions based on the information beyond traditional financial statement analysis.


The presenter leads analytic innovation at Dun & Bradstreet. Her primary responsibilities are prototyping new analytic solutions, deriving business insights from new data sources, applying new statistical methodologies to solve business problems, and macroeconomic insights and econometric forecasting. Her current focus is capital market analysis, supply chain management solution, unstructured data analytics, location analytics, and econometric forecasting. She manages a team of Ph.D. statisticians, economists, and engineers. She also establishes analytic innovation partnership with industry leaders and top universities. Prior to her current role, she has broad experiences in building standard scores and custom solutions for customers across different industries, including financial services, telecommunications, insurance, Hi Tech and manufacturing. The presenter has a Ph.D. degree in Economics from University at Albany – State University of New York, and an undergraduate degree in International Trade from Renmin University of China.