1 Introduction
Through examples and a case study, we shall learn how to apply data analytics to supply chain management with the intention to diagnose and optimize the value generation processes of goods and services, for significant business value.
A supply chain consists of all activities that create value in the form of goods and services by transforming inputs into outputs. From a firm’s perspective, such activities include buying raw materials from suppliers (buy), converting raw materials into finished goods (make), and moving and delivering goods and services to customers (delivery).
The twin goals of supply chain management are to improve cost efficiency and customer satisfaction. Improved cost efficiency can lead to a lower price (increases market share) and/or a better margin (improves profitability). Better customer satisfaction, through improved service levels such as quicker delivery and/or higher stock availability, improves relationships with customers, which in turn may also lead to an increase in market share. However, these twin goals have the potential to affect each other conversely. Improving customer satisfaction often requires a higher cost; likewise, cost reduction may lower customer satisfaction. Thus, it is a challenge to achieve both goals simultaneously. Despite the challenge, however, those companies that were able to achieve them successfully (e.g., Walmart, Amazon, Apple, and Samsung) enjoyed a sustainable and long-term advantage over their competition (Simchi-Levi et al. 2008; Sanders 2014; Rafique et al. 2014).
- 1.
Seasonality and uncertainty in supply and demand and internal processes make the future unpredictable.
- 2.
Complex network of facilities and numerous product offerings make supply chains hard to diagnose and optimize.
Fortunately, supply chains are rich in data, such as point-of-sale (POS) data from sales outlets, inventory and shipping data from logistics and distribution systems, and production and quality data from factories and suppliers. These real-time, high-speed, large-volume data sets, if used effectively through supply chain analytics, can provide abundant opportunities for companies to track material flows, diagnose supply disruptions, predict market trends, and optimize business processes for cost reduction and service improvement. For instance, descriptive and diagnostic analytics can discover problems in current operations and provide insights on the root causes; predictive analytics can provide foresights on potential problems and opportunities not yet realized; and finally, prescriptive analytics can optimize the supply chains to balance the trade-offs between cost efficiency and customer service requirement.
Supply chain analytics is flourishing in all activities of a supply chain, from buy to make to delivery. The Deloitte Consulting survey (2014) shows that the top four supply chain capabilities are all analytics related. They are optimization tools, demand forecasting, integrated business planning and supplier collaboration, and risk analytics. The Accenture Global Operations Megatrends Study (2014) demonstrated the results that companies achieved by using analytics, including an improvement in customer service and demand fulfillment, faster and more effective reaction times to supply chain issues, and an increase in supply chain efficiency. This chapter shall first provide an overview of the applications of analytics in supply chain management and then showcase the methodology and power of supply chain analytics in a case study on delivery (viz., integrated distribution and logistics planning).
2 Methods of Supply Chain Analytics
Supply chain management involves the planning, scheduling, and control of the flow of material, information, and funds in an organization. The focus of this chapter will be on the applications and advances of data-driven decision-making in the supply chain. Several surveys (e.g., Baljko 2013; Columbus 2016) highlight the growing emphasis on the use of supply chain analytics in generating business value for manufacturing, logistics, and retailing companies. Typical gains include more accurate forecasting, improved inventory management, and better sourcing and transportation management.
It is relatively easy to see that better prediction, matching supply and demand at a more granular level, removing waste through assortment planning, and better category management can reduce inventory without affecting service levels. A simple thought exercise will show that if a retailer can plan hourly sales and get deliveries by the hour, then they can minimize their required inventory. One retailer actually managed to do that—“Rakuten” was featured in a television series on the most innovative firms in Japan (Ho 2015). The focus on sellers and exceptional customer service seems to have paid off. In 2017, Forbes listed Rakuten among the most innovative companies with sales in excess of $7 billion and market cap more than $15 billion.1 Data analytics can achieve similar results, without the need for hourly planning and delivery, and it can do so not only in retail but also in global sourcing by detecting patterns and predicting shifts in commodity markets. Clearly, supply chain managers have to maintain and update a database for hundreds of suppliers around the globe on their available capacity, delivery schedule, quality and operations issues, etc. in order to procure from the best source. On transportation management, one does not have to look further beyond FedEx and UPS for the use of data and analytics to master supply chain logistics at every stage, from pickup to cross docking to last-mile delivery (Szwast 2014). In addition, there are vast movements of commodities to and from countries in Asia, such as China, Japan, and Korea, that involve long-term planning, sourcing, procurement, logistics, storage, etc., many involving regulations and compliance that simply cannot be carried out without the tools provided by supply chain analytics (G20 meeting 2014).
The supply chain is a great place to apply analytics for gaining competitive advantage because of the uncertainty, complexity, and significant role it plays in the overall cost structure and profitability for almost any firm. The following examples highlight some key areas of applications and useful tools.
2.1 Demand Forecasting
Demand forecasting is perhaps the most frequent application of analytics to supply chains. According to the Chief Supply Chain Officer Report (O’Marah et al. 2014), 80% of executives are concerned about the risks posed to their supply chain by excessive customer demand volatility. Demand volatility causes problems and waste in the entire supply chain from supply planning, production, and inventory control to shipping. In simple terms, demand forecasting is the science of predicting the future demand of products and services at every level of an organization, be it a store, a region, a country, or the world. Demand forecasting is essential in planning for sourcing, manufacturing, logistics, distribution, and sales. The sales and operations planning modules of ERP systems help to bring several disciplines together so that forecasts can be created and shared to coordinate different activities in the supply chain. These include the obvious ones such as inventory levels, production schedules, and workforce planning (especially for service industries). The less obvious ones are setting sales targets, working capital planning, and supplier capacity planning (Chap. 4, Vollmann et al. 2010). Several techniques used for forecasting are covered in Chap. 12 on “Forecasting Analytics.”
One notable example of the use of forecasting is provided by Rue La La, a US-based flash-sales fashion retailer (Ferreira et al. 2015) that has most of its revenues coming from new items through numerous short-term sales events. One key observation made by managers at Rue La La was that some of the new items were sold out before the sales period was over, while others had a surplus of leftover inventory. One of their biggest challenges was to predict demand for items that were never sold and to estimate the lost sales due to stock-outs. Analytics came in handy to overcome these challenges. They developed models using which demand trends and patterns over different styles and price ranges were analyzed and classified, and key factors that had an impact on sales were identified. Based on the demand and lost sales estimated, inventory and pricing are jointly optimized to maximize profit. Chapter 18 on retail analytics has more details about their approach to forecasting and inventory management.
Going forward, firms have started to predict demand at an individual customer level. In fact, personalized prediction is becoming increasingly popular in e-commerce with notable examples of Amazon and Netflix, both of which predict future demand and make recommendations for individual customers based on their purchasing history. Several mobile applications can now help track demand at the user level (Pontius 2016). An example of the development, deployment, and use of such an application can be found in remote India (Gopalakrishnan 2016). As part of the prime minister’s Swachha Bharat (Clean India) program, the Indian government sanctioned subsidies toward constructing toilets in villages. A volunteer organization called Samarthan has built a mobile app which helps track the progress of the demand for construction of toilet through various agencies and stages. The app has helped debottleneck the provision of toilets.
2.2 Inventory Optimization
Inventory planning and control in its simplest form involves deciding when and how much to order to balance the trade-off between inventory investment and service levels. Service levels can be defined in many ways, for example, fill rate measures the percentage of demand satisfied within the promised time window. Inventory investment is often measured by inventory turnover, which is the ratio between annual cost of goods sold (COGS) and average inventory investment. Studies have shown that there is a significant correlation between overall manufacturing profitability and inventory turnover (Sangam 2010).
Inventory management often involves the planning and coordination of activities across different parts of the supply chain. The lack of coordination can lead to excessive cost and poor service levels. For example, the bullwhip effect (Lee et al. 1997) is used to describe the upstream amplification and variability in demand of a supply chain due to reactive orders placed by wholesalers, distributors, and factory planners. There are modern tools that can help reduce the effect of such actions by increasing demand visibility and sharing of information (Bisk 2016).
A study by IDC Manufacturing Insights found that many organizations that utilized inventory optimization tools reduced inventory levels significantly in 1 year (Bodenstab 2015). Inventory optimization plays a critical role in the high-tech industry where most products and components become obsolete quickly but demand fluctuates significantly. The ability to predict demand and optimize inventory or safety stock is essential for survival because excess inventory may have to be written off and incur a direct loss. For instance, during the tech bubble burst in 2001, the network giant Cisco wrote off $2.1 billion of inventory (Gilmore 2008).
Inventory management can be improved through acquiring better information and real-time decision-making. For example, an American supermarket chain headquartered in Arkansas had a challenge to improve customer engagement within several of their brick-and-mortar locations. Managers were spending hours in getting the inventory of products in position instead of spending time on customer engagement. The R&D division developed a mobile app that fed real-time information to concerned employees. This mobile app provided a holistic view of sales, replenishment, and other required data that were residing in multiple data sources. They also developed an app for the suppliers to help them gain a better understanding of how their products were moving. Likewise, one of the leading retail chains of entertainment electronics and household appliances in Russia was able to process POS data in real time, which helped in avoiding shortages and excessive stock (Winckelmann 2013). The processing of inventory data of over 9000 items in 370 stores and four distribution centers is complex and time consuming. Their use of the SAP HANA2 solution with in-memory real-time processing and database compression was a significant asset in improving the results.
2.3 Supply Chain Disruption
One of the biggest challenges to supply chain managers is managing disruption. It is important to predict potential disruptions and respond to them quickly to minimize their impact. Supply chain disruption can have a significant impact on corporate performance. At a very high level, firms impacted by supply chain disruptions have 33–40% lower stock returns relative to their benchmark and suffer a 13.5% higher volatility in share prices as compared to a previous year where there was no disruption. Disruptions can have a significant negative impact on profitability—a typical disruption could result in a 107% drop in operating income, 7% lower sales growth, and 11% higher cost (Hendricks and Singhal 2015).
Supply chain disruptions can be caused by either uncontrollable events such as natural disasters or controllable events such as man-made errors. Better information and analytics can help predict and avoid man-made errors. For example, one shipping company that was facing challenges of incomplete network visibility deployed a supply chain technology that gave them a seamless view of the system. The technology enabled managers to get shipping details and take preventive or corrective actions in real time. In this example, prescriptive analytics could have also provided better decision support for the managers to assess and compare various options and actions. The benefits of improved efficiency, more reliable operations, and better customer satisfaction could have aided in the expansion of their customer base and business growth (Hicks 2012).
Connected Cows is a widely reported example of technology being used to aid farmers in better monitoring their livestock (Heikell 2015). The cows are monitored for well-being 24 h a day. This technology not only helps in taking better care of the livestock but also in reducing the disruptions in the production of dairy houses. Connected Cows helps farmers to determine which cows are sick and take timely action to nurse them back to good health with minimal effect on production. A similar concept can be applied to other assets of an organization where creating connected assets can draw valuable insights and provide needed preventive or corrective actions that minimize supply chain disruptions.
All of the aforementioned examples have had considerable historical data that helped in identifying supply chain disruptions and risk assessment. At times, this is not the case, and rare events such as Hurricane Katrina, epidemics, and major outages due to fire accidents may occur. Such events have high impact but low probability without much historical data, and hence the traditional approach cannot be used. The HBR review paper (Simchi-Levi et al. 2014) has addressed this issue by developing a model that assesses the impact of such events rather than their cause. In these extreme cases, the mitigation strategy takes center stage. They visualize the entire supply chain as a network diagram with nodes for the supplier, transportation center, distribution center, etc. where the central feature is the time to recovery (TTR)—“the time it would take for a particular node to be fully restored to functionality after a disruption.” Using linear optimization, the model removes one node at a time to determine the optimal response time, and it generates the performance index (PI) for each node. There are many benefits for this approach; most importantly, managers gain a thorough understanding on the risk exposure of each node. The risk can subsequently be categorized as high, medium, and low, and corresponding prescriptive actions can be initiated. This model also depicts some of the dependencies among the nodes and the bottlenecks. There are certain cases where the total spending is low but the overall impact of disruption is significant—a carmaker’s (Ford) spending on valves is low; however, the supply disruption of these components would cause production line to be shut down. This methodology was used by the Ford Motor Company to assess its exposure to supply chain disruptions.
2.4 Commodity Procurement
The price and supply of commodity can fluctuate significantly over time. Because of this uncertainty, it becomes difficult for many companies that rely on commodity as raw materials to ensure business continuity and offer a constant price to their customers. The organizations that use analytics to identify macroeconomic and internal indicators can do a more effective job in predicting which way prices might go. Hence, they can insulate themselves through inventory investment and purchases of future and long-term contracts. For example, a sugar manufacturer can hedge itself from supply and demand shocks by multiple actions, such as contracting out production on a long-term basis, buying futures on the commodity markets, and forward buying before prices upswing.
Another example is the procurement of ethanol that is used in medicines or drugs. Ethanol can be produced petrochemically or from sugar or corn. Prices of ethanol are a function of its demand and supply in the market, for which there is good degree of volatility. The price of ethanol is also affected by the supply of similar products in the market. As such, there are numerous variables that can impact the price of ethanol. Data analytics can help uncover these relationships to plan the procurement of ethanol. The same analytics tools and models can be extended to other commodity-based raw materials and components (Chandrasekaran 2014).
The last example is the spike in crop price due to changing climate. Climate change is likely to affect food and hunger in the future due to its impact on temperature, precipitation, and CO2 levels on crop yields (Willenbockel 2012). Understanding the impact of climate change on food price volatility in the long run would be useful for countries to take necessary preventive and corrective actions. Computable general equilibrium (CGE) is used by researchers to model the impact of climate change, which has the capability to assess the effects external factors such as climate can have on an economy. The baseline estimation of production, consumption, trade, and prices by region and commodity group takes into account the temperature and precipitation (climate changes), population growth, labor force growth, and total factor productivity growth in agricultural and nonagricultural sectors. The advanced stage simulates various extreme weather conditions and estimates crop productivity and prices subsequently.
The examples provided barely touch upon the many different possible applications in supply chain management. The idea of the survey is to provide guidance regarding main areas of applications. The references at the end of the chapter contain more examples and descriptions of methods. In the next section, we describe in detail an example that illustrates inventory optimization and distribution strategies at a major wireless carrier.
3 VASTA: Wireless Service Carrier—A Case Study
Our case study was set in 2010 where VASTA was one of the largest wireless service carriers in the USA and well known for its reliable national network and superior customer service. In the fiscal year of 2009, VASTA suffered a significant inventory write-off due to the obsolescence of handsets (cell phones). At the time, VASTA carried about $2 billion worth of handset inventory in its US distribution network with a majority held at 2000+ retail stores. To address this challenge, the company was thinking to change its current “push” inventory strategy, in which inventory was primarily held at stores, toward a “pull” strategy, where the handset inventory would be pulled back from the stores to three distribution centers (DCs) and stores would alternatively serve as showrooms. Customers visiting stores would be able to experience the latest technology and place orders, while their phones would be delivered to their homes overnight from the DCs free of charge. The pull strategy had been used in consumer electronics before (e.g., Apple), but it had not been attempted by VASTA and other US wireless carriers as of yet (Zhao 2014a, b).
As of 2010, the US wireless service market had 280 million subscribers with a revenue of $194 billion. With a population of about 300 million in the USA, the growth of the market and revenue were slowing down as the market became increasingly saturated. As a result, the industry was transitioning from the “growth” business model that chased revenue growth to an “efficiency” model that maximized operational efficiency and profitability.
Comprehensive national coverage
Superior service quality and reliable network
High inventory availability and customer satisfaction
Lower inventory turnover and higher operating cost when compared to competitors.
Services and products priced higher than industry averages due to the higher operating costs
The main challenge faced by VASTA was its cost efficiency, especially in inventory costs. VASTA’s inventory turnover was 28.5 per year, which was very low compared to what Verizon and Sprint Nextel achieved (around 50–60 turns per year). Handsets have a short life cycle of about 6 months. A $2 billion inventory investment in its distribution system posed a significant liability and cost for VASTA due to the risk of obsolescence. In the following sections, we will analyze VASTA’s proposition for change using sample data and metrics.
3.1 Problem Statement
To maintain its status as a market leader, VASTA must improve its cost efficiency without sacrificing customer satisfaction. VASTA had been using the “push” strategy, which fully stocked its 2000+ retail stocked to meet customer demand. The stores carried about 60% of the $2 billion inventory, while distribution centers carried about 40%.
VASTA’s old distribution model. Source: Lecture notes, “VASTA Wireless—Push vs. Pull Distribution Strategies,” by Zhao (2014b)
VASTA’s proposed new distribution model. Source: Lecture notes, “VASTA Wireless—Push vs. Pull Distribution Strategies,” by Zhao (2014b)
The “push” and “pull” strategies represent two extreme solutions to a typical business problem in integrated distribution and logistics planning, that is, the strategic positioning of inventory. The key questions are as follows: Where to place inventory in the distribution system? And how does it affect all aspects of the system, from inventory to transportation and fulfillment to customer satisfaction?
Pros and cons of the two distribution models
Pros | Cons | |
|---|---|---|
Push | • Customer satisfaction | • Significant inventory investment |
• Batch picking at DCs | • Risk of obsolescence | |
• Batch, 2-day shipping to stores | ||
Pull | • Significant inventory reduction | • Customers have to wait for delivery |
• Faster switch to new handsets | • Unit picking at DCs | |
• Unit, express overnight shipping to individual customers |
While the push strategy allowed VASTA to better attract customers, the pull strategy had the significant advantage of reducing inventory and facilitating the fast introduction of new handsets, which in turn reduced the cost and risk of inventory obsolescence. However, the pull strategy did require a higher shipping and warehouse fulfillment cost than the push strategy. In addition, VASTA had to renovate stores to showrooms and retrain its store workforce to adapt to the change.
Intuitively, the choice of pull versus push strategies should be product specific. For instance, the pull strategy may be ideal for low-volume (high uncertainty) and expensive products due to its relatively small shipping and fulfillment cost but high inventory cost. Conversely, the push strategy may be ideal for high-volume (low uncertainty) and inexpensive products. However, without a quantitative (supply chain) analysis, we cannot be sure of which strategy to use for the high-volume and expensive products and the low-volume and inexpensive products; nor can we be sure of the resulting financial impact.
3.2 Basic Model and Methodology
Basic model of costs
For product i at store j | Costs (per unit of time) |
|---|---|
Average inventory level | I ij |
Inventory cost | IC ij = h i × I ij |
Weekly sales volume | V ij |
Shipping cost | SC ij = s j × V ij |
Fulfillment cost | FC ij = f(V ij) |
Total cost | IC ij+SC ij+FC ij |
We need to estimate all cost parameters and sales (demand) and inventory level statistics for each product–store combination from the data.
3.3 Cost Parameter Estimates
To calculate the costs, such as store inventory, shipping, and DC fulfillment (e.g., picking and packing) cost for each product–store combination, we need to estimate the inventory holding cost rate, h i; the shipping cost rate, s j; and the fulfillment cost function, f(V ij). We will use a previously collected data set of sales (or demand, equivalently) and inventory data at all layers of the VASTA’s distribution system for 60 weeks. One period will equal 1 week because inventory at both the stores and DCs is reviewed on a weekly basis.
Inventory cost rate:
Inventory holding cost per week = capital cost per week + depreciation cost per week
Capital cost per week = Annual capital cost/Number of weeks in a year
Depreciation cost per week = [Product value − Salvage value]/Product life cycle
Features of phones sold by VASTA
Smartphones (expensive) | Feature (inexpensive) phones |
|---|---|
• Average product value: $500 | • Average product value: $200 |
• Salvage value at store: 0% | • Salvage value at store: 0% |
• Annual capital cost: 7% | • Annual capital cost: 7% |
• Inventory cost/week: $19.90 | • Inventory cost/week: $7.96 |
Shipping costs of phones
Pull | Push | |
|---|---|---|
Shipping method | Overnight express to customers | 2-day batch to stores |
Shipping cost rate s j | $12/unit | $2.88/unit |
DC fulfillment cost: Distribution centers incur different costs for batch picking and packing relative to unit picking and packing due the economies of scale. For VASTA’s DCs, the pick of the first unit of a product costs on average $1.50. If more than one unit of the product is picked at the same time (batch picking), then the cost of picking any additional unit is $0.1. We shall ignore the packing cost as it is negligible relative to the picking cost.
3.4 Analysis, Solution, and Results
High-volume and expensive products, that is, hot-selling smartphones
High-volume and inexpensive products, that is, hot-selling feature phones
Low-volume and expensive products, that is, cold-selling smartphones
Low-volume and inexpensive products, that is, cold-selling feature phones
Representative sales of types of phone
Average | Average | |
|---|---|---|
Weekly sales volume (unit) | On-hand inventory (unit) | |
High volume and expensive (hot-smart) | 99 | 120 |
High volume and inexpensive (hot-feature) | 102 | 110 |
Low volume and expensive (cold-smart) | 2.5 | 15 |
Low volume and inexpensive (cold-feature) | 7.3 | 25 |
Savings for “hot-smart” phones between pull and push strategies
High volume and expensive (hot-smart) | Pull | Push |
|---|---|---|
Inventory level | 5 (I ij) | 120 (I ij) |
Inventory cost | $99.52 (I ij × h ij = 5 × $19.90) | $2388.46 (I ij × h ij = 120 × $19.90) |
Weekly sales volume | 99 (V ij) | 99 (V ij) |
Shipping cost | $1188 (V ij × s ij = 99 × $12/unit) | $285.12 (V ij × s ij = 99 × $2.88/unit) |
Fulfillment cost | $148.50 | $11.30 |
(V ij × $1.50 = 99 × $1.50) | ($1.50 + (V ij | |
− 1)*0.1 = $1.50 + 98*0.1) | ||
Total cost | $1436.02 | $2684.88 |
Savings | – | 46.51% |
The calculation shows that we can save 46.51% of the total landed cost for this high-volume and expensive product if we replace the push strategy by the pull strategy. This is true because the savings on inventory cost far exceeds the additional cost incurred for shipping and DC fulfillment.
Savings for “hot-feature” phones between pull and push strategies
High volume and inexpensive (hot-feature) | Pull | Push |
|---|---|---|
Inventory level | 5 (I ij) | 110 (I ij) |
Inventory cost | $39.81 (I ij × h ij = 5 × $7.96) | $875.77 (I ij × h ij = 110 × $7.96) |
Weekly sales volume | 102 (V ij) | 102 (V ij) |
Shipping cost | $1224 (V ij × s ij = 102 × $12/unit) | $293.76 (V ij × s ij = 102 × $2.88/unit) |
Fulfillment cost | $153.00 (V ij × $1.50 = 102 × $1.50) | $11.60 ($1.50 + (V ij − 1)*0.1 |
= $1.50 + 101*0.1) | ||
Total cost | $1416.81 | $1181.13 |
Savings | – | −19.95% |
Savings for “cold-smart” phones between pull and push strategies
Low volume and expensive (cold-smart) | Pull | Push |
|---|---|---|
Inventory level | 2 (I ij) | 15 (I ij) |
Inventory cost | $38.81 (I ij × h ij = 2 × $19.90) | $298.56 (I ij × h ij = 15 × $19.90) |
Weekly sales volume | 2.5 (V ij) | 2.5 (V ij) |
Shipping cost | $30.00 (V ij × s ij = 2.5 × $12/unit) | $7.20 (V ij × s ij = 2.5 × $2.88/unit) |
Fulfillment cost | $3.75 (V ij × $1.50 = 1.5 × $1.50) | $1.65 ($1.50 + (V ij − 1)*0.1 |
= $1.50 + 1.5*0.1) | ||
Total cost | $73.56 | $307.41 |
Savings | – | 76.07% |
Savings for “cold-feature” phones between pull and push strategies
Low volume and inexpensive (cold-feature) | Pull | Push |
|---|---|---|
Inventory level | 2 (I ij) | 25 (I ij) |
Inventory cost | $15.92 (I ij × h ij = 2 × $7.96) | $199.04 (I ij × h ij = 120 × $7.96) |
Weekly sales volume | 7.3 (V ij) | 7.3 (V ij) |
Shipping cost | $87.60 (V ij × s ij = 7.3 × $12/unit) | $21.02 (V ij × s ij = 7.3 × $2.88/unit) |
Fulfillment cost | $10.95 (V ij × $1.50 = 7.3 × $1.50) | $2.13 ($1.50 + (V ij − 1)*0.1 |
= $1.50 + 6.3*0.1) | ||
Total cost | $114.47 | $222.19 |
Savings | – | 48.48% |
Savings for all types of phones between pull and push strategies
Cold-smart | Cold-feature | Hot-smart | Hot-feature | |
|---|---|---|---|---|
% Savings | 76.07% | 48.48% | 46.51% | –19.95% |
Consistent to our intuition, the pull strategy brings the highest savings for the low-volume and expensive product (cold-smart), and the lowest savings (even a loss) for the high-volume and inexpensive product (hot-feature). In general, the pull strategy tends to bring less savings for products with a higher volume and/or a less cost.
Store inventory investment for pull and push strategies
Pull | Push | ||||
|---|---|---|---|---|---|
Category | # of products | Inventory level | Inventory investment | Inventory level | Inventory investment |
Hot-smart | 22 | 5 | 5 ⋅ $500 ⋅ 22 | 120 | 120 ⋅ $500 ⋅ 22 |
= $55,000 | = $1,320,000 | ||||
Hot-feature | 20 | 5 | 5 ⋅ $200 ⋅ 20 | 110 | 110 ⋅ $200 ⋅ 20 |
= $20,000 | = $440,000 | ||||
Cold-smart | 5 ⋅ $200 ⋅ 20 | 110 ⋅ $200 ⋅ 20 | |||
15 | 2 | = $15,000 | 15 | = $112,500 | |
Cold-feature | 11 | 2 | 2 ⋅ $200 ⋅ 11 | 25 | 25 ⋅ $200 ⋅ 11 |
= $4,400 | = $55,000 | ||||
Total | 68 | – | $94,400 | – | $1,927,500 |
From this table, we can see that inventory investment per store under the pull strategy is only about 5% of that under the push strategy. Thus, the pull strategy can reduce the store-level inventory by about 95%. Given that store inventory accounts for 60% of the $2 billion total inventory investment, the pull strategy will bring a reduction of at least $1 billion in inventory investment as compared to the push strategy.
Total costs for pull and push strategies
Per store per week | Pull | Push |
|---|---|---|
Total inventory cost | $3757.85 | $76,729.33 |
Total shipping cost | $52,029.60 | $12,487.10 |
Total picking cost | $6,503.70 | $528.78 |
Total cost | $62,291.15 | $89,745.21 |
The table shows that the inventory cost reduction outweighs the shipping/picking cost inflation and thus the pull strategy results in a net savings per store of about 31% relative to the push strategy.
3.5 Advanced Model and Solution
As shown by our prior analysis, the pull strategy does not outperform the push strategy for all products. In fact, for high-volume and inexpensive products (hot-feature phones), it is better to satisfy a portion of demand at stores. Thus, the ideal strategy may be hybrid, that is, the store should carry some inventory so that a fraction of demand will be met in-store, while the rest will be met by overnight express shipping from a DC. The question is how to set the store inventory level to achieve the optimal balance between push and pull.
T: the review period
D(T): the demand during the review period:
E[D(t)] = μ: the mean of the demand during the review period
STDEV[D(t)] = σ: the standard deviation of the demand during the review period
![$$ E\left[{D}_1\right]=E\left[\max \left\{0,D(T)-S\right\}\right]=\sigma \left[\phi \left({z}_{\alpha}\right)-{z}_{\alpha}\left(1-\alpha \right)\right], $$](../images/428412_1_En_24_Chapter/428412_1_En_24_Chapter_TeX_Equ8.png)
is the standard normal probability density function.![$$ E\left[{D}_2\right]=E\left[\min \left\{D(T),S\right\}\right]=E\left[D(T)\right]-E\left[{D}_1\right]. $$](../images/428412_1_En_24_Chapter/428412_1_En_24_Chapter_TeX_Equ9.png)
![$$ EI=E\left[\max \left\{0,S-D(T)\right\}\right]=S-\mu +E\left[{D}_1\right]. $$](../images/428412_1_En_24_Chapter/428412_1_En_24_Chapter_TeX_Equ10.png)
- 1.
Batch and 2-day shipping of the quantity E[D 2] from DCs to the store
- 2.
Express overnight shipping of the quantity E[D 1] from DCs to customers
- 1.
Batch picking at DCs for the quantity of E[D 2]
- 2.
Individual picking at DCs for the quantity of E[D 1]
Estimates of representative phones at a representative store
Weekly sales average | Weekly sales standard deviation | On-hand inventory | |
|---|---|---|---|
High volume and expensive (hot-smart) | 99 | 40 | 120 |
High volume and inexpensive (hot-feature) | 102 | 53 | 110 |
Low volume and expensive (cold-smart) | 2.5 | 2.3 | 15 |
Low volume and inexpensive (cold-feature) | 7.3 | 9.1 | 25 |
Comparison between push and pull strategies for hot phones
Comparison between push and pull strategies for cold phones
The pull strategy is best for both the hot- and cold-smartphones.
For feature phones, it is best to use a hybrid strategy (refer to Table 24.14).
Savings from moving to pull strategy
Hot-feature | Cold-feature | |
|---|---|---|
The best Type 1 service (α) | 50% | 50% |
% demand met by store | 79% | 50% |
Saving from pull | 28% | 8.4% |
From the store perspective, the savings gained from switching from pull to hybrid is 13.4% (or $8081.90) per store per week.
3.6 Customer Satisfaction and Implementation
So far, our analysis focuses on the total landed cost, which is smaller under the pull strategy than the push strategy. Despite this cost efficiency, a fundamental issue remains: Will customers accept the pull strategy? More specifically, will customers be willing to wait for their favorite cell phones to be delivered to their doorstep overnight from a DC?
- 1.
Overnight free of charge
- 2.
Overnight with a fee of $12
- 3.
Free of charge but 2 days
- 4.
Free of charge but store pickup
Different options have significantly different costs and customer satisfaction implications; they must be tested in different market segments and geographic regions. To ensure customer satisfaction, VASTA had decided to start with option 1 for all customers.
- 1.
Converting retail stores to showrooms and retraining sales workforce
- 2.
Negotiating with carriers on the rate and service of the express shipping
- 3.
A massive transformation of the DCs that will transition from handling about 33% individual customer orders to nearly 72% individual customer orders (the indirect sales, through third-party retail stores such as Walmart, can be fulfilled by batch picking and account for 28% of total sales)
Despite the renovation costs and training expenses, showrooms may enjoy multiple advantages over stores from a sales perspective. For instance, removing inventory can save space for product display and thus enhance customers’ shopping experiences. Showrooms can increase the breadth of the product assortment and facilitate faster adoption to newer handsets and thus increase sales. Finally, they can also help to reduce store-level inventory damage and theft, thereby minimizing reverse logistics.
Implementation plan of pull strategy
Phase I | • Implement the pull strategy for one DC and some target stores • Negotiate shipping contracts with carriers • Review savings, service levels, and impact on customers |
Phase II | • Implement the pull strategy to all stores served by the DC • Experiment the options of store pickup and 2-day free home shipping |
Phase III | • Full-scale implementation of the pull strategy to all three DCs • Review savings, service levels, and impact on customers |
3.7 Epilogue
In 2011, VASTA implemented the pull strategy in its US distribution system. FedEx overnight was used. System inventory reduced from $2 billion to $1 billion. Soon after, other US wireless carriers followed suit, and the customer shopping experience of cell phones completely changed in the USA from buying in stores to ordering in stores and receiving delivery at home. In the years after, VASTA continued to fine-tune the pull strategy into the hybrid strategy and explored multiple options of express shipping depending on customers’ preferences. VASTA remains as one of the market leaders today.
4 Summary: Business Insights and Impact
In this chapter, we showcase the power of supply chain analytics in integrated distribution and logistics planning via a business case in the US wireless services industry. The company, VASTA, suffered a significant cost inefficiency despite its superior customer service. We provide models, methodology, and decision support for VASTA to transform its distribution strategy from “push” to “pull” and eventually to “hybrid” in order to improve its cost efficiency without sacrificing customer satisfaction. The transformation resulted in $1 billion savings in inventory investment and helped the company to maintain its leadership role in an increasingly saturated marketplace.
The conflicting goals of cost efficiency and customer satisfaction are hard to sort out qualitatively. Quantitative supply chain analysis is necessary to strike the balance.
System thinking: Distribution strategies can have a significant impact on all aspects of a system: inventory, shipping, customer satisfaction, as well as in-store and warehouse operations. We must evaluate all aspects of the system and assess the net impact.
One size does not fit all: We should customize the strategies to fit the specific needs of different products and outlets (e.g., stores).
All the datasets, code, and other material referred in this section are available in www.allaboutanalytics.net.
Data 24.1: Vasta_data.xls
Ex. 24.1 Reproduce the basic model and analysis on the representative store for the comparison between push and pull strategies.
Ex. 24.2 Reproduce the advanced model and analysis on the hybrid strategy for the representative store.
Ex. 24.3 For NYC and LA stores, use the basic and advanced models to find out which strategy to use for each type of product, and calculate the cost impact relative to the push strategy.