A decision-making framework for automating distribution centers in the Retail supply

Abstract

Warehouse/distribution center (DC) automation technology for the retail industry promises to reduce operational costs, improve flexibility and response time for customers, and help improve network productivity, thus making it very relevant for omni/multichannel (OC/MC) settings. However, the investment required to acquire the DC automation technology is high, and hence, the investment decision must be operationally and financially comprehensive. In fact, an automated DC has a network-wide impact: it can benefit players in the network, but in turn is exposed to network risks and the investment must be safeguarded. While the need for a comprehensive decision-making framework and safeguarding strategy is stressed by scholars, such a framework is lacking. Further, corresponding integrated sub-frameworks for key elements in the OC/MC value chain are also missing. In this paper, we address these gaps and contribute by providing a) generalized and integrated three-part framework, b) corresponding sub-frameworks, c) discrete event, economic, and math programming models, d) rapid-sizing/analysis tools based on: i) analysis at the DC-level, ii) network level, iii) economic/business level, and iv) contract level (sustainable supplier/distribution relationship). In this reference, we investigate a new generation ‘full-case’ technology that has been recognized as a key to warehouse automation. The insights from our research inform several strategic tradeoffs (extent of automation, investment in labor vs. capital, response vs. efficiency, and sustainable supplier management) relevant for decision-making and safeguarding an expensive asset such as an automated DC. Our analysis is based on interviews (retailers, automated and conventional DCs, and DC equipment suppliers), on-site observations, secondary data, and learning from analytical models. We also present an illustrative real-life application/case study of the framework and the modeling details in the E-component.

Highlights

• •Present integrative framework (OC/ MC setting) for investment decision-making for DC, considering degree of automation
• •While we focus on ‘full case’ technology, our work has general appeal.
• •Three sub-frameworks include i) DC sizing ii) Network analysis ii) Economic analysis; rapid-sizing provided in E-component
• •We identify four levels of analysis: i) strategic, ii) supply/distribution chain, iii) DC, iv) store.
• •We validate our work through: a) organizations in item 5 above, b) conferences and c) trade publications.

1. Introduction

Research scholars have considered operations effectiveness (flexibility/responsiveness and costs) as key to competitiveness in the retail industry, given the trend toward multichannel (MC/OC) retail [1,2]. While distribution centers have been considered as a key element to the distribution/supply chain network (from functionality and costs perspectives),1automated distribution centers assume even more significance, given the more recent trends in MC/OC retail. Scholars note lack of an integrated treatment of important aspects of OC value chain [3]. Understanding the key variables involved in successfully implementing OC important for our proposed investment decision-making framework. While automated DCs provide significant benefits, given the context, they require high investments (in the range of $50 M to $150 M for medium-sized DC2) but surprisingly a holistic and systematic framework for investment decision-making is not available – from practice or academic perspective. Moreover, since such an integrated investment framework is lacking, in practice, investment decisions are made based on myopic considerations (such as labor cost savings alone). This is a key motivation for our work – to provide a framework and tools that are helpful for managers in the retail industry (with general appeal to supply distribution networks) and to contribute to research in this context by providing sub-frameworks and corresponding models.

Furthermore, in the retail context, automated DCs have the potential to affect transportation costs, inventory costs, and labor costs at the supply/distribution network level and store costs/benefits at the store level. This link has not highlighted in extant literature and requires further research, especially in the OC/MC context. Further, we note that stores now play a more important role3 since they serve both online and offline customers [[4], [5], [6],7,8]. In this context, there is an increasing interest among retailers to improve DC productivity and flexibility while reducing DC operation costs, and DC automation provides a promising solution – this area would gain from further research. Moreover, the retail sector (especially the “predictable physical” activities) has a high potential for automation. In addition, more recently, an important factor driving the change is the reorganization and integration of channels [[9], [10], [11], [12],8]: Multi-channel (MC) and Omni-channel (OC). Investigation of supply and distribution network from this perspective in retail context is nascent area – from practice and academic perspective. We find a variety of distribution and fulfillment strategies in the US and Europe (from pure play in bulk and single-item to OC – which is concerned with completely integrated channels). While the e-commerce part of total retail trade is relatively low (for example, in the case of Germany: 7 % in 2012 – [10]; 15.9 % in 2019 – Center for Retail Research4), the growth in retail trade, for all the EU economies (including the UK) that is attributed to e-commerce is 47.5 % between 2015 and 2019 (for Western EU economies). (Center for Retail Research, 2019 report now updated to 2022). Scholars have reported on different strategies by retailers: Bricks and Mortar [BM] continuing to stay as BM, e-commerce pure-plays moving to Bricks-and Click [BC], BM moving to BC [12,13,14]. These scholars highlight the importance of a holistic view involving: a) warehousing and distribution center-design and automation, b) supply management, and c) distribution/supply network design to ensure requisite inventory (at DCs and stores), flexible delivery schedules to stores, and returns (from DCs and stores) in a dynamically evolving retail industry [[14], [15], [16]], and d) importance of stores in MC/OC setting [4,5,9,8]. In this paper, we provide such a holistic perspective and a corresponding integrated decision-making framework for automated DCs that influences all key elements of the supply/distribution network. The need for such an integrated framework has been highlighted by researchers and practitioners alike [9,17,18]. We respond to this gap in the literature and extend the work of these scholars by i) presenting a three-part framework and sub-frameworks [for items a), b), c) above], ii) providing an in-depth case study based on our long-term association with a retailer (US), technology provider and equipment manufacturer (EU), and benchmarked warehouses (US and EU), iii) offering several analytical tools and models that will help further future research and will help practitioners employ them to make automated DC and warehouse-related decisions, and iv) providing an added dimension that is seldom studied – we address the issue of safeguarding investments in the network through sustainable supply chain management and contacting framework (one of the three sub-frameworks). We do not consider last-mile delivery explicitly. Our models assume that customers pick up from local retail stores (the network has small, medium, and large-sized stores in a city) and the DC/warehouse supply to the retail stores.

We refer to warehouse or distribution center automation [19],5 in the sense of automation of the automated storage and retrieval systems (ASRS). As per [[20], page 397], “A warehousing system consists of hardware, which can be further subdivided into storage devices (e.g., a rack), material handling systems (e.g., a conveyor belt), and picking tools (e.g., a pick-by-voice solution or a picking workstation), and processes defining the workflow along the applied hardware elements.” While we also take this DC perspective, we discuss three different automation modules with different levels or degrees of automation. Our scope includes a) automated as well as manual DCs (it is important to discuss the ability of a warehouse to handle pallets, cases, and sub-cases in an MC setting), b) network transportation that works with automated (case-based) or manual (pallet-based) DCs, and c) developing and managing a sustainable supply network to ensure safeguarding the investment in such key network assets as automated DCs. Thus, we take a longer-term perspective to ensure an acceptable and consistent return on investment (ROI) in automated DCs.

Our work is motivated by the growing adoption of new case-based [‘full-case': [[21], [22]]] distribution center (DC) automation technology and the need for a comprehensive framework for decision-making for investment in warehouse automation – given the need to meet the demands of the MC setting. Specifically, based on the literature, interviews, field experience, and analysis we develop a framework (Fig. 1a: Framework, Fig. 1b: Framework operationalization), supporting sub-frameworks (see Fig. 2a, Fig. 2ba/b, Fig. 3a, Fig. 3ba/b, Fig. 4a, Fig. 4ba/b), corresponding analytical models and rapid-sizing tools, and discuss how these could be deployed in practice (see E-component for analytical models). Thus, while we focus on the framework, strategic issues, and managerial insights in the paper, our work has an analytical foundation – based on our empirical work. Although we focus on full-case technology, our framework and sub-frameworks have a generic appeal. Our approach considers different types of automation modules suitable for different categories of product demands (e.g., slow vs. fast-moving SKUs and single-SKU pallets vs. mixed-SKU pallets). Since we do not find such a study that links DC design to network cost strategy given the MC context and full-case technology, we believe that the contributions would be valuable for theory and practice. We believe the framework and sub-frameworks are general enough that with some/little modification, these could be adapted for different technologies. We illustrate our framework through a detailed case study (Table 2 to Table 5 present various computations involved in the decision-making framework) (see Fig. 3a, Fig. 3ba,b).

Fig. 1a.

The Three-Part-Framework for evaluating investment in automated DCs

Note:

1. Each of the three parts of the framework have been discussed separately under three sub-frameworks: Refer Fig. 2a, Fig. 2ba and b (1 in above figure), 3a and 3b (2 in above figure) and Fig. 4a, Fig. 4ba and b (3 in above figure)

2. Importance of positive impact of automation on stores is highlighted, given MC setting.

Fig. 1b.

Operationalization of the Three-Part framework (1, 2, 3: refer to respective sub-frameworks and corresponding models in E-component)

Note: For operationalization, we collect data, develop analytical models, analyze the primary and secondary data, and illustrate the results.

Fig. 2a.

Three areas automated warehouse/DC evaluation: strategic, design, and operations evaluation

Explanatory Notes for automated DC design/modeling sub-framework:

FL-A: forklift assisted; Car-A: car assisted; R–A: Robot-assisted; M/S/A: Manual/Semi-automated/Automated

1. Evaluation at strategy level leads to setting MC-alignment, design and operation performance goals

2. Design level evaluation involves technology selection to meet strategic goals, various equipment sizing, DC layout (finally leading to buffer sizing) and storage/sequencing/picking policies for each module. Comparison of this to performance goals would lead to identification performance gap.

3. Measurement of performance parameters of various modules leads to performance evaluation against set goals.

Fig. 2b.

Aggregate-level steps in DC sizing analysis

Explanatory Notes:

1. Demand analysis involves analysis of mean, peak, and seasonal demands for each SKU

2. Post business strategy, supply chain strategy, and technology selection (Fig. 2a), retailer must decide upon extent of automation (and hence automation module selection must be completed).

3. Demand analysis, business and supply chain level strategies and the decision on extent of automation (given a technology) would lead to consideration of different scenarios that need to be considered before decision on investment and location of DC is made.

4. Various computations, as indicated in different sections of the paper, are then made.

5. The iterative process would lead to optimal investment decision.

Fig. 3a.

Elements of decision-making framework and link to Network Analysis sub-framework.

Fig. 3b.

Simplified outline for distribution and backhaul routines.

Fig. 4a.

Sub-Framework and 5-Step process for sustainable supplier engagement/management

Explanatory Notes:

1. Step 1 allows for strategic supplier selection, building (joint planning) of relationship over time, data collection, and for assimilation of feedback.

2. Step 2 allows for cost-based analysis, based on data collection, assessment of rewards/penalties, and contract policies. Risk assessment is ex-post quantity allocation (i.e., a post-hoc analysis)– based on percentage of a part ordered from a set of suppliers. NQCI is an index that describes such risk (potential for failure). Step results in allocation of a product portfolio to a subset of identified suppliers.

3. Step 3 is feasible when relationship allows for further data sharing/collection so that benefits (offered to the buyer) and associated risks could be assessed resulting in modified allocations in Step 2.

4. While Step 4 ensure that delivered goods and services and assessed against contract terms, Step 5 identifies the gaps between current/future needs and current delivery. This gap then guide future joint-action.

Fig. 4b.

Outline of type of data and modeling

Step-2: Referring to Fig. 4a, Step-2 involves engagement with supplier at a non-strategic level. Information exchange is more generic in nature. Most buyer-supplier relations would fall into this category. But for sustainable buyer-supplier relationship and for safeguarding it is preferable to have strategic buyer-supplier relationship. Over time, even such relationship could become sustainable as business exchange would provide first-hand cost data.

Step-3: Referring to Fig. 4b, Step-3 involves strategic buyer-supplier relationship. Information exchange is more specific and at a deeper level. Supplier costs, along with benefits and risks could be better defined.

Table 2a.

Scenarios for 1-DC network configuration and combination of three modules.

Explanatory Notes:1. Four scenarios have been described, each with a different distribution of peak demand over the Overflow and RCHS; the overflow is then distributed over IRCHS, and LCSH categories2. RCHS could be designed to a specific level of peak demand and excess would be shifted to LCHS or IRCHS.

Table 5A.

Payback period computations for Scenario 1.

DC (or warehouse) automation includes automation of the processes involved in receipt, storage, picking (to customer order), and packaging of materials in ready-to-ship conditions [23]. These include automated storage and retrieval systems (AS/RS), vertical lifts, carousels, conveyors, automated guided vehicles (AGVs: e.g., Kiva Systems used by Amazon), mini-load systems, picking robots/devices, and picking technologies [24]. We find different degrees of DC automation (the extent to which processes are automated) in practice. Manual DCs (all processes are manually handled) typically handle pallets. Further, an automated DC may handle only pallets or a mix of pallets, cases, or ‘eaches’ (single items from a case). In addition to the pallets-only case, we discuss mixed pallets as a) the latter case is representative of a general case since it includes a broader set of processes and b) proven technologies are now available to handle a mix of pallets (same items), cases (a mix of different cases), sub-cases (less-than-case: i.e., a few items picked from a case), and pallets. From an effective transportation perspective, an automated DC can handle a single-item pallet or mixed pallets, it can help include ‘last-minute’ information can be included – eliminating the element of uncertainty (who needs what and when in a network), improving the cross-docking operations in the network, and also facilitating supply (through backhaul). We develop a network analysis module based on such information. We also consider SKU velocity (slow vs. fast-moving SKUs) and the DC modules are selected to handle these. Thus, a pallet to be delivered to a store would contain exactly what is needed, thus keeping the store and network inventory low. Further, automation helps deliver ‘aisle-friendly’ pallets to stores. Such pallets help reduce handling and damage costs at the store level. Thus, automation has a network-wide impact, as discussed in this paper.

We present the models for the three sub-frameworks: A) the first approach for DC-level analysis employs a) discrete event modeling for the automated and semi-automated DCs -Fig. 2a (equipment capacity models for developing characteristic curves for different equipment and DC process model), b) mathematical programming based optimization model for evaluating aisle height vs aisle length for different DC automated modules, c) economic models for i) labor, ii) ROI computations, and iii) labor vs capital equipment tradeoff computations; B) the second approach for network-level analysis employs three mathematical programming-based models for analyzing the network – Fig. 2b: a) distribution, b) supply/backhaul, c) cross-dock; and C) the third approach focuses on safeguarding network-level assets and discusses three economic modeling and mathematical programming models sustainable supplier engagement and management.

In this paper, our goal is not to discuss the modeling details (we provide representative modes and results in E-component for interested readers), but to provide an approach to modeling in the context of the sub-frameworks presented to inform the decision-making for investment in automated DCs.

We identify and address the following research questions: RQ1: What are the important elements of an integrative framework for decision-making for an automated DC, given an MC setting? To address this research question, we participated in many meetings/discussions with academia and industry (through formal/informal meetings, conferences, field visits) to understand the big picture, key drivers, and important issues facing scholars and industry – which helped us in identifying important elements of the framework. RQ2: How can the higher-level framework be linked to the DC level, network level (distribution and supply analysis), and supplier engagement sub-frameworks? To address the question, we dived deeper into each key element of the framework and identified the related sub-frameworks based on academic literature. RQ3: How can we operationalize such a framework (using analytical models, tools, and guidelines) so that a manager would be able to make an informed investment decision? To address this question, we developed analytical tools by employing the literature and by developing innovative techniques to solve the real-world issues in real time.

In light of the gaps identified, we contribute in several ways: a) to the DC automation literature by developing a three-part investment decision-making framework for automated warehouses (a general framework with a particular focus on ‘full-case’ technology6) that extends and integrates similar contributions (theoretical) of other scholars -given the MC context, b) by documenting voice of industry (retailer, stores, and technology supplier) and using the voice of customers to develop analytical models and for validation, c) by developing three supporting sub-frameworks, d) by developing mathematical programming-based, economic theory-based, and discrete event modeling based models – that have been validated (in reference to work of other scholars and industry conference presentations), e) by developing tools that could be rapidly deployed by a retailer for DC automation investment analysis, f) by illustrating and validating the framework with a real-life example – by deploying it with a retailer, and g) by providing a generalized framework for OC value chain that identifies sustainable supplier relationship management as an integrated (and missing) element (Table 1D) and h) by identifying future research directions based on our research work.

Table 1D.

Framework oriented review papers highlighting the important components in multichannel/omnichannel value chain.

The rest of the paper is organized as follows. We first provide a literature review of the relevant topics. The need for a decision-making framework along with the requisite models and tools, from a research and practice perspective, is described (based on literature review and field inputs). This is followed by a description of the approach to data collection (primary and secondary data were collected). We then describe how the decision-making framework is operationalized by employing sub-frameworks. These are followed by a summary of the strategic issues, DC design and sizing model, network cost analysis model, and how this information is integrated into the payback period computations. We then illustrate the framework for 1-DC and 2-DC networks through a case study. This is followed by a discussion/conclusion of our findings, potential future work and limitations of our work. Next, we provide a review of the relevant literature and how it supports our framework (Fig. 1a).

2. Literature review

We present a literature review for important elements of automated distribution center evaluation and analysis in retail supply and distribution networks context, given OC/MC setting (other similar e-commerce notions include ‘online market place’, ‘online retail’, ‘internet retail’, ‘electronic platform’, ‘e-tail’, especially from a customer perspective where difference between an online or an offline channel is blurring/disappearing [3]). Although we do not find framework papers that deal with decision-making for automated DC investment and sustainable management in OC context, we discuss review articles that highlight key aspects/variable essential for the OC value chain (Table 1D) and those articles that are relevant to respective sub-frameworks. We divide the literature review into four sections: a) Strategic issues and the need for a systematic analysis for the evaluation of an automated DC, b) DC design and sizing using simulation, c) Network analysis for evaluating the transportation and inventory DC costs, and d) Supplier contract analysis for ensuring a sustainable supply-chain and safeguarding investments in network assets (such as an automated DC). For brevity, we focus on essential and relevant elements the literature. While the role of IT infrastructure is critical for achieving channel integration, we are unable to discuss this important facet due to space constraints. Our framework (Fig. 1a) is based on the literature and data collection from the field (primary and secondary data).

2.1. Strategic issues relevant to the evaluation of an automated DC in MC setting

Although investment in warehouse automation has increased [26] over the years, scholars have emphasized the need for research articles for a systematic and scientific evaluation of an overall design process for distribution centers ([26], [27]; [18], p. 515; [28], p. 3; [27], p. 285). We find very limited number of journal articles that discuss strategic issues and their impact on the decision-making process for investment in an automated warehouse. For example [17,29], focus their study on within-DC operations strategies (storage and pick), while [18] discuss the need for evaluation at the strategic level. In this paper, we highlight the importance of analysis at different levels (strategic/contracts, DC design/layout, network, and stores) and how it relates to investment decision-making thus extending the work of scholars working in the area and fill this gap in the literature.

To evaluate the strategic issues for automated warehouses [30], conducted interviews of six European organizations and found that agility (rapid and flexible response to changes) is important for managing supply networks and that distribution centers must possess three types of agilities: a) time variances (urgent orders), b) volume variance (changes in demand due to seasonality, promotions, product end of life, and end-consumer demand), and c) quantity size variances (less-than-case-sized or a few-cases-at-a-time shipments). More recently [20], highlight the importance of DC/warehousing functions that are relevant in an e-commerce era – providing an MC perspective. Important features such as a) small order sizes, b) large assortments, c) tight delivery schedules, d) varying workloads have been mentioned. The importance of long-tail items (especially for the internet buyers) is pointed out by scholars (for example, [31]), on the other hand, proximity to the consumers (which is afforded by BM stores) is also considered an important factor [14]. Given the OC/MC setting, we see the need for important tradeoffs that a retailer must consider in her decision calculus: a) while OC/MC provides a better mix to the consumers, this may lead to channel conflicts [32]; b) while e-fulfillment is important for online customers (physically reaching the ordered goods to the consumer), it is critical and expensive operationally [14] c) location of delivery and return points [15,33] at: warehouse/s (which interfaces with suppliers and stores received goods), DCs (distributes goods received from warehouse – not focused on storage as much as a warehouse), cross-docking points (only facilitates transfer of goods between vehicles), and stores (that serve as pick-up/return nodes for consumers -largely focused on display/sales/marketing function); d) the extent of automation (manual, semi-automatic, automatic modules), e) developing a supply network that is not only responsive but also works effectively on the three sustainability dimensions – environmental, social, and economic aspects.

While we find that variances (in time, quantity, size) are important considerations, we find that the DC automation addresses these well. For example, due to automation, the DC is able to respond to changes in needed quantity and is able to do so in short notice. In addition, a mixed pallet can accommodate a variety of sizes. Further, other distribution center design and sizing related issues noted from engagement/interviews with warehouse management include: a) inventory costs due to supplier lead times, b) peak demand, c) seasonality, d) the ability to address demand growth, e) ability to provide ‘aisle friendly’ pallets (for efficient shelving), and f) time variance of items received in stores or distribution centers. The importance of stores in the network, given MC setting, has been highlighted by several scholars [1,[4], [5], [6], [9]]. We specifically connect stores to DC automation by presenting the impact of automation on store operations. From a management perspective we uncovered that important issues include: convincing suppliers to adapt to automation needs, network coordination, managing promotions so that operations could support the change in demand. We have tabulated our experience regarding these key factors (Table 1A1, Table 1A2, Table 1BB). We also address these strategic, tactical, and tradeoff related issues mentioned above in the three sub-frameworks. These issues affect warehouse and network design and sustainable supplier relationship management and contracts.

Table 1A1.

Approach to primary and secondary data collection.

Table 1A2.

Approach to secondary data collection.

Table 1B.

Learning from interviews: Specific features of conventional and automated DCs.

2.2. DC design and sizing using simulation

We focus on the literature that addresses system-level issues related to DC automation. Well-known scholars in the field [17,21,29,34] have reviewed important issues associated with automated DCs [17]. mention that there is a need in the field to consider demand characteristics and to validate of cost models.7 More recently [20,[8], [35], [36], [37]], highlight the importance of DC/warehousing functions that are relevant in the e-commerce era. Important features, especially from an OC/MC context, such as a) small order sizes, b) large assortments (including long tail/slow moving items), c) tight delivery schedules, d) varying workloads (ability to meet uncertain demand) have been mentioned [38]. provide a review of how simulation approaches have been employed by scholars in the field [39,40]. presented the simulation and design-of-experiments based approach to identify the factors affecting the performance of DC systems. Only a few scholars discuss systems-level analysis while [38] discuss a discrete simulation-based study to compare storage assignment policies. While [41] provides a review of the high-level design of DC [42,43], discuss department functional details of a DC. We find that scheduling, dwelling, and storage issues and policies have been addressed by several scholars. For example [44], presents a simulation-based study for comparing dwell-point policies. These largely have been limited to single-aisle cases (with a few exceptions). We do not find an integration of simulation-based (or other analytical evaluation) with end-to-end DC operations into a decision-making framework. We attempt to address this gap in the literature.

Although a DC itself may be well-designed, the rest of the network operations also must be effective. Hence, we discuss the literature in the area of network analysis next. Network design (number and location of DCs in the network) is considered an important issue in supply and distribution networks (location of DCs and cross-docks in the network: [45,46]. We refer to the network cost analysis literature, as it is an essential element in estimating network costs (a key element in financial analysis). We assume a given network design.

2.3. Network analysis: inventory routing and cross-docking related costs

We focus on the network analysis literature in the OC/MC retail context [[47], [48], [49]] for a) large-scale multi-item inventory routing problem, b) considering integrated backhaul and line haul, and c) considering cross-docks in the network. We include perishable products for distribution [50] as same warehouses may carry them along with regular products. We specifically refer to a research gap: articles that consider automation in this context (impact of pallet-based analysis, which relates to manual/conventional DC operations vs. case-based analysis which relates to automated DC operations). We assume a two-week horizon and meet the demand requested by the stores (especially slow-moving items), given a few days’ notice. Thus, we avoid complicated forecasting techniques [51,52,53] which helps in implementation stage. Inventory routing [[54], [55], [56], [57]] is relevant to our problem at hand. We refer to earlier works [58,59] that consider a single item delivered to multiple stores and into fixed partitions (a few stores in a given area that can be on a delivery route [60,61], in the retail context, and allow for direct deliveries into fixed partitions [62]. discusses fixed and variable transportation costs but does not consider fixed partitions while [63] analyzes delivery patterns in t the retail context using IRP [inventory routing problem]. While scholars have investigated distribution in retail context [[64], [65], [66]], a joint supply-distribution problem is seldom considered. The backhaul problem also has been discussed in early retail literature [59,[67], [68], [69]] where authors consider different approaches to backhaul. For example [69], employs a greedy heuristic to solve the multi-terminal backhaul problem where a returning vehicle could consider visiting one of the other terminals. We also consider backhaul routes, similar to the notion of peddling [67,68] in a zone/store cluster. We consider the EOQ approach (first stage) that is similar to that employed by Refs. [[70], [71], [72]]. But we do not consider the notion of an aggregate product. While [73,74] assume fixed frequencies, we assume that the deliveries need to be made within a few (2–3 days) of order receipt. Further, unlike scholars who have employed trucks that visit all stores in a cluster [61,75], we visit all or fewer stores. We share similarities with [75] who divide the problem into three stages. We find a few scholars solving an integrated distribution-backhaul problem [76] for a large network in a retail context with the goal of integrating the analysis into a decision-making framework.

We do not find studies referring to DC automation and corresponding issues in cross-docking. We refer to pre-distribution (outbound) and post-distribution (inbound) cross-docks [77] discussed in the literature. The cross-docking problem has been addressed more recently in the literature [78,79]. Review articles discuss network level (CD-location, CD-networks, and CD-vehicle routing [78] and operational level issues (CD-layout, dock door assignment, and truck scheduling [see [[78], [79]]]. We specifically refer to articles that discuss cross-dock location [[80], [81], [82]] and direct vs. shipment through cross-dock [83].

2.4. Sustainable supplier engagement: safeguarding network assets (e.g., an automated DC)

Sustainable supplier-buyer network contracts [25] have recently become a topic of interest and are considered important to the success of all network members [76,84,85]. Several scholars [86,87] have discussed single sourcing and consider it risky. But, Eastern cultures, that are largely characterized by relationship (for example, [88,89]) focus, are likely to form long-term oriented (and likely, single-source) relationships due to high relationship trust factor. Multiple-sourcing supplier management strategies have been discussed by scholars, e.g., pricing issues and capacitated suppliers [90,91,92]. More recently, sustainability principles have been considered important in a multi-sourcing quantity allocation decision [93]. We also consider this as a critical angle in safeguarding the network investments such as that in an automated DC/s. Our formulation allows us to consider either a single-sourcing and/or multiple-sourcing situation ([94], Table 1). Further, we find that few scholars [95] employ math programming (MP, MIP- mixed integer programming) techniques [96, Fig. 1; 97, 98, 99] for developing a framework for SSCM. Other scholars develop closed-form analytical relationships relating to SSCM contracts [[100], [101], [102], [103]]. We develop a constrained optimization-based technique which employs a piece-wise linear formulation (fixed cost and transportation cost) to estimate total costs. We not only capture subjective inputs but our work also differs from the multi-criteria decision making (Fig. 1, [95,96,104]) approach by capturing subjective inputs. Further, we consider the risk characteristics of individual decision-makers involved. This is critical for quantity allocation, given the combination of cost, benefit, and risk profile of suppliers.

3. The need for a decision-making framework, associated models, and tools

As highlighted by literature review by eminent scholars, OC/MC is a nascent field and value chain driven and integrative literature reviews are lacking [[3], [10], [11], [13], [15]], although that on specific areas exist [3], as indicated in Table 1D. Hence it is not a surprise that there is also a need for a decision-making framework for investment in DC automation driven by industry requirements and as indicated by the above review of academic literature [10,16,33,105]: a) highly cost-competitive business environment and hence the need for efficiency of operation is important (important to consider suitable tradeoff between labor and capital, reduced picking and packing times, higher throughput), b) flexibility in warehouse operations (ability to handle pallets, cases, and subcases) and delivery (optimally filling the truck for delivery to one or multiple stores), c) ability to handle slow (‘long tail’) as wells as fast moving items, e) lower response time and higher delivery frequencies, f) ability to return goods (using backhaul through supplier locations and vehicle routing through multiple stores to collect return products – [14] identify this need on page 350), g) ability cross-dock to facilitate material handling for distribution, supply, and return, h) awareness and preference on the part of the consumer to buy from retailers that promote sustainable business practices (care about environmental and social aspects).

Based on the above review of literature and our experience from the field, we propose the 3-part framework and the related sub-frameworks (Fig. 1a, Fig. 1ba and b) so that managers would find a stronger basis for investment decision-making and scholars would be inspired to guide their research by employing an integrated approach so that further insights could be obtained to guide such investment decision-making.

Our framework addresses the needs of a generic DC: it includes fully automated, partially automated, and manual modules. We identify and address several research gaps. First, we find few scholars addressing the issue of a generic framework for investment in DC automation. Second, DC automation not only impacts the DC but also the distribution and supply networks, but we do not find an integrative study from a framework perspective (including safeguarding of investment in high-value assets such as an automated DC/warehouse). The need for such a framework and the research gap has been identified by several scholars [[[41], [106]]; [18], p. 515; [28], p. 3). Although scholars have addressed DC automation at the DC level [17,23], a decision-making framework with a more holistic focus is missing. Thus, we extend the work discussed in the extant DC automation literature but include the network-wide impact of automation and consider four levels of analysis (strategic level, DC level, network level, and store level). Third, we develop sub-frameworks a) for guiding the design and sizing at the DC level, with a focus on MC retailing, b) for guiding distribution-supply network modeling (with a focus on linehaul and backhaul to facilitate returns), and c) for guiding sustainable supplier engagement and management so that the specific asset (automated DC) is well utilized. This is critical since supplier failure would increase uncertainty in the network, increase retailer coordination costs, decrease service levels at the stores (customers), and invite litigation (if the supplier does not follow sustainability-driven production practices). Fourth, we provide models for the three sub-frameworks – thus providing an analytical basis for the sub-frameworks that support the high-level 3-part framework. Fifth, we present business cases to illustrate the application of the framework to different network configurations.

In practice, we find that the DC automation equipment manufacturers make the case for automation based largely on labor savings as they do not have access to the retailer's strategic (or tactical/operations) information. Without guidelines to envision relevant scenarios and tools that enable the above-mentioned analysis, the retailer would not be able to assess the payback period/return on investment for an automated DC. Our integrated approach, based on the three sub-frameworks (interested readers are referred to the E-component) and corresponding analytical models, helps in creating a comprehensive framework for evaluating the investment opportunity.(see. Table 2a, Table 2ba,2b)

Table 2b.

Sizing (number of equipment) and investment computations -Example (Scenario 1 only).

Explanatory Notes:1. One of the four scenarios have been described in the table.2. Corresponding number of modules and cost of each module has been listed.3. Total cost of automation equipment is then computed.

4. Approach to data collection for analysis

Our goal in the project was to understand key components of the framework (and sub-frameworks) and model the cost elements involved over the lifetime of an asset [[107], Fig. 5] such as an automated DC. In addition, to ensure that the substantial investment in an automated DC is safeguarded, we investigate and propose a suitable sub-framework. Although we present ‘full-case’ technology, which has growing acceptance, we benchmark other technologies for generalization. We consider the case where the contract is build-to-operate and includes service contracts over the life of the assets. This is driven by the fact that the technology is complex. We interviewed a DC technology provider and obtained primary data (quotes for different configurations). We benchmark automated DCs (six DCs working on the full-case technology – and obtain published trade reports, interview transcripts, and interview practitioners in RILA conference). Further we visited one operating automated DC that employs the technology (four individuals made observations on the floor for 8 h) and obtained key installation/operation information from the retailer and the DC technology provider. Similarly, we obtained data on conventional DCs (investment, maintenance, operations, and salvage). We also collect primary data at stores (customers). The summary of the data collection approach is depicted in Table 1A1 and Table 1A2. The learnings are summarized in Table 1B (not relevant for [25]). Further, we consider environmental costs (secondary data used for finding the total cost of transportation in the supplier selection and quantity allocation [108]. We also show the approach to data collection on the store floor (Figure_Table 3) and corresponding analysis (data and analysis not relevant for [25]) for estimation value offered by automation to a store (Table 3). Further, data was collected on the distribution center floor (Fig. 4a, Fig. 4b) and analyzed to estimate the cost of LCH and IRCH modules (data and analysis not relevant for [25]). The analysis is useful in estimating the tradeoff between automation and manual operations for a module. Estimates for the cost of hardware for an automated DC and conventional DC were obtained for a very reputable automated DC OEM (equipment manufacturer M1) who we worked with (data and analysis not relevant for [25]).

Fig. 5a.

Cost/case for different investment levels (target demand levels) for 3-module automated DC

Explanatory Notes

1. Three scenarios that allow for meeting same demand quantity (160k cases/day) have been considered

2. Full automation (RCHS only) offers best solution at design point but is expensive if demand falls.

3. Other two options are less sensitive either way.

Table 3.

Metrics for current and future state for pallet breakdown and shelving operations.

Explanatory Notes: (Also see Figure_Table3 below)1. Anticipated distance/time between unloading dock and staging would increase post automation as aisle-friendly pallets are delivered to the stores (customer). But the savings from average time for moving between staging area and to shelves advantage is more significant. While the distance between unloading area and the staging area is not favorable, but the distance travelled per trip is significantly reduced.2. Thus, significant savings result from automation – as see from the customer (store) perspective.Figure_Table3: Flow chart for process and for finding number of touches that pallets received at a store.

Our approach to data collection is guided by a) existing literature: journal articles and trade literature b) interviews (Table 1A1 and Table 1A2), c) Field observations: in stores and in DCs (automated and manual) (for example, Figure_Table 3 describes our approach to collecting data on store floor), d) Analysis (for example, simulation of DC operations, network analysis – supply and distribution, analysis of store operations – Table 3), and e) regular interaction with CXO level personnel over three years – where we shared our analysis and improved upon our understand based on their feedback. Data collection and analysis were key to developing the framework (Fig. 1a) and corresponding sub-frameworks. While we are unable to share the primary data directly (qualitative and quantitative) due to non-disclosure agreement (NDA), our approach to analysis and inputs from the industry (similar publicly available inputs from other retailers) is shared. We also validate our work by comparing the results of our analysis with industry practice (published warehouse specifications, design, layout, etc.) and publicly shared data. Further validation of our analysis comes from regular meetings with retailer R1, other automated and conventional DCs (R2-R4 and M1).

First, we would like to state (for transparency reasons) that the extensive primary and secondary data we compiled from multiple sources not only aided in the development of the framework described in this paper, but also served to motivate the research addressed in Ref. [25]. Specifically, primary data from one retailer and semi-structured interviews with retail industry professionals were used in this paper and in Ref. [25]. The aim of this paper is to develop an analytical decision-making framework for investment in automated DCs while that of [25] is the discuss the theoretical framework/theory for coordinating supply/distribution networks. Thus, the two papers have different goals requiring different types of analyses, and have been reported as two separate papers in order to maintain coherence with their respective research objective. We provide further details in Table 1C and explanatory notes that follow.

Table 1C.

Table highlights the similarity and difference between how the data was employed in this paper and in.Ref.[25]

Notes: First, we would like to state (for transparency reasons) this extensive data (primary and secondary data from a variety of sources) collected over time (as stated above) not only help us understand issues related specifically to this paper (the three-part framework), but also served to motivate us to think about topics related to Ref. [25]. Thus, while some parts of the primary data (Item 1 and 7 Table 1A1) directly influenced Section 6.6 of this paper, they also motivated us to go beyond (thus laying the foundation for the need of [25]). Similarly, some parts of the secondary data (item 11, Table 1A1 and item 8, Table 1A2) are more directly linked to Ref. [25] but also served to motivate this paper and confirmed our belief in the philosophical importance of Section 6.6. We refer to Table 1C for specific differences and similarities between how the data was employed in this paper and [25]. The purpose of this paper is to discuss the overall frameworks and the supporting analytical frameworks for investment in automated DCs while that of [25] is the discuss the theoretical framework/theory for coordinating supply/distribution networks. Thus, the two papers have very different goals and require different types of analyses. The analytical approach for network coordination described in Section 6.6 in this paper could also be applied to help coordinate the networks discussed in Ref. [25]. Thus, the tools discussed in Section 6.6 would help implement (at least partly) the theory developed in Ref. [25].

Primary data was collected, in part, over time during our meetings with CXOs (five CXOs, over 2 years) of retailer R1 and through two graduate internships (CXO is a general term for a senior manager – for example, a CEO [Chief Executive Officer] or others such as CIO/CMO/CFO/COO [Chief Information/Marketing/Financial/Operations Officer - respectively). These were in terms of meeting notes, comments on our presentations and models, and our validation of such inputs from other meetings and secondary data sources (such as retail conferences, trade publications, and reports describing the implementation of warehouse automation). In addition, interviews (semi-structured and unstructured – on regular intervals, thus building on previous knowledge) with automated (R2, R3, R4) managers, employees working in the DC) and conventional/semi-automated DCs (with retailer R1 managers and employees working in the DCs) were conducted to understand the ground realities, tactics, and strategies. These locations were selected for us by the CEO of R1 as they were most relevant to our study. We also selected an automated DC based on mutual interest for on-site study and interviews with senior managers (a few DC owners did not want to expose their DC operations due to privacy reasons and since it takes several hours of their time to engage with us – we had request full day engagement). We requested (and obtained) inputs and engagement from CXOs/managers (retailers, DCs, equipment manufacturers) who have very relevant experience. Further primary data was collected (in terms of how an automated DC works, key equipment and their functions, and associated cost elements) by discussions with automated DC equipment manufacturer M1 (a CEO and a few senior managers/managers) over a two-year period. We refer the interested readers to sample questions (a part of much larger set of queries/questions for which responses were collected) (Appendix 1) that were asked in semi-structured or unstructured manner. These are largely objective in nature and designed to understand the technical aspects which, along with inputs from other sources (secondary data such as published interviews, trade journal articles reporting case studies and trends in industry, and research articles), form the basis for the decision-making framework discussed in this paper. We refer to Table 1A1 (items 1–7) for type of respondents involved in this data collection effort. While this part of the data is directly relevant to this paper, it served as a motivation and background material for [25], which has a theoretical focus and employs the learning as a context/background but goes beyond the technical aspects discussed in this paper. Further, we point out that some of the primary data collected, while directly relevant for this paper, had some relevance for [25]. For example, of the sample questions listed in Appendix 1, ‘D. Sustainable supply network contract related questions’ are more relevant and motivational for [25]. Thus, the current study, specifically Section 6.6 (with a focus on sustainable networks), has served to motivate (philosophically) the key ideas of [25].

We also collect secondary data from secondary CXO interviews of warehouse automation providers (M1, M2, and M3). We continuously applied this knowledge in building models and validating them through secondary data available through white papers and trade journal reports. While we cannot share the primary data collected (a mix of semi-structured and unstructured informal interviews which were not recorded but key points and learnings noted and verified), we validated the data through secondary data collected (Retailer R5-R10). We share excerpts of the secondary data (Table 1A2). Many of the quotes (we provide quotes from thirteen CXOs) from secondary sources (that we share) second our learning from our discussions with retailer CXOs (primary data). We again point to the fact that sample secondary data (a part of the whole) is presented in Table 1A2 and some of it is relevant for [25]. We provide in Appendix 1 further details of similarities and differences in how the secondary data, which was collected over two years, was employed differently for this paper and [25]. For example, Table 1A2, item 8, is also relevant for [25].

Further, we collected data from the stores (R1S1, R1S2, R1S3) – observing receipt and handling/shelving processes (these are specific to this paper and no bearing on [25]). We also combined data from disparate sources (for example data from the Alibaba website8) to inform our models thus reducing our dependence on any single data source. Several suppliers were contacted to obtain data from this website and primary (through quotes) and secondary data were collected to understand cost curves (price vs. scale) for the supply side. On the distribution side, while we obtained data from R1 on various costs (distribution cost per mile, labor costs – fixed and variable costs for different functions, cost of material handling at important points in the network, carbon cost for transportation through various modes) we verified this data through secondary sources and employed representative costs for models. Similarly, we obtained production costs (equipment cost, cost/case for processing different extents of automation) from R1 and other sources and employed a range for sensitivity analysis. We also shared the results of our analysis at conferences (modified data was employed). In addition to above discussion on how the data, collected over time, was deployed to largely support this paper but also motivate and support [25] we also note here that [25] employs an additional survey to validate the propositions.

For this paper, we did not need any ethics approval for data collection as interviews conducted were with business managers (mostly at senior levels [20+ yrs experience with the retailer] hence above 18 yrs), primary data was collected from quotes, and secondary data was collected from supplier websites. Since our research reports (and articles for publications) employed our own contributions (models/tools that we developed – which did not exist before) and data from secondary sources (which helped us validate our models), our retailer and warehouse equipment partners did not object to our research publications.

Next, we provide further details regarding the framework and its connection to sub-frameworks. We first outline the approach to analysis. The three-pronged approach is at the core of the decision-making framework (Fig. 1a) and its operationalization (Fig. 1b). The importance of strategic analysis (also, Fig. 2a) and the iterative nature of decision-making, given different scenarios, is highlighted. We first assume a first-cut network design (Item 2, Fig. 1b) followed by network analysis (different types of costs are shown in Fig. 3a) -assuming an automated DC serves the network. The type of analysis (network optimization) with some details is shown in Section 2.0 in the E-component. We then develop a mechanism to size the automated DC (Item 1, Fig. 1b) given a variety of inputs (Fig. 2a, Fig. 2ba and b) including strategic inputs, demand distributions for different SKU types, the extent of automation, and the type of automation technology, equipment type and their numbers, initial DC layout, etc. The type of analysis (discrete-event-based simulation) and results are shared in Section 1 in the E-component. The third leg of the framework consists of math programming (optimization) and economic modeling-based analysis to inform sustainable buyer-supplier contracts. The outline of the approach to analysis is shown in Fig. 4a, Fig. 4ba and b and the type of analysis (with some details) is shown in the E-component (Section 3.0). Other analytical tools/approaches were developed (for example, scenario analysis -Table 2a, Table 2ba and Table 2b) for store-level analysis – Table 3 and Figure Table 3, DC-level analysis -Fig. 5a, Fig. 5ba and 5b, etc.) to provide inputs to investment analysis (Table 5A, Table 5Ba and 5b).

Fig. 5b.

Observations and retailer data based computations for estimating labor costs for LCH module (manual interface)

Notes:

1. The computations are helpful in estimating the tradeoff between automated RCH and LCH with manual interface

2. The computations also help in evaluating combination of automated and semi-automated modules for designing suitable solutions, given characteristics of a DC.

Table 5B.

Payback period for different scenarios.

Explanatory Notes:1. Table 5A describes to payback period computations2. Investment costs (based sizing and design of automated DC), transportation costs (based on network analysis), and savings (due to improvement in store operations, DC labor savings, etc.) have been included.3. Table 5B describes the payback computations for different scenarios.4. Thus, our work provides tools for comparison of different extent of DC automation.

5. Operationalizing the framework: inter-relationships between sub-frameworks

DC automation can deliver enhanced capabilities that impact not only the DC operations, but also the supplier (upstream), and store (downstream) operations. The enhanced capabilities and performance of these systems may require a redesign of the entire distribution network, major changes to the existing supplier engagement, supply chain management (including procurement, inventory management, and transportation) and contracting policies, and changes to the existing standard store operations. Our framework and sub-frameworks (Fig. 1a, Fig. 1ba and b) are derived from the above literature review and our interactions with the personnel in the field.

We depict the three key elements of a high-level, three-part decision-making (Fig. 1a illustrates the three-part framework) framework and important steps in operationalizing the framework (highlighting the iterative nature of the framework in Fig. 1b) in Fig. 1. The first deals with designing and sizing the DC. The second part focuses on network costs (transportation, inventory, fixed and variable labor costs, and overheads). The third part focuses on supplier engagement and management with the goal of ensuring that the asset (automated DC) is well utilized. For operationalizing the framework, we consider five levels of analysis: i) Level 1-strategic level: A retailer would consider the macro-level inputs first: business strategy (economic growth plans, state of competition, social aspects, environmental aspects, etc.) and other key internal factors (such as labor contracts, competitive advantage, etc.); ii) Level 2-network level: Next, the retailer would decide on the supply/distribution network level design (number of automated DCs and cross-docks, their location in the network, and policy distribution/supply/backhaul) and would complete demand analysis. The retailer then would create different scenarios of interest that must be evaluated. For example, given the network structure, the retailer must decide which items to stock at which locations, inventory policies at the DCs and stores, supplier allocation to the DCs, and how to utilize the cross-docks. Given appropriate data, the retailer then would be able to perform network analysis for assessing various costs. iii) Level 3 – DC level: The retailer would consider DC level issues here: the extent of automation, DC design and sizing, including alternative sizing strategies (evaluating a combination of different modules to handle a demand scenario), and may need to use rapid sizing tools to analyze equipment needed for a given scenario (we consider three types of items: regular, irregular, and less-than case); iv) Level 4 -Store level: Next, the retailer would consider customer-level ((store)-level) inputs and evaluate the impact of DC automation; v) Level 5- Strategic network assets: Finally, the retailer would develop a sustainable supplier engagement and supply chain management strategy to ensure that the investment is well utilized, given global supplier networks (supplier selection, keeping needs of an automated DCs in mind, deciding on transportation modes, a mechanism to address/manage costs, benefits, and risks).

To enable the business case evaluation, we develop worksheet-based tools that help integrate costs and benefits (from the above four levels of analysis) into the payback period computations. For manual (conventional) DCs, we assume a pallet as a unit of analysis, while for an automated DC, we assume a case as a unit of analysis. By simultaneously considering the opportunities and benefits that DC automation can provide at all the five levels, the retailer can gain insight into the potential synergies and trade-offs that can maximize the return on investment in automation. Further, the retailer would be able to leverage the knowledge and insight gained through the analysis for writing contracts with automated DC equipment manufacturers and sustainable supplier contracts for (items sold by the retailer) to safeguard the significant investments not only in the DC but also in the network (Level 5 analysis). Since we find that retailers are keen on introducing store brands and are getting directly involved in manufacturing and supply management, sustainable contracts [109] have become more relevant.

Next, we discuss the key elements of the framework and how they come together to help make ROI computations. We also outline the methodology involved in completing the analysis for each of the key elements.

6. Key elements of the framework and the relationship to payback period computation (decision-analysis tool)

This section provides an overview of the key elements involved in the higher-level framework (Fig. 1). These involve reviews of a) strategic issues, b) DC sizing and design-related sub-framework and corresponding modeling tools, c) network costs involved in distribution (including back-haul and cross-docking costs) and corresponding modeling, d) analysis of store operations, e) sustainable supplier management to ensure that the significant investment in network asset (the automated DC) is safeguarded. We discuss the sub-framework and corresponding economic and mathematical models.

6.1. Strategic issues relevant to investment decision-making

A retailer needs to consider strategic issues that are relevant to making an investment in an automated DC. We find very few studies discussing the topic [41,30,110,109]. While authors [105] consider the strategic issues regarding decision-making, they do not discuss automation. On the other hand, a few studies [111] discuss decision-making frameworks for automation but do not discuss strategic issues. In Ref. [109] authors discuss strategic issues related to automation but are focused on technology selection. Competencies in responsiveness, accuracy (errors in shipment), handling variety, and cost are considered important factors (volume variance, time variance, and quantity variance: [30]). Further, from a strategic planning perspective, growth (target future geographical markets-potential new stores), and product mix (which SKUs – cost and variety) are considered important. This requires competitor analysis and identification of value offered to the end customer. Market entry strategy that allays competition fears and ensures success may need to be considered in competitive regions (or product categories). Relevant market segments that would be served best by the retailer would need to be assessed. For example, would the retailer carry a large number of slow-moving items or only a small number, and what corresponding service levels would be important? Next, the corresponding supply chain strategy to support the business strategy needs to be evaluated. For example, the location of the DCs would be chosen based on the needs of the current and future demand centers (stores). The location of cross-docks (if any) and their role would need to be assessed (distribution-oriented role or supply consolidation-oriented role). In addition, different scenarios (which SKUs need to be supplied to the new DCs from which DCs or suppliers) for supply to the new DC/s need to be considered. As BCG,9 strategy consultants and retailers10 report how competitive pressures could be met by warehouse automation. Further, the retailer needs to consider the choice of technology, how it would affect the customer and employee interfaces, and how the retailer would manage the transition to automation (considering the technical feasibility and pace of automation in the retail sector).11 Next, the retailer would need to assess the availability of labor (quantity and quality and relationship with the union) and then decide on the extent of automation and corresponding module scenarios (M1, M2, and M3 report how automation helps make warehouses competitive, Table 1A2). Rapid-sizing tools could be employed at this stage for assessing the type and number of equipment necessary. The retailer would need to ensure that the supplier (such as an automation equipment supplier, strategic suppliers, transactional suppliers, etc.) relationships are sustainable to ensure the long-term success of the stakeholders involved. Next, we provide an overview of design and sizing analysis tools that are critical for rapid-design and scenario analysis for investment decision-making.

Methodology: Literature review and largely based on primary and secondary data collection (interviews and data collected over time; secondary interviews), conference presentation (trade and research), and work of other scholars (Table 1A/B).

6.2. DC sizing analysis and relationship with economic analysis

We refer to the frameworks proposed by noted scholars in the field and the lack of an integrated approach for warehouse design [17,30]. We provide a summary of the sizing sub-framework in Fig. 2a/b. We provide a brief on the different automation modules (fully automated modules and partially automated modules). We assume that strategic business analysis, supply chain analysis, and technology selection for an automated DC have been completed at this stage. A retailer first needs to classify the demand as follows a) items packed in regular-sized packages: these are items that are packaged into standard-sized cuboid boxes, b) items packed in irregular-sized packages: items that are packaged, say, in sacks (e.g., dog food), bags, etc. and are bulky or heavy, and c) individually picked items. We assume that every incoming pallet is broken down and items stored as cases/individual packs/bags in the DC. When an order is received, a pallet is assembled to a set standard.

Next, we provide a brief description of the three automated DC modules that cover a wide variety of SKUs: a) Regular case handling system (RCHS) for handling regular cases, b) Irregular case handling system (IRCHS) for handling irregular cases, and c) Less-than-case handling system (LCHS) for handling less-than case items. For a detailed discussion on specific technologies, we refer to the following - AGVs (automated guided vehicles - wire-guided or self-learning vehicles): [[112], [113], [114]]; Robotics12 (static robotic arms): [115]; Shuttle-based storage and retrieval systems [116]. Next, we briefly discuss the key issues related to various aspects of full-case technology and the trade-offs involved.

We briefly discuss the three modules next, each of which describes a varying amount of automation. For further details we refer the interested reader to the E-component (Section 1.1).

• a.Regular case handling system (RCHS) module: The module function is described in the E-component (Figs. A1–2a). The module is useful for SKUs that are amenable to automation. These SKUs must fit a specific size (L x D x W). This module can handle slow as well as fast-moving items. Aggregation of slow-moving items at the DC helps reduce system inventory. Further, quick response and the ability to create mixed pallets help deliver the needed items to the stores. A retailer must decide on the number of RCHS modules and the extent of over (under) investment if any, considering peak demand. Excess demand must be handled by other modules.
• b.Irregular case handling system (IRCHS) module: In this module, the pick-slots are filled using automated equipment and the picking function is largely manual (the operator uses pick-up carts/cars). Automation vs. manual operation cost tradeoff needs to be evaluated in this case.
• c.Less-than case handling system (LCHS) module: In this module, the pick slots are filled using automated equipment and the picking function is manual (the operator walks to the slot). Automation vs. manual operation cost tradeoff needs to be evaluated in this case.

Modeling methodology: We employ discrete event simulation and inputs (primary data) from the retailer, DCs, stores, and equipment supplier to guide our modeling effort. We also employ a math programming and a marginal analysis-based approach along with the results of the simulation, to define a) the dimensions (sizing) of the DC, i.e., length, height, and depth of different aisles (depending on the number of cranes needed), b) characteristics curves (curves that inform about the number of different machines needed, given certain throughput needs), c) buffer capacities. The characteristic curves serve as rapid sizing tools for a retailer – who could employ the curves to estimate equipment needed for different potential demands. The mathematical programming details are provided in the attached E-component (Figs. A1–2b).

Next, we provide a brief on the approach for computing network-related costs.

6.3. Assessing the network costs using an optimization-based approach

The important elements of a typical retail distribution and supply network we consider include a) Distribution center(s) b) stores c) suppliers (including backhaul routes) and d) cross-docks. The DCs could be: i) Single automated DC ii) Two automated DCs or iii) a combination of manual and automated DCs. The automated DC itself may include one or more of the three automation modules. We first present an outline of the network elements and their relevance to the framework in Fig. 3a/b.

6.3.1. Manual DC (pallet-based transportation analysis)

For manual DCs, a pallet-based analysis is conducted while for automated DCs a case-based analysis is considered. We first formulate the total cost problem as a ‘fixed cost’ problem used for pallet-based analysis (Modified-EOQ). We consider a 1-year planning horizon, but for delivery purposes, we consider a 15-day period for demand aggregation for multiple SKUs. We consider a variety of costs that are relevant in a retail context: the approach makes a trade-off between transportation costs, inventory costs, ordering costs, labor costs for processing a truck at the DC and a store, stock-out costs (with fixed-cost formulation) for a network with direct routes. For fast-moving items, full trucks are scheduled as needed. For slow-moving items, SKUs are aggregated and trucks are then scheduled. This approach provides us values for delivery quantity ($\hat{Q})$ and re-delivery quantity (r*) under the assumption that the DC makes direct delivery to a store (DC-store-DC) and the empty truck comes back to the DC. For partial pallet SKUs, we assume manual labor is used to make mixed-pallets which are then aggregated to compute transportation costs using direct shipments. We refer interested readers to the E-component (Section 2.1) for manual DC analysis: Fixed Cost Model) for further description of the transportation costs (inbound and outbound) for the pallet-based (manual DC) scenario.

6.3.2. Automated DC (case-based transportation analysis)

We first employ the ‘fixed cost’ approach as discussed in 6.3.1, with one difference: the analysis is case-based i.e., we assume that an automated DC will be involved and mixed-pallets (any number of cases of a particular SKU could be included). Next, we ascertain what fraction of $\hat{Q}$ goes on a truck such that the total cost is minimized. First, a route is selected using a ‘cluster first’ heuristic. Once the stores on a route are known, we use a truck-filling-distance-computation heuristic to finalize the store visited (not all stores may be visited). The E-Component describes (Section 2) the Fixed Cost Model (FCM) which is at pallet level and rough-cut analysis followed by a non-linear model which is suitable for case-based distribution analysis. We refer interested readers to the E-component (Section 2.2) for further description of the linear distribution model. The backhaul transportation costs and cross-docking are not included for brevity.

Our goal in developing the above analysis was several-fold: a) provide a mechanism to find total network costs for i) a manual DC serving a medium or a large-sized network, ii) an automated DC serving a small or medium-sized network (say 50 stores and 1000 SKUs), iii) an automated DC serving a large network (200+ stores and 5000+ SKUs); b) develop a heuristic that will be able to run in less than 12 h so that delivery could be scheduled multiple times in a week; c) be able to improve upon the total cost solution by proposing an optimal fraction of $\hat{Q}$ (using the top 1000 SKUs, selected by highest value –say, for the week); d) providing a tool to the retailer, which runs in real-time and provides adequate accuracy, for benchmarking. The analysis could be used to determine inbound, backhaul, cross-docking, and outbound transportation costs, stock-out costs (given the required inventory level at stores), and variable and fixed labor costs, at the DC and stores, related to transportation and inventory management.

Modeling methodology: We collect secondary data, interview store managers, and DC managers (conventional and automated DCs). This helped us to collect data and inputs for analysis and informed us about important costs and issues. We employ mixed-integer, linear, and non-linear programming for analysis [117]. In addition, we also develop a closed-form analytical solution. We refer interested readers to Section 2.0 in the E-component analytical details.

6.4. Analysis of impact of DC automation on store operations

Store operations are key to ensuring online as well as offline customer demands are effectively met. Automation provides a way to assemble pallets more intelligently so that they are aisle-friendly and this allows efficient (lower cost and effort) and improved (lower damage and reduced injuries) movement of materials to exactly where they are displayed in the aisles. Table 3 and Figure_Table 3 display the metrics developed to estimate operations improvement and the corresponding approach respectively. We are unable to describe the details due to space constraints and privacy reasons.

Modeling methodology: We collect primary data/make observations in the field (store operations at night when customer traffic is minimal) from stores (three) and discuss issues on the floor with the store managers and senior executives (primary data) and perform Excel-based analysis.

6.5. Economic analysis for decision making

The economic impact of each of the sections above (in Sections 7.1, 7.2, 7.3) is captured (in Table 2 to Table 4) and the payback period is then computed (Table 5A/B). For example, the scenarios discussed (Table 2) are the result of the strategic evaluation (Section 6.1) and further detailed analysis (Section 6.2, Section 6.3). The outcome of the analysis is captured and summarized (for example, Table 5A/B) for decision-making. Section 6.2 helps the retailer in evaluating equipment costs (investment) and labor costs (line items in Table 5). Capital-labor tradeoff is discussed in Fig. 4A/B. This is helpful in understanding the extent and type of automation.

Table 4.

Labor cost in automated 1-DC network.

Explanatory Notes:1. Two scenarios have been described in the table above 2. For each scenarios, three different modules have been considered3. For each module, different types of labor costs have been described.

Modeling methodology: For the above work we collect secondary data and perform Excel-based analysis.

6.6. Sustainable supplier engagement for safeguarding network assets

A retailer would be able to use the insight from the above analysis for writing contracts (developing specifications, identifying performance targets, specifying maintenance contracts, etc.) with the equipment provider. Further, with such knowledge and by considering key issues related to supplier-retailer relationship, retailer requirements for ensuring automated DC is well engaged, the importance of local vs. global sourcing, and sustainability principles the retailer would be able to ensure that the supplier network supports investment in the specific asset (automated DC). We refer to the sub-framework in Fig. 4a/b and the outline of the modeling method in Fig. 4b. In the E-component (Section 3.0) we provide the details of corresponding modeling approach.

Modeling methodology: We collect secondary data from the Alibaba website (Alibaba.com). We develop economic models and solve them using mixed-integer linear programming.

Next, we provide a summary of the tools and their illustration through a business case analysis.

7. Business case analysis

7.1. DC inventory evaluation

In this section, we briefly describe the tools that help analyze automated DC inventory costs, the benefits of an automated DC at the store level, and payback period computations based on the assessment of the described costs and benefits. We obtained the costs of the conventional DC-related elements from the retailer, in addition to drawing on our analysis. We employ it as a reference for the payback period computations.

Given an automated distribution center, we compute the on the hand (OH) inventory based on a minimum order quantity (MOQ), i.e., the minimum number of units the supplier would agree to supply or the retailer must order (depending on retail needs and supplier economy of scale and negotiations). Further, we employ materials requirement planning (MRP) logic to assess order quantity. We illustrate the MRP logic below.

• 1.For each item, MOQ (${MOQ}_i)$ is chosen, a parameter that is mutually decided upon (between a supplier and a retailer based on supplier lead time, capacity, and transportation agreements between a retailer and a supplier).
• 2.We have, ${OH\_quantity\left(t+1\right)}_i$ = ${OH\_quantity\left(t\right)}_i$ – ${Demand\left(t\right)}_i$, where ${\left(t\right)}_i$ denotes time period for item $i$.
• 3.We assume daily review and consider daily demand ${Demand\left(t\right)}_i$.
• 4.If the ${OH\_quantity\left(t+1\right)}_i$ < ${LT}_i$*(${Average\_dail{y}_{demand}\left(t\right)}_i)$, we order ${MOQ}_i$, assuming ${LT}_i$ to be lead time for item $i$ , based on discussion with the retailer.
• 5.${Average\_OH}_i$ quantity is computed as the average of ${OH\_quantity\left(t\right)}_i$ over the year.

To keep computations consistent, we keep the same policy for safety stock in the automated DC (10 % of the on-hand inventory). The average cost of OH inventory = $\sum _{i=1}^{N_{SKU}}{N}_i*{MOQ}_i*C_i/2$; Where ${\mathrm{N}}_{\mathrm{i}}$ is an integer multiple (${\mathrm{N}}_{\mathrm{i}}\ge 1)$ of the ${MOQ}_i$ quantity ordered as suggested by the MRP algorithm. Our assumption is that the suppliers can deliver the inbound materials in cases and layers, especially for slow moving items. Employing the standard practice observed, that each store gets two or more deliveries per week, we assume that the safety stock to be 10 % of the on-hand inventory. Further, based on observation and practice, we compute the store inventory to be half-the amount of that at the DC.

7.2. Labor analysis for an automated DC

Based on our observation and for simplicity purposes, we employ only three labor types (manual, technical staff for operations/error correction, and maintenance) for handling varying tasks in an automated DC. We also assume that it would be necessary to employ supervisory/office staff. Typical employee activities include: a) material receipt, b) loading the material on conveyors for storage, c) retrieving materials from conveyors on receipt of an order, and once a pallet is made (using automated processes) and moved to the shipping dock, d) driving the materials using forklift to a truck and e) loading materials on to a truck. The technical staff (operations and error correction employees) constantly monitor operations during the shifts to ensure smooth operations. Maintenance staff maintains the equipment/facilities during the maintenance shifts. The maintenance shift works for approximately 4 h a day while the production shifts (two shifts) work for the remaining 20 h (two 10-h shifts). Table 4 summarizes the results from such an analysis.

7.3. Store savings analysis: travel time savings from pallet breakdown and shelving operation

We first make observations (see Figure_Table 3 of approach to making observations on store floor) on the current state (conventional DC-driven distribution network) by observing practices in a store for a regional retailer. We then summarize our observation on the time needed for the break down and shelving operation of a representative pallet. We seek to develop a set of metrics that would help characterize store operations. We present a comparison of metrics (for current and future states, i.e., post automation, where we assume that aisle-friendly pallets would be delivered to the store) in Table 3. Information for each metric related to the distance travelled, time involved in different elements, number of employees needed, etc. are shown. The estimate the metrics for a future state are based on a given layout of a store, given the fact that the pallets are aisle-friendly13 pallets assembled at an automated DC. We observe uniform (same item) and mixed pallets (different items in one pallet). We use the metrics above to compute savings from changing over to an automated DC.

For interested readers, we refer to the E-component that depicts the approach to data collection on current and future states (different time and cost elements involved) that includes automation.

7.4. Store savings analysis: damage costs

In an automated DC, the damage to packaged items could occur in the high-bay warehouse, in the tray warehouse, or in the shipping and receiving docks. If cross-docks are part of the distribution network (in case of multiple DCs), then the damage may occur at cross-docks. During transportation, damage could occur if the pallets are not arranged properly for transportation in the trailer. At the stores, the damage could occur while unloading from the trailers, while moving the pallets to the staging area, during the pallet break-down operation, while moving the packages to the shelving area, and while shelving. We find, that based on our interaction with retailers and DCs, the following issues affect product damage: a) Pallet handling (receipt, storage, and processing once the demand is registered), b) Structure of the pallets, c) Pallet build/assembly, d) Operator training in material handling, e) Pallet break-down and shelving operations in stores. An automated DC helps reduce the chances of damage in the DC as well as in the stores. We assess damage in stores (observing and documenting DC operations is much more complex and is a part of our future research) which is largely due to pallet break-down and shelving operations.

8. Discussion and conclusion

8.1. Discussion on results and practical implications

We briefly review several insights that we obtain from our analysis. From strategic level analysis, we find that from Table 5B it is clear that under different scenarios, the 1-DC approach is better than the 2-DC approach. Sources of advantage for the 1-DC approach are a) the on-hand inventory (due to the consolidation of inventory in one location the consolidated DC faces a higher demand rate and more frequent supplier deliveries, given a negotiated MOQ), b) labor (due to elimination of redundancy in supervision/overheads), and c) inbound transportation cost (a supplier ships only to one DC and the location of the 1-DC is such that the supplier-DC distance is favorable for overall supplier-shipments, i.e., the inbound cost decrease is more than the corresponding cost increase). In addition, d) the investment needed in the 1-DC approach is lower due to better utilization of the DC equipment. The source of disadvantage is e) the outbound cost as it is inefficient for one DC to ship items to distant stores. Another source of disadvantage would include f) the risk of consolidation: a disruption in distribution could occur if the single DC operations are impaired due to either manpower issues, natural calamity, or machine failure-related issues. Moreover, g) a centralized automated DC (1-DC) is less flexible since it all depends on the equipment and the equipment must work else, the DC deliveries to stores suffer. A conventional DC is less prone to such a failure but is more prone to disruptions due to labor unrest. In addition, natural calamities and labor unrests affecting transportation (in the DC area) could affect transportation costs (to and from the single DC). We find that h) both of these issues could be addressed by a more flexible DC design (moving away from 100 % automation), improved management practices, and planning (a mix of self-owned and outsourced transportation).

We note that deeper insights into the costs/benefits of an automated DC were obtained by the cost analysis at the mentioned four levels: a) the strategic level, b) supply chain level, c) DC level, and d) store level. At the supply chain level (given the structure of a network), a retailer (network coordinator) should: i) evaluate changes in costs related to - a) transportation, b) supplier contract, and c) projected fuel usage; ii) conduct impact analysis of layer-based vs. pallet-based truckloads in order to understand the benefits of a shift to an automated DC; iii) evaluate the benefits that accrue due to a) improved backhaul opportunities and b) trailer packing efficiency (achieved due to the automated pallet assembly process). Added benefits would depend on the network structure features such as location of suppliers, stores, and the DC/s; we also find that iv) the payback period analysis shows lower sensitivity to fuel costs, damage costs, but higher sensitivity to labor costs. We refer to Fig. 5a where, given a labor cost, the retailer could consider different combinations of automation modules based on anticipated demand. Thus, labor-capital tradeoff could be evaluated: v) thus, if the demand has a high variance, the retailer would be better off investing in a combination of modules (instead of just RCHS/full automation). A retailer would benefit from scenario analysis (Table 4) for 1-DC, 2-DC or other such networks depending on the network design and must then consider detail computations as in Table 5A, Table 5Ba and 5b.

Employing the DC-level analysis we find that inventory considerations are important for evaluating improvements due to an automated DC. Following insights were obtained from this perspective: i) the supplier capability to supply layered pallets and the DC capability to process orders at a case level (employing an automated DC) are critical to deciding the inventory levels at the DC; ii) additionally, if the response rate of the suppliers to supply layered pallets is high along with a similar high response rate that is obtained by employing an automated DC, the network inventory, including that in stores, could be reduced; iii) that the labor costs contribute to a significant portion of cost savings/benefits that accrue by moving to an automated warehouse (for example, in the case of Scenario 2 the saving proportions are: 47 % from labor, 29 % from inventory, 2 % from transportation, and 22 % savings from moving to automation; here we assume a 2-DC model).

At the store level we have the following insights: the savings (from moving to DC automation) result from i) reduced damage costs, ii) reduced labor costs due to use of aisle-friendly pallets, and iii) reduced employee injuries that occur due to excessive material handling; iv) in the case of aisle-friendly pallets, a forklift could deliver the pallet to a pre-decided location in the aisle (where the items are needed) resulting in reduced shelving time since the sequence of items in the pallet matches the sequence in the aisle (automated assembly process is fed the planogram of the aisle and knows what is missing where and the pallet is accordingly assembled); v) an automated DC would also be a natural choice if the stores change planograms often.

Finally, vi) the retailer must plan for maintaining needed customer service levels for each item while shifting between conventional to an automated DC technology. Highly trained employees are needed to man an automated warehouse on a daily basis and hence it is critical that suitable contracts are developed and agreed upon with the equipment manufacturer. Thus, quantitative and qualitative issues both contribute to successful consideration for investment in an automated DC.

8.2. Conclusion: managerial contributions/implications

Our research has several implications and contributions for practicing managers: a) We identify several aspects of retail operation (as indicated in Table 5A) that a manager must consider before computing payback period, b) the proposed three-part framework and the three associated sub-frameworks would guide mangers in choosing a suitable automation solution and deploying it, c) we propose that the business strategy and supply chain strategy, and its impact on automation must be considered, d) impact of automation on suppliers and customers (stores) must be thoroughly evaluated to understand benefits, costs, and risks (along the three proposed dimensions) of automation, e) we propose that the choice of technology (automation and IT) and extent of automation need to be carefully thought through based on OC/MC needs, f) the proposed simulation and math-programming based analysis can help in investment estimation, in layout finalization, and in buffer sizing, g) the proposed math-programming based analysis/approach can help improve inventory at stores, truck efficiency, delivery frequency of an SKU, cross-dock analysis, and total cost in the network for case-based (automated DC) set-up, h) safeguarding such investment is key to managing the network and a hence the proposed sustainable contract management framework is key to ensuring long-term and sustainable supplier relationships: supplier cost, benefits, and risk could be considered in an integrated approach to solving product portfolio allocation, given a global supply base.

8.3. Conclusion: theoretical contributions/implications

From a theory perspective, we contribute to the DC automation literature (theory) by developing a three-part investment decision-making framework for DC automation (a general framework that we illustrate for ‘full-case’ technology) that extends and integrates similar contributions (theoretical) of other scholars. Further, we propose corresponding sub-frameworks that are key to furthering research in the field. For example, simulation of DC operations (DC-design sub-framework) leading to a well-designed and assembled mixed-pallet could guide transportation (routing and delivery quantity) decisions (network analysis sub-framework) as well as composition of pallets to ensure aisle-friendly pallets. We also show the link between transportation networks (network analysis framework) and inputs needed for sustainable contracts and for ROI computations. We are motivated by the resource-based view (RBV), the Triple Bottom-line perspective for sustainable supply chain management (SSCM)/supply chain risk management (SCRM), and contract theory (Transaction Cost Economics and Relational Exchange theory). It is essential that the retailer (a network coordinator) utilize the scarce resources that they have access to obtain a competitive advantage that could be sustained (cannot be easily copied). Thus, if a retailer deploys the right combination of automation technologies (appropriate for the market) in the DC and, based on that, deploys appropriate transportation and inventory in the network so that labor, service levels, and out-of-stock costs are at the levels that are better than those in the market, the retailer will be able to compete well in the long run. We contribute to sustainable supplier management literature by proposing an integrated supplier selection, transportation, and behavioral model that would be essential to managing the network successfully, safeguarding the network assets, and supporting customer needs in OC context.

Our math and economic models contribute to the modeling literature and help illustrate how different costs in retail networks could be modeled and used for optimization models.

8.4. Limitations and future research direction

The damage costs at DC (conventional and automated) need further investigation. We expect these would contribute to savings for an automated DC. We have observed and estimated damage costs at the stores since these were considered a cause of concern by the retailer. But, we find that a more detailed ongoing effort in data collection would be needed. Direct delivery of aisle-friendly pallets to the shelving locations also needs to be investigated further. Cross-dock considerations would require further analysis and considered a topic for further research. We also could not investigate the impact of ‘hub-spoke’ networks (hubs act as cross-docks, employ smaller trucks, and make smaller and more frequent deliveries to nearby stores) and believe it will help reduce inventory at stores and transportation costs while contributing to improved service levels. We also did not consider the cost-benefit impact of futuristic technologies such as a) robotics that would replace manual labor in truck loading/unloading, pallet breaking, and shelving at the stores; b) drones for transporting items from stores to the end customers; c) driver-less trucks. These technologies are in the pilot stage and significant research would be needed to uncover the technical and managerial impact.

Direction for future research includes: In this paper, we have made an attempt to highlight the complexity of key elements that go into the decision-making framework for investment in DC automation. As is evidenced by our work, the issue consists of solving multiple complex problems that must come together, considering both theory and practice, given a OC/MC setting. We believe the field will benefit from research that focuses on: a) We find that there are only a few research articles linking business strategy to supply/distribution strategy (research integrating all key elements of the a successful OC value chain) in OC context, given the objective of identifying which business segments will benefit from a technology, b) a comparative analysis of dominant technologies (for example, robotics) and their suitability for specific type of application need to be investigated with a goal of automated DC investment, c) extent of automation is important factor and needs to be investigated, given business segments (case studies, system-level modeling and economic analysis), d) for a given/chosen technology, DC and network level sub-frameworks need to be evaluated, e) impact on DC layout, key machinery, labor, and other resources need to be further investigated – through what-if and scenario analyses, f) impact of DC on customer (stores) needs further investigation (different pallet breakdown and shelving approaches, impact on damages, and worker injuries), g) further investigation is needed from network perspective to include multiple centers of production (warehouses that could reconstitute pallets to meet demand on short notice), cross-docs, and stores that serve as small warehouses for supporting stores in a local region, and h) further modeling of sustainable supplier contracts needs to performed – that includes network risks, costs, and benefits with a goal of investigating safeguarding network assets, i) sustainability principle must also be investigated as OC customers value such principle and may not continue to buy from a retailer/supplier that does not follow such principles.

Data availability statement

The authors do not have permission to share data:

The data was made available to us under the condition that it would not be shared and we have signed a non-disclosure agreement. In return the retailer, the manufacturer, and the DCs have shared the data and also helped validate the results. We are allowed to publish results of our research that is representative of our observations and data employed.

Has data associated with your study been deposited into a publicly available repository? Response: No.

Has data associated with your study been deposited into a publicly available repository? Response: The authors do not have permission to share data.

Statement of consent with the ethics information.

The organizations (retailer and equipment manufacturer) and personnel that participated in this study provided their consent through non-disclosure agreements which were signed with each organization.

CRediT authorship contribution statement

Vivek Kumar Dubey: Writing – review & editing, Writing – original draft, Validation, Resources, Methodology, Investigation, Formal analysis, Conceptualization. Dharmaraj Veeramani: Writing – review & editing, Validation, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: