Bachelor Thesis on the topic of: How can Predictive Shipping be successfully implemented for Online Retail? Simulating the profitability of Predictive Shipping on an exemplary distribution network 1 Table of contents Introduction ..................................................................................................................................... 3 Theoretical Background .................................................................................................................. 4 Overview ............................................................................................................................. 4 Opportunities and Challenges .............................................................................................. 8 Opportunities ........................................................................................................... 8 Challenges ............................................................................................................. 16 Implementation methods and standings ............................................................................ 21 Examples ............................................................................................................... 21 Current Progress .................................................................................................... 22 Research Method ............................................................................................................... 24 Problem Description ...................................................................................................................... 25 Predictive Shipping Simulation ..................................................................................................... 27 Model ................................................................................................................................. 27 Prediction Algorithm ............................................................................................. 27 Order Simulation System....................................................................................... 27 Monte Carlo Simulation ........................................................................................ 28 Results and managerial implications ................................................................................. 31 Conclusion and future outlook ...................................................................................................... 33 Declaration of originality/Statutory declaration ............................................................................ 33 Reference list .................................................................................. Error! Bookmark not defined. Table of figures .............................................................................................................................. 38 2 Introduction The modern Supply Chain is undergoing enormous transformational changes (Alicke et al., 2016, p. 1). This can be seen both in terms of global trends such as the “[…] maturation of globalization, the exponential rise of e-commerce, the constant threat of technology disruption, and, most recently, the coronavirus pandemic […]” (DHL Customer Solutions, 2020) as well as new technological advancements such as “cloud computing, collaborative robotics, big data analytics, artificial intelligence, and the Internet of Things […] (DHL Customer Solutions, 2020) Increased costs, individual preferences and supply chain complexity are the top external trends putting pressure on Supply Chains (BVL, 2017, p. 1) Furthermore, today customers demand faster deliveries (Ericsson Industrylab, 2020, p. 3) and increased traceability of their orders to have a detailed understanding of when the ordered product arrives. As Ericsson describes: “These expectations are also evident in the B2B sector; of the companies surveyed, 88 percent say that their customers currently request same-day delivery”. (Ericsson Industrylab, 2020, p. 3). Faster deliveries are in many parts due to the revolutionary way Amazon has changed customer expectations: “[…] now everyone wants everything within two days, or the next day, if they’re in the US. Amazon came in and changed the game for everyone.” (Ericsson Industrylab, 2020, p. 3) It thus becomes evident that the speed of your logistics is an essential competitive advantage to attain and retain more customers (Zhalgassova, 2014, p. 1). Customers expect fast delivery after order placement. Thus delivery performance is critical to enhancing the customer shopping experience in omnichannel commerce. (Lee, 2017, p. 593). To put it simply, whoever can deliver first, will be preferred by the customer (Zhalgassova, 2014, p. 1) As E-Commerce sales are increasing worldwide (Fig. 1), there is now growing pressure to deliver more products faster, and solutions have to be found. Figure 1: Retail e-commerce sales worldwide from 2014 to 2023 (Simmons, 2020) One of the most innovative and critical technological advancements in supply chain management is a mixture of Big Data Analysis and Artificial Intelligence in Predictive Analytics. My research 3 focuses on how a subset of Predictive Analytics, Predictive Shipping, which focuses on sending a product near you before you ordered it, could be implemented in real life to combat the Issues above. During my research, I have observed a lack of research concerning the actual implementation of Predictive Shipping. Therefore, this thesis aims to show the current research on Predictive Shipping and then create a model to simulate Predictive Shipping in the real world. The goal is to understand whether Predictive Shipping can improve the delivery process and maintain profitability. I will first explain what Predictive Shipping entails and the opportunities and challenges that impact Predictive Shipping. Then, I will show how Predictive Shipping is currently being used and what progress has been made. Next, I describe my research approach and the exact problem in Predictive Shipping that I attempt to solve. Afterward, I will demonstrate my model to simulate Predictive Shipping and analyze the results. Finally, I will give a conclusion and future outlook for both technology and research. Theoretical Background Overview As mentioned previously, Predictive Shipping is a subset of Predictive Analytics, which is defined as follows: “Predictive Analytics is a set of business intelligence (BI) technologies that uncover relationships and patterns within large volumes of data that can be used to predict behavior and events. Unlike other BI technologies, Predictive Analytics is forward-looking, using past events to anticipate the future.” (Eckerson, 2007, p. 5) Instead of asking “What happened?” like descriptive analytics or “Why did it happen?” like diagnostic analytics, Predictive Analytics try to predict “What will happen?” in the future (compare Fig. 2). Predictive Analytics is a substantial field used in many other different industries, especially healthcare and finance. Predictive Analytics and the application of artificial intelligence hold enormous potential for optimizing logistics processes. (Kersten et al., 2017, p. 14) Figure 2: Predictive Analytics in comparison (AIMultiple, 2019) 4 There are many subsets of Predictive Analytics in supply chain management, such as Predictive Maintenance, Predictive Picking&Packing or Predictive Shipping. The “Bundesvereinigung Logistik” describes Predictive Analytics as a highly relevant current technology already being implemented. (compare Fig. 3) Figure 3: A graph depicting the current standings and relevance of modern technology trends. (BVL, 2017, p. 4) One key aspect of all these subsets of Predictive Analytics is that they allow for a completely different approach from traditional analytics. As a blog from SupplyChain24/7 puts it: “Without predictive analytics, you’re planning under the assumption that the past will repeat itself, instead of building a data vision and creating a model for the future.” (SupplyChain24/7, 2021) Thus Predictive Shipping uses current data, allowing for better planning of, for example, Vehicle Capacity or Inventory Management in the future. 5 It is thus not surprising that a recent survey of 621 logistic providers found Predictive Analytics to be the 3rd most invested technology, with only a small margin of a few % to the most popular technology cloud logistics (compare Fig. 4). Predictive Analytics and their associated subsets have future potential and are thus very relevant in the modern landscape. Figure 4: Most essential technology investments (Simmons, 2020) Predictive Shipping, also known as Anticipatory Shipping, originates from the patent US8615473B2, which Amazon filed in 2012 and then approved in 2013. Their most common definition in the patent states: “According to one embodiment, a method may include packaging one or more items as a package for eventual shipment to a delivery address, selecting a destination geographical area to which to ship the package, shipping the package to the destination geographical area without completely specifying the delivery address at the time of shipment, and while the package is in transit, completely specifying the delivery address for the package.” (Spiegel et al., 2013) The way it works is quite simple in theory. The description in the patent (visible in Fig. 5) shows how products get sent to a hub close to you before an order is made. If the order is made during this process, the vehicle is rerouted dynamically to your address. Besides dramatically lowering delivery times, Predictive Shipping could reduce shipping costs, as heavily demanded items can be sent via standard shipping beforehand over expedited shipping (Kundu and Tripti, 2021, p. 1). As Last-Mile Delivery is still the most expensive part of supply chains, Predictive Shipping could thus be a very profitable innovation (DHL Customer Solutions, 2020, p. 51). 6 Figure 5: How Predictive Shipping works (Spiegel et al., 2013) There is another application of Predictive Shipping that could be applied in more fringe conditions and has unique use applications. While the one I just described focuses on identifying significant trends from personal preferences and then shipping the goods from around the world as close to you as possible, the other use case involves sending it directly to you. The full text describing this in the patent states: “After determining the potential cost of returning or redirecting the near proximity package, the package is provided to the potentially interested customer at a discounted price, where the discounted price depends on the determined potential cost of return or redirection. For example, rather than incur the possible cost of returning or redirecting a package without a sale, some or all of the potential cost may be offered as a 7 discount to a potentially interested customer, such as via an e-commerce portal, as an inducement to convert the potential interest into an order. In some instances, the package may be delivered to a potentially interested customer as a gift rather than incurring the cost of returning or redirecting the package. For example, if a given customer is particularly valued (e.g., according to past ordering history, appealing demographic profile, etc.), delivering the package to the given customer as a promotional gift may be used to build goodwill. (Spiegel et al., 2013) For this thesis’s sake, it is vital to differentiate between the Predictive Shipping used to deliver goods to hubs closer to you and having goods delivered directly. In the following passages, I primarily focus on the latter; however, statistics describing Predictive shipping will refer to the former, which is far easier to implement in the real world, with much less controversy. The current response to this very innovative technology was mixed. Many believe in its great potential and state: “If this gets good customer response, Amazon will achieve a competitive advantage in shipping time. Few other online merchants have the resources to match Amazon’s logistics capabilities” (Kaplan, 2014). A blog by SmartData Collective goes as far as to say that: “If it does work, it will significantly alter the landscape and encourage many other retailers and e- commerce brands to develop something similar.” (Nichols, 2018) However, there are many critical voices due to the many challenges facing Predictive Shipping. Can Predictive Shipping be implemented in the real world? To further understand Predictive Shipping and structure my findings, I analyze the potential opportunities and challenges. Opportunities and Challenges Opportunities Speed of Delivery The main benefit of a successful implementation of Predictive Shipping is obvious. First, this technology could reduce transportation time and thus also transportation costs. Second, customer satisfaction could increase immensely by having their desires met faster. Quick delivery speeds are critical as, according to Fig. 6, only 1% of all cross-border bought online products had same-day shipping, and only 3% had next-day shipping. Yet, there is a need for faster deliveries in the population. For example, 65% of 314 logistics companies mentioned needing to increase delivery options, and 49% focused on higher speeds for their last-mile deliveries. (compare Fig. 7) If Predictive Shipping was used globally, trends could be deduced early and products shipped to places of high demand way before demand occurred, drastically reducing transportation times. If one imagines the COVID-19 pandemic, for example, if a need for masks had been identified beforehand, then shipments from around the world, especially China, could have arrived far earlier in Europe and saved many lives. Similarly, newly emerging trends such as the Fidget Spinner in 2017/2018 could have been identified beforehand. As those goods can rise and drop extremely quickly in demand, quick transportation times and early deliveries across borders are vital to match demand and manage stock. (Gesing et al., 2018, p. 25) Transmetrics describes this process as follows: “By using 8 predictive solutions to generate supply and demand forecasts, companies will be able to make the right operational decisions proactively. This approach can also allow for the rebalancing of assets across any logistic network at a minimal cost “(Transmetrics, 2019). Faster, more accurate deliveries with fuller trucks could also reduce carbon emissions and increase the sustainability in the last-mile delivery. Figure 6: Cross border delivery date (Simmons, 2020) Figure 7: Most crucial last-mile delivery areas (Simmons, 2020) 9 Competition with Brick-and-Mortar Stores E-Retailers can use Predictive Shipping to compete with different physical retailers, such as brick-and-mortar stores. As the GTG Technology Group puts it on the example of Amazon: “Amazon’s objective is to more effectively compete with brick and mortar retailers that can better satisfy customer needs for instant gratification (GTG Technology Group, 2015, p. 10)”. In addition, these stores can have close interactions with customers and purchase products within minutes (Kaplan, 2014). Therefore, a faster delivery performance is critical in enhancing the customer shopping experience (Lee, 2017, p. 593) and increasing customer retention. The high amount of returns There is currently an issue with a large number of returns and destroyed items for E-Retailers, which Predictive Shipping might help. A recent report shows: “Online giant Amazon is destroying millions of items of unsold stock every year, products that are often new and unused, ITV News can reveal.” (Pallot, 2021, p. 1) The destruction of perfectly useable items leaves much room for more potential. Of the 130.000 items destroyed per week, 50% are still unopened, while the other 50% are returned items in good conditions (Pallot, 2021, p. 2). It is essential to mention that this was not a case exclusive to Amazon. Purdy shows: ”In France, a TV station found more than 3 million products destroyed in 2018. In Germany, one warehouse sends out a truckload every week for destruction. And those are just the stories that rose to the surface”. (Purdy, 2021, p. 3) The reasons for these actions are simple: “It is eventually cheaper to dispose of the goods, especially stock from overseas, than to continue storing the inventory.” (Pallot, 2021, p. 2). Labarre further explains: “Most objects in financial accounts are classified as stock, a type of asset. Stocks depreciate over time, until ultimately – when the cost of storage outweighs potential return – that unsold stock becomes a liability. Organizations can choose how to dispose of their liabilities but, in balance-book terms, the stock has become waste: worthless objects to be discarded (in the cheapest possible way)”. (Wishart, 2021, p. 2) As Pallot states, these actions have been heavily criticized: “Products that were never sold or returned by a customer. Almost all could have been redistributed to charities or those in need. Instead, they are thrown into vast bins, carried away by lorries (which we tracked), and dumped at either recycling centers or, worse, a landfill site. (Pallot, 2021, p. 2) (see Fig. 8) Figure 8: Unopened Products about to be destroyed (Merchant Fraud Journal, 2021) 10 Returns have been an increasingly significant Issue for E-Retailers, as numbers keep increasing. A report by Amazon showed: “In today’s retail climate, returns account for 50 percent of items purchased in the US alone — and amount to a cost of $350 billion a year, with 5 billion pounds of clothing and textiles winding up annually in US landfills.” (Amazon Staff, 2021, p. 3) The report concludes with the statement: “Customer returns are a fact of life for all retailers, and what to do with those products is an industry-wide challenge,” said Libby Johnson McKee, director of Amazon WW Returns, ReCommerce, and Sustainability. (Amazon Staff, 2021, p. 2) This increase in returns is in large part due to the: “Increased e-Commerce adoption, driven by COVID shutdowns, leads to higher overall return rates because digital purchases are more likely to be returned store purchases” (Blair, 2020, pp. 1–2) However one cannot simply remove returns to save money as “return policies and experiences have a significant impact on customer loyalty. In a survey of consumers conducted by Doddle, 84% said a positive returns experience encourages them to shop with a retailer again. Conversely, 73% of consumers responding to a survey by Returnly said they would not shop with a brand again after a poor returns experience.” (Blair, 2020, p. 2) Return policies are also mandated by law: “Many countries require online retailers to give their customers the right to return products within a certain period of time after purchase. For instance, the minimum required by law in all member states of the European Union is 14 days.” (Urbank P et al., 2015, p. 2) Although a well-managed return system can improve sales and the customer experience, nowadays, the problem gets exaggerated by an expectancy of free returns. The Wharton University found that: “Consumers are coming to expect they can return things and that it will be easy. They’re coming to expect that returns will be free, that when something is delivered to them, there will be a return label or return envelope for sending things back. There is this major upward trend in terms of consumers returning and expecting to return.” (Wharton University of Pennsylvania, 2020, p. 1). 67% of online customers expect free returns. If they had to pay for returns, they would decrease their subsequent spending with the same retailer by 75 -100% by the end of 2 years, whereas customers who had free returns increased their spending by 158 - 457%. (Jack et al., 2019, p. 3) Most companies are not equipped to deal with a large number of returns. As a result, transparency is low, and returns are not being managed correctly. For example, a report by the ECR shows: “Of the 10 retail clients understudy, all were found to be underestimating their return rate – one by as much as 150 percent, with an average return rate discrepancy of over 80 percent …resulting in additional costs of over $462 million to a retailer doing $10 billion in annual revenue.” (Jack et al., 2019, p. 3). A bit further in the report, they continue: “a surprising 39% of respondents said they have no visibility into returns—they just show up. A similar number (38%) receive scheduled reports, and 32% track point of sale information at their returns center. More surprising, 40% of respondents couldn’t determine how much reverse logistics saved their company, and another 36% aren’t sure. Finally, only 24% said they were able to determine how much their reverse logistics operation is saving their company, with an estimated average annual revenue savings of 16.5%.” (Jack et al., 2019, p. 7) Finally, the last issue with returns concerns fraudulent returns. They increase unsold merchandise, create losses, and affect many products in Predictive Shipping. Customers and companies have vastly different expectations for what a legitimate return contains (Wharton 11 University of Pennsylvania, 2020, p. 3). People use “Faulty” or “Change of mind” return categories to give back a TV after a tournament or Lawnmowers at the end of summer, effectively “renting” them. (Jack et al., 2019, p. 11) The result is fraudulent returns: “For eCommerce companies, an astonishing 14% of returns are fraudulent. This shows that while generous returns policies are important for customer satisfaction, it is important for company profitability to make returns barriers high and be tough in enforcing them” (Jack et al., 2019, p. 4) Fraudulent returns are an increasingly significant Issue. Skapa reported in 2013 that: “According to the survey carried out among 111 retailers in the US, the estimated loss caused by return fraud increased from $9.59 billion to $13.95 billion.” (Škapa, 2013, p. 380) Today this figure has risen even higher: “Consumers returned an estimated $428 billion in merchandise to retailers last year, approximately 10.6 percent of total U.S. retail sales in 2020. Of those returns, roughly 5.9 percent were fraudulent, equating to $25.3 billion, according to a report released today by the National Retail Federation and Appriss Retail.” (NRF, 2021) (compare Fig. 9) Figure 9: Merchandise Return Fraud (Merchant Fraud Journal, 2021, p. 2) This issue, in particular, is very complicated and severe with no one-size-fits-all solutions. For the sake of this thesis, I would propose that simply more accurate predictions would have to be made, which also include the risk of fraudulent returns. Besides including previous purchasing behavior and demographics, a proposal by Urbank et al. seems the best to me. They propose: “a system of prediction and targeted intervention that can identify consumption patterns associated with an extremely high rate of product returns and prevent such transactions from taking place. Such consumption patterns might be related to fraud or impulse shopping. “(Urbank P et al., 2015, p. 2) However, in the end, multiple different solutions and policies will have to be used to minimize the problem. The question is, how does one tackle all these significant challenges and turn them into an Opportunity? One solution might be a new program proposed by Amazon themselves: “Amazon has launched two new Fulfillment by Amazon (FBA) programs designed to make it easier for its third-party sellers to resell customer-returned items or overstock inventory while also giving more products a second life.” (Amazon Staff, 2021, p. 1) This reselling gives one hope of a circular economy, where products get reused instead of destroyed. “These new programs are 12 examples of the steps we’re taking to ensure that products sold on Amazon — whether by our small business partners or us — go to good use and don’t become waste. Along with existing programs like FBA Donations, we hope these help build a circular economy, maximize reuse, and reduce our impact on the planet.” (Amazon Staff, 2021, p. 2). Amazon has high hopes for this program as: “once fully rolled out, the company expects these programs to give more than 300 million products a second life each year.” (Amazon Staff, 2021, p. 1) Although this program seems very promising, I believe that Predictive Shipping could also significantly impact decreasing destroyed inventory and help manage the increase in returned goods. The idea is that Predictive Shipping could induce demand and sell products at heavy discounts before the product gets thrown away. Thus, the high number of returns becomes an opportunity to implement Predictive Shipping on less wanted goods. Amazon states in the Predictive Shipping patent that products shipped through Predictive Shipping could be offered at a significant discount or as gifts to build goodwill. This could dramatically reduce the number of destroyed items and returns. Instead of being destroyed, these items now have another chance of a resell. I strongly support Predictive Shipping in combination with returned items because predictively shipping an utterly new item might result in more losses, as it might turn out to be a “lost sale” which was given at a higher discount for no reason. This issue does not exist with items that technically no longer have a value associated with them. Another report strengthened the possibility of gifting being used in Predictive Shipping to reduce costs and improve returns. “For example, shipping products back in some instances costs more than a retailer could recoup through vendor credits, liquidation, or resale. Suppose customer service representatives have that cost data on hand and can combine it with the purchasing and returns history of the customer. In that case, they could be allowed to tell the customer to keep the product, not send it back.” (Blair, 2020, p. 4) Predictive Shipping could be used to create demand and thus reduce inventories of unsellable items, effectively becoming a part of the reverse logistics ecosystem. “(...) pioneering companies (…) identify (…) reverse logistics as more than what amounts to disaster remediation, and see it as a way to improve cost savings, customer satisfaction, profitability, environmental viability, and ultimately competitive advantage and profits.” (Jack et al., 2019, p. 18) Marketing function&Internet of things Adopting Predictive Shipping might mean getting the early adopter benefit. Predictive Shipping is a way to differentiate yourself from the market and offer same-day shipping in a world where free two-day shipping is becoming more and more of an industry standard. Predictive Shipping could be seen as an innovative, revolutionary method that further incentivizes people to choose Amazon over competitors, even if it turns out to be a loss leader (GTG Technology Group, 2015, p. 13). One of Amazon’s strengths has always been its focus on customer experience and shopping convenience. Due to the difficulty of using Predictive Shipping on a large scale, Amazon could be a market leader for a decent time. Amazon would not only be able to implement one or two-day shipping on a more significant number of products but even reduce those times to within hours or before a purchase was made. 13 Besides the dominant market position, there are multiple marketing-related benefits associated with Predictive Shipping. A blog on Medium explains: “Predictive Analytics can help e- commerce businesses drive targeted promotions to their customers by closely analyzing campaigns that have worked well in the past. Advertisements can be offered to relevant customer segments in real-time to encourage customers to complete a purchase or retrieve an abandoned shopping cart.” (IQLECT, 2018, p. 3) To make those recommendations more subtle, one can mix accurate suggestions with other tangentially related items (Hill, 2012, p. 4) A multitude of methods to bring recommendations closer to your home already exists. In the example of Amazon, it would be Amazon’s Alexa platform, which allows you to order products automatically or on a schedule. (Nichols, 2018) However, Alexa could be used to gain even more data. As the Harvard Business Review (HBR) describes: “One could also use Alexa to order your parent a bouquet when you remember that her birthday is next week.” (HBR, 2016, p. 2). Thus, the HBR concludes: “intelligent assistants and connected devices will learn from user habits and pick up on behavioral and environmental patterns to make these experiences more predictive. Devices like the Echo will access data from everyday interactions to predict specific opportunities for a transaction.” (HBR, 2016, p. 2). The internet of things could be a driving force to increase prediction certainty. Thus predictions for common household goods can be made with high certainty (PYMNTS, 2019, p. 2). Other ways to achieve more accurate predictions include using Predictive Shipping on trendy goods, where the chance of a misprediction is lower (smartphones), and very time- sensitive items, where the immediate shipping might be more appreciated (best seller books) (Zhalgassova, 2014, p. 1) The Adoption of Predictive Shipping may be more straightforward if one implements it for people who are already devoted customers and incorporated in the ecosphere and environment of the company. Kaplan described this by bringing up the example of Amazon Prime: “Amazon might choose to limit Predictive Shipping to its Prime customers who pay an annual fee for free shipping. The company could use this as an added benefit for Prime subscribers, especially if Amazon raises Prime annual subscription fees as it stated it might.” (Kaplan, 2014). Weingarten mentioned in their description of Predictive Shipping that: ” Model application could also be limited to customers with high order frequency to reduce the cost of erroneously sending products, as results indicate that those customers are easier to predict.” (Weingarten, 2020, pp. 8– 9) Furthermore, gifting products could be used as a marketing tool to retain customer loyalty. “[…] a provider could improve customer loyalty and push its products into the market. Attaching this service to selected products would incentive customers to purchase critical stock items such as over-stock, perishable or phase-out goods” (CAMELOT Blog, 2019, p. 1). Instead of having items be destroyed, the inventory could be decreased ahead of time through Predictive Shipping if sales targets weren’t met. Amazon has tried gift programs in the past as trial runs in the form of Amazon samples (Fuchs, 2019, p. 1). It is unknown whether this was a trial for Predictive Shipping systems or whether it was evaluated as unprofitable. It was very similar to how Predictive Shipping would work for the customer in real life and was temporarily used as a marketing tool for advertisement and to generate customer goodwill. Amazon samples could 14 have been used as an automated tool for soon-perished items. Extensive marketing campaigns using this method could have sold items directly before they became unfit. (Fuchs, 2019, p. 3) Overall Predictive Shipping might mean a considerable increase in comfortability. Kundu and Dhote conclude: “This strategy, if implemented well, would increase the base of loyal customers, as they are in this process outsourcing their purchases and making themselves free of the purchasing hassles and it is a win-win situation for the Customers and the E-Commerce companies.” (Kundu and Tripti, 2021, p. 8) Implementation of AI To implement Predictive Shipping in the real world, there are many prerequisites. One of them is the know-how and application of Big Data Analysis (BDA) and Artificial intelligence (AI). These are both trends that are very well suited to the Retail and Transportation Industry, as there are is usually a vast amount of high-quality information readily available. (Gesing et al., 2018, p. 14). The DHL reports: ”Predictive logistics remains the most critical AI application for industry professionals, given the abundance of supply chain data, as well as better machine- learning algorithms from which to draw predictive insights.” (DHL Customer Solutions, 2020, p. 51) What exactly is meant under BDA and AI? According to Weingarten: “BDA involves the application of advanced analytics techniques, such as statistics, simulation or optimization, to gain insight from big data to enhance decision-making and increase business value and firm performance” (Weingarten, 2020, p. 1). AI has many definitions and is an incredibly vast field with many different applications, however for the sake of this thesis, it will simply be a tool to improve the efficiency of BDA. As both these trends are on a steady rise, it is to be expected that the importance of Predictive Shipping increases soon. The benefits of BDA have already been shown: “businesses that already use BDA report a 5% increase in productivity and 6% increase in profitability, compared to those that do not” (Weingarten, 2020, p. 1), and “applying AI to supply chains will result in $1.4 trillion of value in the next 20 years.” (DHL Customer Solutions, 2020, p. 51) The GlobalTradeMagazine reports: “Over 90 percent of 3PLs and shippers believe data-driven decision-making is essential to supply-chain activities” (GlobalTradeMagazine, 2016, p. 1). A study by the Council of Supply Chain Management Professionals showed that: ” […] 93% of shippers and 98% of third-party logistics firms feel like data-driven decision-making is crucial to supply chain activities, and 71% of them believe that big data improves quality and performance (Transmetrics, 2019) DHL, in combination with IBM, found that: “AI can help the logistics industry to redefine today’s behaviors and practices, taking operations from reactive to proactive, planning from forecast to prediction, processes from manual to autonomous, and services from standardized to personalized.” (Gesing et al., 2018, p. 15) They concluded that “companies, deciding not to adopt AI run the risk of obsolescence in the long term, as competitors seize and effectively use AI in their business today” (Gesing et al., 2018, p. 14) Predictive Shipping thus has clear future potential and is a natural evolution of current logistics. Figure 10 shows how Pre-Emptive or Predictive Logistics are the last step in the development path. 15 Figure 10: Development Path necessary for Predictive Shipping (Ericsson Industrylab, 2020, p. 7) Although these trends are on the rise, there hasn’t been widespread adoption. “Kersten found: To date, nearly half of the companies we surveyed have not planned to transform their business models digitally yet.” (Kersten et al., 2017, p. 15) Challenges Costs The most critical challenges that Predictive Shipping could face are related to increased costs. The Camelot blog says: “[…] the price pressure is becoming more intense, and this service can potentially cost a great deal, as unsold products must be returned and re-handled. (CAMELOT Blog, 2019, p. 1) If a previously unsold product is returned once again, the problem is only exacerbated even further. One needs to be very accurate in their predictions as else: “It would be entirely possible for Amazon to fudge things up and send a huge shipment of a product or item to an area where it’s not in demand. This is undoubtedly risky and could spell disaster in some instances — namely if the goods involved are time-sensitive or perishable.” (Nichols, 2018, p. 2). Making accurate predictions and reducing returns is especially important as we previously established that cost pressure is one of the most relevant current trends. Price sensibility and transparency are increasingly putting pressure on more efficient deliveries. (Kersten et al., 2017, p. 20) If Predictive Shipping fails to improve profitability, it will not be easy to implement. 16 Although there are established ways to improve the models through Big Data Analysis, targeted marketing, and implementation of assistant technologies, there is always a chance of misprediction. “It’s possible that Anticipatory Shipping could be wrong much of the time. While Big Data and predictive analysis have certainly come a long way, a computer cannot predict what a human wants with 100 percent accuracy. This could result in logistical nightmares and confusion if the shipping system isn’t as accurate as first anticipated.” (GTG Technology Group, 2015, p. 9) Wrong predictions could not only reflect in lost sales and returns but also Issues for Inventory Management, as wrongly sent items could result in stock insufficiencies (Weingarten, 2020, p. 9) Lastly, implementing this technology has many prerequisites and could be unrealistic for less advanced retailers such as Amazon. “For retailers with few warehouses and a widespread customer base, resulting in long transportation time, Predictive Shipping could be tough to implement.” (Weingarten, 2020, p. 9) Implementing Predictive Shipping would first mean vastly improving the efficiency, capacity, and number of existing warehouses and customers. However, this can be a dangerous and costly plan for many, as the DVV Media Group found out in a study: “ […] one-third of those surveyed, rate digital transformation as carrying high to very high risk.” (Kersten et al., 2017, p. 14) However, if all the issues are solved, Predictive Shipping could have a positive benefit on costs. Kundu and Tripti describe: “Predictive Shipping would enable the E-Commerce companies to reduce the costs and optimize the logistics and supply chain to a great extent, as now they would be independent, compared to previously when they were able to rely on a very limited number of shipping companies for timely deliveries. (Kundu and Tripti, 2021, p. 8) Social Aspect & worries Not only does one need accurate predictions to lower costs, but one also has to send the correct type of items. Plenty of examples of incorrect examples will be shown in the next section. Predicting purchases could creep your customers out, as it feels like an invasion of privacy, resulting in losing customers. (Hill, 2012, p. 3) An example of that will be given down below. The loss of customers could be very costly as “Acquiring a new customer is almost always more expensive than retaining an existing one (SmartData Collective, 2017). Fear due to a lack of privacy might mean that the technology has to be run in trial runs first or with a more technology- friendly selected target audience. The practical application could also result in many issues if a product were to be gifted while unattended; theft and piracy could run rampant. A good example is an Issue called Porch pirates in the USA. “Suburban areas have their challenges, such as the fear of “porch pirates” – thieves stealing parcels left by the door when no one is home to accept them. (Ericsson Industrylab, 2020, p. 5) . Theft poses a serious challenge, as Stickle et al. describe: “package theft is an emerging crime type due to the tremendous growth in online shopping and the delivery of goods directly to a home. Unattended delivery creates an opportunity for thieves to steal packages after delivery and before the resident collects them. It is believed that these types of incidents are increasing dramatically, and media attention has amplified awareness and concern of ‘porch pirates.’ (Stickle et al., 2020) 17 Customers might also feel bothered by having to pick up or return an unwanted item without ordering anything. In addition, there may not be sufficient space at the customer’s home to deliver the product, or the customer may not be at home to receive the package, further increasing complexity and potential issues. As seen in Fig. 11, the majority of packages are delivered directly to the home. However, the receiving process might be complicated if the item is too large for the postbox, and there is no good place to put the product if shipped pre-emptively. Therefore, one of the most realistic applications for Predictive Shipping might be using Smart Parcel Lockers, which would automatically notify the customer that a Product had been sent and what type of discount had been applied. Figure 11: Most used methods in package delivery worldwide in 2019 (Simmons, 2020) Cooperation of Stakeholders There are many different actors in the global supply chain. If an E-Retailer were to implement an entirely new shipping strategy and try to reroute trucks on the fly adaptively, they would have to share this information with all the different actors. As Ericsson explains: “Interconnectivity and communication throughout the entire logistics chain are key requirements when enabling a pre- emptive logistics solution.” (Ericsson Industrylab, 2020, p. 11). The Global Trade Magazine (GTM) underlines the importance of information, saying that “access to real-time information can be pivotal to improve supply-chain efficiency.” (GlobalTradeMagazine, 2016, p. 1), while Kersten found: “Digitalization of business processes and transparency in the supply chain are the most important trends, and ones that companies will need to develop considerably in the future” (Kersten et al., 2017, p. 14) This interconnectivity and communication are not always given. 34% of surveyed businesses by Ericsson said that lack of easy exchange of information is a significant obstacle in improving 18 their logistics. The opacity of the supply chain is due to many different standards, software, systems, and the number of other actors. Even if everyone is on the same page, there might be problems due to poor data hygiene (BVL, 2017, p. 2). Poor Data hygiene can result from various factors, but one of the most major is obsolete technologies. In Germany, 20 percent of the workers use VHF radio daily. VHF is outdated technology and another standard that increases complexity in communication. (Ericsson Industrylab, 2020, p. 5). To fix this problem is not easy as “digital transformations might be very risky for companies due to the high investment costs, personal shortages, and lack of expertise” (BVL, 2017, p. 1). Nonetheless, although the need for data is currently not satisfied by many, there appears to be a growing willingness to share data (Kersten et al., 2017, p. 14) In the case of Amazon’s patent for Predictive Shipping, their cooperation with different logistics companies would prove to be a challenge. “It’s unclear how FedEx and UPS would react to a system that includes complete shipping addresses or labels. According to information found in the patent, Amazon would initially only place zip codes and city names on a package – updating it with more descriptive information as it is en route. Would rates go up? Would these companies even allow this process? Both answers to these questions are unknown. (GTG Technology Group, 2015, p. 14) Cooperation is necessary for many to function in this industry. However, Amazon might be able to do things differently due to its colossal size: “Amazon launched its branded cargo planes in 2016 to increase the speed of deliveries and decrease its reliance on third-party logistics providers. For example, Amazon expanded Amazon Air last year after FedEx announced ending its express delivery contract with the company. (Shu, 2020, p. 2) Decreasing reliance on others over increasing cooperation might be necessary for Amazon, as they hinted in its 2018 10-K annual filing: “We rely on a limited number of shipping companies to deliver inventory to us and completed orders to our customers. If we are not able to negotiate acceptable terms with these companies or they experience performance problems or other difficulties, it could negatively impact our operating results and customer experience.” (Technology and Operations Management, 2018) Challenges resulting from lack of cooperation, communication, and interconnectivity can be solved. Ericsson’s report presents two excellent figures, which you can find below that describe what information has to be shared with whom (Fig. 12) and how technology can be used to implement these changes in different sections to implement Predictive Shipping (Fig. 13). These will have to be implemented to enable Predictive Shipping and meet new visibility requirements by customers. 19 Figure 12: Shared Information and Stakeholders (Ericsson Industrylab, 2020, p. 9) Figure 13: How technologies can be used to improve Cooperation(Ericsson Industrylab, 2020, p. 11) 20 Nonetheless, even these solutions have Issues associated with them, as a truly global supply chain will always have a big Issue with complying with “varying consumer data protection regulations and different cultural views on privacy altogether.” (Ericsson Industrylab, 2020, p. 12) In the end, the Issues of cooperation can only be solved if it’s economically beneficial to do so, Ericsson concludes: “The resulting benefits of improved efficiency and customer experience for everyone in this system need to outweigh the risks of potentially exposing crucial data to competitors” (Ericsson Industrylab, 2020, p. 12) Implementation methods and standings Examples Many companies already use Predictive Analytics for different needs such as forecasting, marketing, customer service, or product recommendations. Tesco in the USA, for example, used it to “[…] mail beer coupons to shoppers who bought diapers. Analysis revealed that new fathers stuck at home tending the baby drank more beer”(Fawcett and Waller, 2014, p. 5). Furthermore, “Tesco countered Walmart’s Asda by lowering prices on a set of 300 items that price-sensitive customers regularly bought, to keep these customers from visiting Asda to comparison shop”. (Fawcett and Waller, 2014, p. 5) Macy also found success in applying Predictive Analytics: “In the first 3 months after the implementation of its analytics solution, Macy’s recorded an 8–12% increase in online conversions.” (IQLECT, 2018, p. 3) These increased online conversions could’ve been combined with Predictive Shipping to increase sales and their delivery performance. Another famous example includes Target, which knew for a fact that a man’s daughter was pregnant before he did (Hill, 2012, p. 3). Target assigns all known data to a Guest ID and then analyzes the resulting patterns between Guests (Hill, 2012, pp. 1–2). Thus it was able to predict his daughter’s purchasing pattern and concluded that she was pregnant. The way those correlations are made is quite fascinating: “Pole’s colleagues noticed that women on the baby registry were buying larger quantities of unscented lotion around the beginning of their second trimester. Another analyst noted that sometime in the first 20 weeks, pregnant women loaded up on supplements like calcium, magnesium, and zinc. Many shoppers purchase soap and cotton balls, but when someone suddenly starts buying lots of scent-free soap and extra-big bags of cotton balls, in addition to hand sanitizers and washcloths, it signals they could be getting close to their delivery date. (Hill, 2012, p. 2) Thus Target was not only able to predict the pregnancy itself but also the due date. These products would be a prime target for Predictive Shipping; however, the highly emotional product could result in disastrous consequences in a miscarriage, for example. How is the current situation for Predictive Shipping? A study of 1923 business executives by Ericsson showed the following: “Two-thirds of the studied companies will use AI and data analysis to match logistics needs with capacity within the next five years, and one in five are already doing it. More than 6 in 10 companies will ship pre-emptively in the next 3–5 years. (Ericsson Industrylab, 2020, p. 2) The closest comparison to Predictive Shipping that I could currently find is a currently used tool by DHL. “DHL has developed a machine learning-based tool to predict air freight transit time delays to enable proactive mitigation. By analyzing 58 different internal data parameters, the machine learning model can predict if the average daily 21 transit time for a given lane is expected to rise or fall up to a week in advance. Furthermore, this solution can identify the top factors influencing shipment delays, including temporal factors like departure day or operational factors such as airline on-time performance. This can help air freight forwarders plan by removing subjective guesswork around when or with which airline their shipments should fly” (Gesing et al., 2018, p. 25). The DHL can more accurately send products closer to distribution centers and reduce shipping times by predicting delays. Another way by the DHL to potentially implement Predictive Shipping is “the SmartTruck routing initiative in the early 2000s to develop proprietary real-time routing algorithms for its fleet operators and drivers. The new soft infrastructure of cities such as digital and satellite maps, traffic patterns, and social media check-in locations are creating a wealth of information that can augment systems like SmartTruck and improve the overall routing of truck drivers on delivery runs” (Gesing et al., 2018, p. 26). The increased routing potential and transparency within the deliveries could ease the difficulty of routing the trucks correctly. HBR gives two examples of how Predictive Shipping might be implemented in the real world: “Imagine you’re about to leave the house to pick up your kids. As you grab your keys, you hear a voice from the device on your coffee table: “It looks like you’ll use the last of your milk tomorrow, and yogurt is on sale for $1.19. Would you like to pick up an order from Trader Joe’s for a total of $5.35?” You say yes, and Alexa confirms. The order will be ready for curbside pickup, on the way home from your kids’ school, in 15 minutes.” (HBR, 2016, p. 1) and “Say you’re on a business trip and realize you forgot your phone charger. You’ll pay a premium for a new one delivered to your hotel room before an all-day meeting.” (HBR, 2016, p. 2) I have analyzed how Predictive Analytics has been used in the past in various examples. Thus, the question becomes: what is the current progress within its implementation in the next decade? Current Progress The GTG believes that Amazon’s culture and vision is a crucial contributor to the viability of Predictive Shipping (GTG Technology Group, 2015, p. 6). A specific size and desire to implement new technologies must be present to be the first to implement Predictive Shipping. Only once Predictive Shipping becomes an industry-standard might others follow to match the same-day shipping service given by Amazon. Before we see widespread adaption, the prerequisites will have to be fulfilled, as Camelot describes: The first step is to predict the right products, time, and place/area. To derive reliable results for one distribution area, the underlying data must be plentiful and of high quality (CAMELOT Blog, 2019, p. 1) Nonetheless, there are reasonable indications to be optimistic about Predictive Shipping. In a Poll by the MHI, Predictive Analytics was used by 31% of surveyed companies, while 48% planned on using it. (Fig. 14) 22 Figure 14: Current Usage in % of Predictive Analytics (MHI, 2021, p. 2) The same study found that Predictive Analytics would be used in large parts for Demand Forecasting, Customer insights, and Supply planning, all necessary steps to introduce Predictive Shipping (compare Fig 15.) Figure 15: Top uses for Predictive Analytics (MHI, 2021, p. 10) Although the trends seem to indicate a rise in Predictive Shipping, the future of Predictive Shipping remains unclear, and many questions have to be resolved first. “[Predictive Shipping] raises a variety of questions. For instance, what are the competitive implications of Predictive 23 Shipping—both for Amazon and its rivals? What does Predictive Shipping mean for returns? What type of supply chain infrastructure does a company need to compete? How will supplier and service provider relationships change?” (Fawcett and Waller, 2014, p. 5) Research Method While working on Predictive Shipping, there was a distinct Issue that would dramatically increase the difficulty of this thesis. The issue was that publicly available data related to Predictive Shipping was close to nonexistent. As my internship had little to do with Predictive Shipping or the retail industry in general, action research was not possible, and I had three options left: case studies, surveys, or simulations. First, Case-Study research wasn’t possible due to the lack of widely disclosed information and literature. Second, I could try to do surveys; however, I didn’t want to risk a lack of responses and thus working data. Finally, I found that the current usage of Predictive Analytics is already surveyed sufficiently in the existing literature and that information concerning Predictive Shipping might be harder to access. These circumstances left me with a simulation/optimization approach. This approach was preferable to me due to my analytical approach and the mathematical nature of Predictive Shipping. The next issue became the question of what to simulate. Simulating an entire worldwide or countrywide distribution network with inventory management, shipping schedules, and transport fleets might be possible with software answers such as anyLogistix; however, those were outside my competencies and would have blown the scope of this thesis out of proportion. Thus, I focused on a specific part of Predictive Shipping, mainly how goods might be shipped to the customer immediately and then sold at a heavy discount or gifted if declined. This was a significantly under-researched part, especially in terms of specific values. I came upon some blogs that described the soft factors but no previous models. During my research, I came upon the importance of returns in E-Retailing. Therefore, I made this the focal point of my thesis and combined this topic with the specific part of the immediate shipment by Predictive Shipping. I will create a model based on multiple assumptions that simulates this process in Excel and explain whether it might be profitable. Hopefully, this model can give managerial insights and be an impetus for future research. 24 Problem Description The overall structure of the Predictive Shipping model is simple: (compare Fig. 16) Figure 16: Predictive Shipping Model To create a model that can simulate the profitability of an immediate shipment of a product, I have to make two distinctly separate parts: 1. The first is a prediction algorithm that associates a likelihood of purchase or correlation with other goods. This algorithm serves to provide values for the next step 2. The second step is an ordering simulation that shows whether a purchase was successful or not to calculate the total profitability. To create prediction algorithms, Collaborative Filtering and Clustering Algorithms are used (Edosio, 2014, p. 7). A clustering algorithm identifies groups of users with similar preferences and clusters them into another unique group. It is then assumed that those within the group would have comparable choices in their item purchases. Lee used clustering Algorithms previously in the form of association rule mining to understand the correlations within groups in their optimization model for Predictive Shipping “Association rule mining is applied to discover the relationship among purchased items from customers in the whole distribution network.” (Lee, 2017, p. 595) Collaborative filtering compares a customer with a database of preferences and then matches the most suitable product in a recommendation. It is more closely related to how Amazon creates recommendations for Predictive Shipping. Edosio describes how it is used: “Search-based (also called content-based search) utilizes a consumer purchase history and rated item to create a search query that finds other things (such as author or similar genres) similar to consumers’ taste. For example: if a customer purchases a DVD called the “God Father.” The product recommender will recommend movies from similar authors, similar genres, and similar directors (Linden et al., 2003). Amazon has access to even more information than just purchase history. Amazon can supplement its Predictive Shipping algorithm through various channels (Zhalgassova, 2014, p. 1). These include: • Customer information: gender, sign-up year, segment (mainly dependent on profitability) • Order information: Order date, products ordered, total number of orders per customer • View information: Number of product page visits of a customer, date, and length of visit • Event information (information on where a customer clicked on a product page): event type (e.g. “click on the image,’’ add to cart), event date, the total number of clicks of a customer 25 • Product information: product category Additionally, the season, month, and weekday on which a customer viewed a product for the first time were included as variables. • The number of times a customer opened a product page, and the number of events on a product page were calculated. • The average decision time per customer was calculated, which is the average time between the first date a product was viewed and the order date. (Weingarten, 2020, p. 3) As I had previously mentioned in section 2.2.1.4, “Implementation of AI”, BDA and AI can significantly benefit a company. One of the ways is, for example, another way of creating a collaborative filtering or clustering algorithm. In Weingarten’s case, they developed different prediction models in the fashion Industry with “different forecasting methods, namely logistic regression (LG), random forest (RF), neural network (NN), and (one-class) support vector machine (SVM).” (Weingarten, 2020, p. 4) For the sake of this thesis, I assumed that any company trying to implement Predictive Shipping would already have some prediction algorithm or collaborative filtering. The exact numbers used are described in more detail in section 5. “Predictive Shipping simulation.” To calculate profitability, I used the following simple formula: Profit = revenue from sale - shipping cost The revenue from sales depends on whether a product is sold, gifted, or a discount was applied. The shipping cost is set to the average shipping cost for last-mile delivery and is counted twice if a return is made. To see whether the product is accepted based on the probabilities that I determined from my collaborative filtering matrix, I used a Monte-Carlo simulation software. The software runs through my prediction algorithm with a set number of trials and gives me distribution of successful or not attempts with their corresponding profitability. I then analyze this distribution for statistical validity and see in which cases Predictive Shipping was profitable. 26 Predictive Shipping Simulation Model The model was made in Microsoft Excel 365, with the “Opensolver” and “Oracle Crystal Ball” add-ins. Due to the lack of data described in section 3 “research method”, many assumptions had to be made that can be filled with actual data to be applied in the real world. Prediction Algorithm I first assumed existing historical data already, e.g., the purchasing data/collaborative filtering matrix. To simulate this, I made a matrix consisting of nine different items, with a value between 0 and 1 between them. The value is the likelihood of purchasing another good through “if -> then” logic. I then created a correlation table through the excel functionality to normalize said data. I now had correlations between all nine items. It is important to note that I manually changed the historical data’s 100% likelihood to a null value in the correlation table for the calculations afterward to function. Both tables can be seen down below in Fig. 17 Figure 17: Historical Data Assumed Figure 17: Resulting Correlation Table Another way of creating this correlation table is making an extensive array of random numbers between the items and then forming it through the same correlation table function in excel. Order Simulation System I now have the likelihood that a particular good will be bought if another one has been purchased. So now I create a model which selects the best possible item for a random customer with an unexpected purchase history. This prediction would not always be reliable in a real-world application, and there is a high chance of false positives. The GTG mentioned while describing algorithms that: “an algorithm or data-driven system cannot predict what a customer wants with 100 percent accuracy. In reality, it will probably be challenging to get anywhere near 80 or 85 percent accurate. (GTG Technology Group, 2015, p. 14) 27 The way I did it in Excel was through the use of “Open Solver”. I first created a “buying matrix”, which shows the customer’s purchase history. In this matrix, 1 represents a sale, and 0 illustrates the opposite. I then made an “ordering matrix” filled with variable binary cells for the Open Solver. In this matrix, Open solver would test out buying each item once and see which had the best correlation with all the other previously purchased items. To confirm that only one item was selected, I created a constraint, which said that the sum of all bought items could not be larger than 1. Open solver found the best possible recommendation by comparing the correlation of each potential item for each customer with each previously bought item through a series of Index Match functions and then adding their correlations together. The item with the highest resulting correlation then got chosen as the product order. The entire excel table used can be found in Figure 18 below. Figure 18: Order Simulation System The result of this optimization problem was as follows: • Customer A: Item 3 recommendation, 0.546 correlation • Customer B: Item 6 recommendation, 0.816 correlation • Customer C: Item 1 recommendation, 0.373 correlation These values were quite good for the sake of this model, as now I had representatives for low (<0.4 correlation), middle (>0.4, <0.8 correlation), and high (>0.8 correlation) cases and could analyze the impact of those on profitability. Monte Carlo Simulation After finding the best possible item for each customer, I simulate the profitability of shipping this good to the customer. As mentioned previously, the profit is the revenue from the item’s sale minus the transportation costs. 28 The simulation is a three-stage process. In the best case, the product gets sent to the customer and gets bought instantly. If this doesn’t occur, the customer receives a discount and another opportunity to purchase the product. If the discount also gets declined, there is a chance that the product gets given as a gift. Finally, if the customer refuses the present, then it gets returned to the warehouse. I’ve illustrated this process in Fig. 19 Figure 19: Process Diagram for Monte Carlo Simulation The excel-table used for this calculation can be seen in Fig. 20. For a Monte Carlo Simulation, one needs variables. In this case, I tried to introduce randomness into the model by making the values follow normal and lognormal distributions: Figure 20: Monte Carlo Simulation Excel Table • Prices of items: Normal distribution (20 mean, 5 std. deviation) • Transportation costs: Normal distribution (10 mean, 1 std. deviation) • The base chance of returning an item (wrong size, color, other attributes): Lognormal distribution (0.15 mean, 0.1 std. deviation) • The increased chance of buying an item due to a discount: Lognormal distribution (0.15 mean, 0.1 std. deviation) • The increased chance of buying an item due to a gift: Lognormal distribution (0.15 mean, 0.1 std. deviation) 29 The transportation costs were based on a study on the average last-mile delivery costs (compare Fig. 21). Figure 21: Average last-mile delivery cost worldwide in 2018 The prices were thus approximated proportionally to the transportation costs so that there would be higher, lower, or no profitability. The chances of returning, discounts, and gifts were all made lognormal, as negative percentages don’t make sense, and some people might be more influenced by those proposals than others. The discount was set to a constant value of 20%, however in the real world: “historical data such as previous purchases, clickstream, cookies, enterprise resource planning systems could be used to set prices of an item or offer customized discounts dynamically.” (Edosio, 2014, p. 8) If the technique was further developed, one might even take it one step further and change prices based on a purchase’s probability and current trends. IQLECT describes this process: “E- commerce stores must continually predict trends in product prices, taking advantage of festivals or periods of high visitor traffic on their e-commerce store. With predictive analytics, e- commerce businesses can create enhanced product pricing models, determining the optimum prices to maximize conversions effectively. Predictive analytics capabilities in a robust analytics solution can analyze historical data for different products, analyze customer responses to past pricing trends, and evaluate competitor pricing, to build suitable pricing models for e-commerce businesses.” (IQLECT, 2018, p. 4) Another necessary exemption in this model is that it was made assuming one warehouse sent it to the customer and back. However, if there are multiple warehouses, decisions must be made from where it should be sent, and prediction confidence becomes an important topic. Lee addressed this during his model for Predictive Shipping: “Taking the confidence of the prediction results into account is very important when improving one factor leads to the scarification of the others. For example, when the cost required for shipping a product type to DC A is lower than DC B, but it is predicted that having a purchase near DC B is of greater confidence than that near DC A, 30 then the retailers have to face a trade-off between the cost and the prediction confidence.” (Lee, 2017, p. 594) Results and managerial implications The simulation was run for a total of 5000 trials. The data used was described in the section “Monte Carlo Simulation.” In Figure 22, one can see the profit frequency distribution for each customer. One can see that the strength of the purchase prediction has a massive impact on profitability. Losses group around the -20 and -10 mark as expected; however, customer A and especially customer C are far more frequently in the negative than customer B. Figure 22: Profit frequency distribution The table below shows the mean, median, and standard deviation for each customer. Customer Mean Median Standard Profitable % of Deviation the time A 0.27 3.82 11.67 58.86% B 6.86 8.09 7.83 85.66% C -5.9 -9.62 12.45 37.21% If one takes the mean as an average for profitability, B was profitable, A barely profitable, and C created more losses than profit. The lower the chance to buy a good, the higher the standard deviation rose. The likelihood to buy an item was surprisingly close to the resulting profitability in %. For example, customer A, who had a 54.6% chance of purchasing product 3, was profitable in 58.86% of the cases, only an approximately 4% difference. Regarding Gifts and Returns, A saw nearly 2000 out of 5000 attempts that couldn’t be sold at a discount, gifted, or returned. It was unbelievably 2982 attempts for C, so 59.64% of the time, no sale would have been made. It is thus not a surprise that C was not profitable. Meanwhile, for customer B, the benefit of getting a gift was so high, that every item that wasn’t sold was accepted as a Gift, and none were returned. The lack of returns was vital for B’s profitability, as return costs didn’t need to be calculated. Nonetheless, it is essential to note that an increase in the item’s sales revenue would not have changed the profitability % of the time. Assuming B cost 60€ instead of 20€ and was sold under the same costs, the profits on existing sales would have been higher, but 14.34% of the time still a gift would have been made. 31 Customer Chance of a gift Chance of a return A 26% 12.3% B 11% 0 C 31% 28.6% What do these results mean? According to my calculations, it seems like one can predict the profitability of delivery of an item sent beforehand quite accurately if the surrounding data is known in advance. Thus it is crucial to have an accurate understanding, as return rates might be completely different based on the industry: ”[…] retail data technology company Clear Returns suggested figures of 25% return rates for apparel by UK consumers and a massive 75% return rate by German customers” (Jack et al., 2019, p. 3). To use this model in a real-world application, one would have to know the base likelihood that an item will be returned how this can be done as described in the “return” section. Furthermore, the price of goods is significant for profitability. If sale revenue can’t cover the transportation costs, then sending the product out preemptively isn’t reasonable. Conversely, if the product prices were significantly higher, this method could be very beneficial with the return chances staying the same. However, this is another assumption, as a more expensive product is probably more likely to be returned due to the capital commitment. To conclude, in this experiment, the most critical factor was prediction certainty. All products behaved similarly as expected to their correlation calculated in the Order Simulation System. Higher confidence of purchase will always give better results, so the primary goal of future Predictive Shipping research should be to make predictions as accurately as possible. If the certainty is high enough, over time, Predictive Shipping should be profitable. 32 Conclusion and future outlook In this thesis, I have analyzed the opportunities and challenges and the current standings for implementing Predictive Shipping. I created a model to simulate Predictive Shipping and found specific scenarios in which profitability can be achieved. The predictability of the profitability rises with the availability of data and confidence of purchase. As no data was available to me during my research, I created multiple assumptions and replications for existing models such as the prediction algorithm. Transportation times and capacity limits were ignored entirely, and prices were held to a reasonable standard. The chances of buying a product were also assumed based on distributions. There is still lacking research on whether Predictive Shipping can be implemented, in large part due to a lack of trial runs and experiments in the real world. However, as transparency and cooperation and the adoption of AI technology within the supply chain increase in the following years, Predictive Shipping is sure to become an essential topic within the near future. Trial runs within select communities would give more accurate information about the acceptance and usage by the general populace and the rate of fraudulent returns. The next step in research would be to apply my model and approach to real-world data and see where data can reduce uncertainty within the model. It would also be beneficial to include more cost components besides transportation costs, such as inventory, handling fees, or administrative expenses, in the cost calculation. Declaration of originality/Statutory declaration I confirm that this assignment is my own work and that I have not sought or used the inadmissible help of third parties to produce this work and that I have clearly referenced all sources used in work. I have fully referenced and used inverted commas for all text directly or indirectly quoted from a source. This work has not yet been submitted to another examination institution – neither in Germany nor outside Germany – neither in the same nor in a similar way and has not yet been published. Berlin, the 09.19.2021 33 Reference list AIMultiple (2019) What is Analytics? How is it evolving? [2021 update], 1 January. Available at: https://research.aimultiple.com/what-is-analytics/ (Accessed: 14 September 2021). Alicke et al. (2016) ‘Supply Chain 4.0 in consumer goods’ (Accessed: 22 April 2021). Amazon Staff (2021) ‘New Amazon Programs Aim to Curtail Rampant eCommerce Return Waste’, Sustainable Brands, 8 October. Available at: https://sustainablebrands.com/read/waste- not/new-amazon-programs-aim-to-curtail-rampant-ecommerce-return-waste (Accessed: 27 August 2021). Blair, A. (2020) ‘How Retailers Can Control Costs Even As Return Rates Climb’, Retail TouchPoints, 9 March. Available at: https://retailtouchpoints.com/topics/fulfillment-last-mile/ how-retailers-can-control-costs-even-as-return-rates-climb (Accessed: 27 August 2021). BVL (2017) ‘15 Trends beeinflussen Strategie und Praxis der Logistik: Der Endkunde bestimmt den Takt der Digitalisierung’ (Accessed: 22 April 2021). CAMELOT Blog (2019) Predictive logistics in e-commerce: Shipped before you click “add to cart” - CAMELOT Blog, 3 February. Available at: https://blog.camelot-group.com/2019/ 02/predictive-logistics-in-ecommerce-shipped-before-you-click-add-to-cart/ (Accessed: 28 June 2021). DHL Customer Solutions (2020) ‘The Logistics Trend Radar: Delivering insight today, creating value tomorrow.’ (Accessed: 22 April 2021). Eckerson (2007) Predictive analytics. Available at: http://download.101com.com/pub/tdwi/files/ pa_report_q107_f.pdf. Edosio, U. (2014) ‘Big Data Analytics and its Application in E-Commerce’. Available at: https:// www.researchgate.net/publication/264129339 (Accessed: 22 April 2021). Ericsson Industrylab (2020) ‘Pre-emptive logistics – the road ahead’. Available at: https:// www.ericsson.com/4a310f/assets/local/reports-papers/industrylab/doc/pre-emptive-logistics- report.pdf (Accessed: 1 September 2021). Fawcett, S.E. and Waller, M.A. (2014) ‘Supply Chain Game Changers-Mega, Nano, and Virtual Trends-And Forces That Impede Supply Chain Design (i.e., Building a Winning Team)’, Journal of Business Logistics, 35(3), pp. 157–164. doi: 10.1111/jbl.12058 Fuchs, J.G. (2019) ‘Amazon entwickelt neues Werbeformat und verschickt Gratisprodukte’, t3n Magazin, 28 February. Available at: https://t3n.de/news/amazon-samples-werbeformat- marketing-1147094/ (Accessed: 18 July 2021). Gesing et al. (2018) ‘Artificial Intelligence in Logistics: A collaborative report by DHL and IBM on implications and use cases for the logistics industry’. Available at: https://www.dhl.com/ content/dam/dhl/global/core/documents/pdf/glo-core-artificial-intelligence-trend-report.pdf (Accessed: 21 April 2021). 34 GlobalTradeMagazine (2016) ‘Big Data = Big Changes in Logistics, Transportation’, Global Trade Magazine, 11 February. Available at: https://www.globaltrademag.com/big-data-big- changes-logistics-transportation/ (Accessed: 10 September 2021). GTG Technology Group (2015) ‘Will Anticipatory Shipping Throw a Wrench In Logistics Management’. Available at: https://gtgtechnologygroup.com/wp-content/uploads/2015/01/GTG- White-Paper-Design_Will-Anticipatory-Shipping-Throw-a-Wrench-In-Logistics- Management.pdf (Accessed: 25 April 2021). HBR (2016) How Predictive AI Will Change Shopping, 17 February. Available at: https://hbr.org /2016/11/how-predictive-ai-will-change-shopping (Accessed: 14 September 2021). Hill, K. (2012) ‘How Target Figured Out A Teen Girl Was Pregnant Before Her Father Did’, Forbes, 16 February. Available at: https://www.forbes.com/sites/kashmirhill/2012/02/16/how- target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/?sh=6f95d7396668 (Accessed: 4 August 2021). IQLECT (2018) ‘The Importance of Predictive Analytics for E-commerce Stores’, The Startup, 14 November. Available at: https://medium.com/swlh/the-importance-of-predictive-analytics-for- e-commerce-stores-d7ef0ce2d32e (Accessed: 14 September 2021). Jack et al. (2019) ‘Buy Online, Return in Store: The Challenges and Opportunities of Product Returns in a Multichannel Environment’ (Accessed: 27 August 2021). Kaplan, M. (2014) ‘Anticipatory Shipping: Amazon’s Approach to Influencing Purchases’, Practical Ecommerce, 3 December. Available at: https://www.practicalecommerce.com/ Anticipatory-Shipping-Amazons-Approach-to-Influencing-Purchases (Accessed: 25 April 2021). Kersten, W. et al. (2017) Chancen der digitalen Transformation: Trends und Strategien in Logistik und Supply Chain Management. Hamburg: DVV Media Group GmbH (Accessed: 22 April 2021). Kundu and Tripti (2021) View of Predictive analytics to offer profitable anticipatory shipping for E-commerce Companies, 27 August. Available at: http://www.modern-journals.com/index.php/ ijma/article/view/733/629 (Accessed: 27 August 2021). Lee, C. (2017) ‘A GA-based optimisation model for big data analytics supporting anticipatory shipping in Retail 4.0’, International Journal of Production Research, 55(2), pp. 593–605. doi: 10.1080/00207543.2016.1221162 Merchant Fraud Journal (2021) Return Fraud: How Merchants Lost $25.3B in Fraudulent Returns, 28 April. Available at: https://www.merchantfraudjournal.com/return-fraud-how- merchants-lost-25-3b-in-fraudulent-returns/ (Accessed: 17 September 2021). MHI (2021) ‘Innovation Driven Resilience: How technology and innovation help supply chains thrive in unprecedented times’, pp. 1–40. Available at: https://locusrobotics.com/wp-content/ uploads/MHI-Annual-Industry-report-2021.pdf. 35 Nichols, M.R. (2018) ‘Amazon Wants to Use Predictive Analytics to Offer Anticipatory Shipping’, SmartData Collective, 16 January. Available at: https://www.smartdatacollective.com/ amazon-wants-predictive-analytics-offer-anticipatory-shipping/ (Accessed: 25 April 2021). NRF (2021) $428 Billion in Merchandise Returned in 2020, 17 September. Available at: https:// nrf.com/media-center/press-releases/428-billion-merchandise-returned-2020 (Accessed: 17 September 2021). Pallot, R. (2021) ‘Amazon destroying millions of items of unsold stock in one of its UK warehouses every year, ITV News investigation finds’, ITV News, 22 June (Accessed: 30 August 2021). Purdy, K. (2021) ‘Destroying Unsold Products Isn’t a Scandal, It’s How Amazon Works’, iFixit, 17 September. Available at: https://de.ifixit.com/News/50886/destroying-unsold-products-isnt-a- scandal-its-how-amazon-works (Accessed: 17 September 2021). PYMNTS (2019) ‘Can Predictive Retail Really Take Off?’ PYMNTS.com, 6 April. Available at: https://www.pymnts.com/news/ecommerce/2019/predictive-retail-amazon-logistics-fulfillment/ (Accessed: 14 September 2021). Shu, C. (2020) ‘Amazon Air adds 12 new aircraft to its cargo fleet, expands its ground operations’, TechCrunch, 6 April. Available at: https://techcrunch.com/2020/06/04/amazon-air- adds-12-new-aircraft-to-its-cargo-fleet-expands-its-ground-operations/ (Accessed: 1 September 2021). Simmons (2020) ‘Live Commerce: Last Mile Delivery’. Available at: https://img1.wsimg.com/ blobby/go/6f4ecff1-2ea2-4a44-b1fd-a8d63b1a6d05/downloads/last-mile-delivery.pdf?ver= 1622190941315 (Accessed: 27 August 2021). Škapa, R. (2013) ‘Application of the theory of planned behavior to fraudulent returning’, Acta Universitatis Agriculturae et Silviculturae Mendelianae Brunensis, 60(7), pp. 379–386. doi: 10.11118/actaun201260070379 SmartData Collective (2017) ‘5 Applications of Predictive Analytics’, SmartData Collective, 9 February. Available at: https://www.smartdatacollective.com/applications-predictive-analytics/ (Accessed: 25 April 2021). Spiegel et al. (2013) US8615473B2 - Method and system for anticipatory package shipping - Google Patents, 22 April. Available at: https://patents.google.com/patent/US8615473B2/en (Accessed: 22 April 2021). Stickle, B. et al. (2020) ‘Porch pirates: examining unattended package theft through crime script analysis’, Criminal Justice Studies, 33(2), pp. 79–95. doi: 10.1080/1478601X.2019.1709780 SupplyChain24/7 (2021) How You Can Use Predictive Analytics to Reduce Parcel Shipping Costs - Supply Chain 24/7, 28 June. Available at: https://www.supplychain247.com/article/how_ you_can_use_predictive_analytics_to_reduce_parcel_shipping_costs/Ecommerce (Accessed: 28 June 2021). 36 Technology and Operations Management (2018) Anticipatory shipping—retail’s crystal ball? - Technology and Operations Management, 15 November. Available at: https://digital.hbs.edu/ platform-rctom/submission/anticipatory-shipping-retails-crystal-ball/ (Accessed: 10 September 2021). Transmetrics (2019) How Predictive Analytics Is Transforming Logistics and Supply Chain - Transmetrics, 22 March. Available at: https://www.transmetrics.ai/blog/predictive-analytics-in- logistics/ (Accessed: 10 September 2021). Urbank P et al. (2015) ‘Predicting Product Returns in E-Commerce: The Contribution of Mahalanobis Feature Extraction’. Available at: https://www.researchgate.net/publication/ 283271154_Predicting_Product_Returns_in_E-Commerce_The_Contribution_of_Mahalanobis_ Feature_Extraction_Completed_Research_Paper (Accessed: 17 September 2021). Weingarten, S.S. (2020) ‘Shortening Delivery Times by Predicting Customers’ Online Purchases: a Case Study in the Fashion Industry’ (Accessed: 22 April 2021). Wharton University of Pennsylvania (2020) The High Cost of Returns: Should Retailers Rethink Their Policies?, 19 August. Available at: https://knowledge.wharton.upenn.edu/article/high-cost- of-returns-should-retailers-rethink-policies/ (Accessed: 19 August 2021). Wishart (2021) ‘Amazon is destroying millions of unsold goods, from smart TVs to laptops’, Fast Company, 7 March. Available at: https://www.fastcompany.com/90652617/amazon-is- destroying-millions-of-unsold-goods-from-smart-tvs-to-laptops (Accessed: 17 September 2021). Zhalgassova (2014) ‘Logistik mit Gedanken lesen: Anticipatory Shipping im Onlinehandel’ (Accessed: 22 April 2021). 37 Table of figures Figure 1: Retail e-commerce sales worldwide from 2014 to 2023 (Simmons, 2020) ..................... 3 Figure 2: Predictive Analytics in comparison (AIMultiple, 2019) .................................................. 4 Figure 3: A graph depicting the current standings and relevance of modern technology trends. (BVL, 2017, p. 4)............................................................................................................................. 5 Figure 4: Most essential technology investments (Simmons, 2020) ............................................... 6 Figure 5: How Predictive Shipping works (Spiegel et al., 2013) .................................................... 7 Figure 6: Cross border delivery date (Simmons, 2020)................................................................... 9 Figure 7: Most crucial last-mile delivery areas (Simmons, 2020) .................................................. 9 Figure 8: Unopened Products about to be destroyed (Merchant Fraud Journal, 2021) ................. 10 Figure 9: Merchandise Return Fraud (Merchant Fraud Journal, 2021, p. 2) ................................. 12 Figure 10: Development Path necessary for Predictive Shipping (Ericsson Industrylab, 2020, p. 7) ................................................................................................................................................ 16 Figure 11: Most used methods in package delivery worldwide in 2019 (Simmons, 2020) .......... 18 Figure 12: Shared Information and Stakeholders (Ericsson Industrylab, 2020, p. 9) ................... 20 Figure 13: How technologies can be used to improve Cooperation(Ericsson Industrylab, 2020, p. 11) .............................................................................................................................................. 20 Figure 14: Current Usage in % of Predictive Analytics (MHI, 2021, p. 2) ................................... 23 Figure 15: Top uses for Predictive Analytics (MHI, 2021, p. 10) ................................................. 23 Figure 16: Predictive Shipping Model........................................................................................... 25 Figure 17: Resulting Correlation Table ......................................................................................... 27 Figure 18: Order Simulation System ............................................................................................. 28 Figure 19: Process Diagram for Monte Carlo Simulation ............................................................. 29 Figure 20: Monte Carlo Simulation Excel Table........................................................................... 29 Figure 21: Average last-mile delivery cost worldwide in 2018 .................................................... 30 Figure 22: Profit frequency distribution ........................................................................................ 31 38
Enter the password to open this PDF file:
-
-
-
-
-
-
-
-
-
-
-
-