Adoption of AI-based order picking in warehouse: benefits, challenges, and critical success factors

AI is a broad term that encompasses various technologies and applications (Helo and Hao 2022). AI can be the underlying element for other technologies, such as robots and automation (Dash et al. 2019). In essence, AI refers to machines or systems that possess intelligence comparable to or, in some cases, surpassing human cognitive abilities (Wang et al. 2023).

Past literature has shown that AI can play a significant role in improving supply chain processes, such as sales, smart connected products, operations, and service maintenance (Helo and Hao 2022). Based on a survey of 250 company executives, Davenport and Ronanki (2018) highlight the targeted benefits and challenges that are often associated with AI for businesses. These benefits are product improvement, enhanced decision making, new product development, operational optimization, increased worker freedom, expansion into new markets, acquisition and application of scarce knowledge, and reduced workforce through automation. Conversely, the challenges identified include complex integration of AI with existing systems and processes, the high cost of AI technology and required skills, limited managerial understanding of AI technologies and their functioning, limited specialized human resources, technological immaturity of AI, and overselling of AI technology in the marketplace.

The previous literature on AI in warehouses explores its applications, benefits, challenges, and critical success factors. Khan et al. (2021) focus on using robots and AI to ensure food safety. Zhang et al. (2021) emphasize the importance of integrating AI elements, such as data, algorithms, and robots with warehouse-specific systems to enhance forecasting, planning, and learning. They also highlight the evolving role of human capabilities alongside AI. Mahroof (2019) identifies benefits, such as reduced travel time and precise picking processes, while also emphasizing challenges, such as technology maintenance, system interoperability, and limitations of managerial knowledge.

In their categorization of AI, Metaxiotis et al. (2003) state that there are three types of AI (as cited in Meyer et al. 2020). The first type is expert systems, which utilize an extensive database on the topic to find an answer to the problem. The second type is artificial neural networks, which search for patterns within the available data. The third type is autonomous agents, which can autonomously act in a given environment.

In warehouses, AI enables automation, which is manifested in different ways, such as automated tridimensional storehouse (Zhang et al. 2021) and machine-based order picking (de Koster et al. 2007). With its pattern-finding capability, AI can help with various warehouse-related tasks, such as demand forecasting (see, for example, Zhang et al. 2021). Moreover, AI-based expert systems can contribute to the management of flexible order-picking systems (Manzini et al. 2005).

Expert systems are AI-driven computer programs that can emulate human thinking to produce an outcome that a human expert could (Olson et al. 1992, as cited in Mirčetić et al. 2016), often much faster and with fewer errors. It is a useful tool to help decision making within a domain where expertise is demanded (Turban 1995). Expert systems enable transforming knowledge into wisdom, thereby assisting the knowledge-gaining process (Shokouhyar et al. 2019). Past literature has demonstrated various applications of expert systems (Yao et al. 2010; Mirčetić et al. 2016; Patriarca et al. 2016; Leung et al. 2018). For example, expert systems can assist with decisions on defining inventory levels (Patriarca et al. 2016) or optimizing the use of forklifts in cargo loading (Mirčetić et al. 2016). The AI-based systems utilized in this paper refer to the expert system type, specifically designed for route optimization in warehouse order picking.

Order picking activity is a critical component of the warehousing process, involving the selection and collection of the correct items in the right quantities based on customer orders (de Koster et al. 2007). There are three primary order-picking strategies: picker to goods where the picker retrieves the customer's order and travels through the warehouse to pick the items, either by foot using trolleys or cages, or by pallet jacks and forklift trucks using pallets; goods to picker, which involves storing the items in narrow aisle racking or in the automated storage and retrieval system (AS/RS) and automatically moving them to the picker’s workstation, thereby reducing picker travel time; automated picking, which means the order picking activity is fully automated, using various advanced technologies and equipment (Richards 2018). Each of the above-mentioned strategies has its own advantages and disadvantages. Among them, picker to goods is the most commonly applied order-picking strategy (de Koster 2004; de Koster et al. 2007; Richards 2018). This strategy is also employed in the warehouse case discussed in this paper.

There are four approaches in the picker to goods strategy: cluster picking, batch picking, zone picking, and wave picking (Richards 2018). The cluster picking approach involves pickers taking on a number of customer orders together, traveling through the warehouse and picking items jointly. The picked items are immediately sorted and grouped into their respective cages or trolleys based on the composition of the orders. Batch picking also includes joint picking for different orders, but the orders are clustered and sorted after the picking is completed. With zone picking, pickers are assigned to different areas/zones in the warehouse to perform picking in their corresponding zones. In this approach, order picking can flow from one zone to the next, either simultaneously or sequentially, depending on the degree of activity and frequency of customers’ orders until the picking process is completed. Finally, wave picking involves the grouping and releasing of orders at specific times throughout the day or in step with cycles of replenishment, departure of vehicles, alterations in shifts, locations and commonality of products, priorities, and requirements for value-adding services. In addition to these four approaches, there is the single order-picking approach in which the picker travels across the warehouse to fulfill one customer order at a time (Lin and Lu 1999).

Recent literature on AI-based order picking reveals significant advancements in optimizing warehouse operations through various AI-driven methods and technologies. Zarinchang et al. (2024) introduced an AI algorithm that reduced picker travel distances and improved efficiency. However, despite efficiency gains, the improvements in rack stability and safety were only moderate. Sodiya et al. (2024) emphasized the significant role of AI-driven autonomous mobile robots in enhancing inventory management and order fulfillment by optimizing routes and adapting to warehouse changes. Kalkha et al. (2024) introduced an AI algorithm that clusters and positions items with similar demand patterns, reducing order preparation time and unnecessary movements. De Assis et al. (2024) recommended further exploration of technologies, such as digital twins, to advance intelligent warehouse management. Zhao et al. (2024) proposed a goods-to-picker system combining AI and robotics to improve picking efficiency. Balzereit et al. (2023) used AI to optimize order picking by merging small orders and re-sequencing them to save time. Zhou et al. (2023) demonstrated how a batching AI algorithm could optimize order-picking schedules in smart warehouses. Cano et al. (2023) showed that a genetic algorithm outperforms ant colony optimization in picker routing, which enhances performance and speed.

The case revolved around the adoption of an AI-based system in the warehouse. The goal was to identify the benefits, challenges, and critical success factors associated with the adoption of the AI-based system in the warehouse.

With the advent of Industry 4.0, the warehousing process gains significant importance in the supply chain. According to Voss (2021), the increase in the customization of products and the potential for rapid fluctuations in demand in the Industry 4.0 era have heightened the complexity and the efforts required for activities such as picking, packing shipment, and resource management. Consequently, the warehouse’s tasks expand, and resource management intensifies to improve supply chain efficiency (Voss 2021). In the context of Industry 4.0, supply chain warehousing needs to be more efficient and responsive due to the growing complexity of the activities involved. Indeed, order picking is the major activity in warehousing and represents a large portion of the warehouse's operating cost. Thus, inefficiency in this activity has a vital impact on the entire supply chain (Coyle et al. 1996; de Koster et al. 2007; Richards 2018). Therefore, investigating a system with such improvement potential, along with its corresponding challenges and critical success factors, is vital to respond to the emerging demands surrounding the warehousing process in Industry 4.0.

The case study’s warehouse is located in Türkiye. Türkiye was selected for several reasons. First, Türkiye’s strategic geographic location, a bridge between Europe, Asia, and the Middle East, makes it a vital hub for international trade and logistics (Özdemir 2010; Acar et al. 2015, 2020). The country's advantageous geographical location and role as a gateway to various markets necessitate the adoption of advanced technologies to optimize warehouse operations and enhance competitiveness (Saatçioğlu et al. 2009; Erol et al. 2021). Secondly, due to the growing demand for efficient and accurate order fulfillment processes, the implementation of AI technologies in warehouse operations has become crucial (Cergibozan and Tasan 2022). Moreover, the Turkish government has actively supported the digitalization of industries, including logistics, through various initiatives and incentives (Duman and Akdemir 2021).

Nevertheless, Türkiye has delivered mediocre results in transitioning to advanced technologies and Industry 4.0. (Izmen et al. 2020). Hence, investigating the benefits, challenges, and critical success factors associated with an AI-based system in a Turkish warehouse aligns well with the current status and characteristics of Türkiye’s transition to advanced technologies and Industry 4.0.

The AI-based system was developed by Kairos Logic AB (hereafter Kairos), a Swedish microservices company providing dynamic order-picking intelligence to help businesses meet logistic and warehouse process needs. Kairos’s AI-driven algorithms optimize warehouse and logistics resources to drive customer operations efficiently. Sienzi Lojistik Antrepo ve Dış Tic. Ltd. Şti. (hereafter Sienzi), a well-established company in the field of warehousing in Türkiye, has adopted this AI-based system, offering a wide range of services to meet customer needs (Sienzi 2023). Sienzi’s warehouse covers an area of 7000 square meters, including 6,000 square meters of closed storage and 1000 square meters of open space. Its 14-m height allows a 500-pallet racking system, and there are various areas for inspection, handling, and reservation, all compliant with customs regulations. The 14 hydraulic ramps enable simultaneous operations for up to 8 trucks while smaller vehicles can be serviced in the open yard area. The company is located near the Muratbey Customs Directorate and ensures timely delivery through regular courier services. Sienzi has served 2,819,165 product entries over the past ten years. Alongside its warehousing capabilities, the company offers value-added services, such as handling, packaging, and palletizing. Sienzi has expanded its operations to include e-commerce logistics, catering to online businesses’ storage and distribution needs. Furthermore, the company provides loading services for international exports. The company operates on a project basis, and during the study of AI adoption in the warehouse, the company project focused on automobile parts.

The warehouse uses a conventional layout structure. This layout arranges racks with parallel aisles (Oxenstierna et al. 2022). Figure 1 depicts the layout of the warehouse. Products are organized from the bottom of the shelves, grouping similar products within each column. Then, another product is placed similarly, following the same bottom-to-top arrangement. Furthermore, the figure illustrates how a warehouse layout is designed specifically for order picking. The collection point is located at the top right corner, and there is a designated tunnel that serves as the primary pathway for pickers to navigate through the aisles to pick up customer orders.

Full size image

Using historical pick runs spanning a month and the warehouse’s floor map, Kairos developed a digital twin simulation of the warehouse. By utilizing the cloud technology, the digital twin can then apply AI algorithms to improve order picking in Sienzi. The integration of the AI-based system with the warehouse’s ERP was achieved through the establishment of an application programming interface (API). The ERP system, developed by Select Bilişim Hizmetleri AŞ (hereafter Select), one of Türkiye's well-established logistics software companies, also includes a warehouse management system (WMS) module. Based on customer order data from the ERP, the AI-based system has produced optimal routes for picking.

The list of orders is received at the warehouse. Then, Sienzi’s operations officer requests an optimized picking solution from the AI-based system. The system provides the solution in real time or in a few minutes, and it is then given to the picker.

The study setup is designed in three modes: traditional single-order-based pick rounds, order batching, and batch-tunnel optimization. The baseline mode was single-order-based pick rounds starting from the collection point, picking items from the shelves for a single order, and bringing the electrically powered reach truck forklift back to the starting point. Using the forklift, the picker lifts a basket with a capacity of 40 items, picks the items, and places them in the basket. The picker repeats the process if the entire single order is not picked in one round due to picking capacity limitations. In the order batching mode, the Joint Order Batching and Picker Routing Problem (JOBPRP) was used (Valle et al. 2017; Gils et al. 2019), which integrates the Order Batching Problem (OBP) (Gademann et al. 2001) with the Picker Router Problem to optimize picking several orders at a time in the shortest distance. For the batch-tunnel optimization mode, an additional tunnel was added to the simulations to evaluate its resulting impact. The tunnel was added to batch optimization to address the limitations imposed by the layout structure on optimization capacity. By incorporating an imaginary pick tunnel positioned across the aisles in the upper section of the warehouse layout, the aim was to evaluate the potential of reducing total travel distance in theory (see Vaughan 1999). The tunnel’s position was carefully selected to minimize its impact on the aisles and to be near the warehouse wall. Orders used for the study’s analysis were randomly selected orders. Conducting real picking in the first two modes (i.e., single order-based pick rounds and order batching) and simulated picking in the third mode (i.e., batch-tunnel optimization), underpinned the empirical ground for the qualitative and quantitative analyses.

Order picking performance was assessed by aggregating total travel distances and the travel time required to pick all orders. Travel time is considered a waste in the context of order picking (Bartholdi and Hackman 2005), and reducing travel time can lead to cost savings and improved service levels (de Koster et al. 2007). Comparative analyses were made based on the total distance and time spent between baseline order-picking rounds, order-batching optimization, and order-batching optimization with an extra pick tunnel. Three hypotheses are proposed to test the effectiveness of the AI-based system in the warehouse’s order-picking process. The first two hypotheses test whether travel distances have been improved, while the third hypothesis examines changes in travel time, as follows:

Numerical data were collected by conducting systematic direct observations in the warehouse during single-order picking and picking with AI-based order-batching optimization. One of the researchers accompanied the warehouse picker during order-picking activities to ensure real-time data collection with high accuracy and reliability in measuring travel distances employing the warehouse layout and a stopwatch to calculate travel time. For the third stage, the data collection involving the AI-based order batching with an additional pick tunnel was conducted using simulation. The results obtained from the warehouse picking operations were subjected to statistical analysis to examine the potential significant differences between AI-based picking and traditional order picking in terms of time and distance. Descriptive statistics for time and distance were calculated for both the order-picking and batch-optimization groups. The selection of statistical tests was guided by the need to ensure rigorous and valid comparisons between groups. Initially, a t-test for two independent samples was considered to assess the significance of differences between the groups in line with the suggestions advanced by Hair et al. (2019). However, nonparametric alternatives were also considered, given the potential violation of the normal distribution assumption, which could result in type I errors (Field 2013).

Prior to conducting the statistical analyses, the normal distribution was assessed for traditional single pick rounds and order-batching optimization using the two-sample Kolmogorov–Smirnov test (Tabachnick and Fidell 2013). The results indicated that there was a significant deviation from the normal distribution for both time (p = 0.036) and distance (p = 0.000). Additionally, single pick rounds included 29 pick rounds, while AI-based order-batching optimization had 25 rounds, with each group having fewer than the threshold value of 30 observations. According to Hair et al. (2019), in samples of 50 or fewer observations, especially when the sample size in each group is below 30, significant departures from normality can greatly affect the results. Therefore, a nonparametric test was more applicable because the data deviated from the normal distribution, and the number of observations was insufficient (Dodge 2008; Hair et al. 2019). The Mann–Whitney U test, the nonparametric analogue of the t-test for two independent samples, was employed to investigate whether there is a difference between the groups regarding distance and time (Sheskin 2020). Statistical significance using a p-value less than 0.05 was considered.

Further analysis focused only on the travel distance metric and was conducted after the tunnel simulation because the simulation could not objectively and realistically cover time. This stage of the analysis used one-way ANOVA for three groups: single-order picking, (AI-based) order-batching optimization, and (AI-based) order-batching optimization with an additional pick tunnel. Prior to conducting the statistical analyses, the normal distribution was assessed using the Kolmogorov–Smirnov test. The results indicated that there was no significant deviation from normal distribution for travel distance (p = 0.200). Additionally, skewness and kurtosis values of the travel distance variable were checked to meet normal distribution assumptions. According to Kline (2011), the absolute value of skewness should not exceed ± 3, and the absolute value of kurtosis should not exceed ± 10, to avoid significant deviations from normality in the data. Values exceeding these thresholds may indicate substantial departures from normality in the data. The skewness and kurtosis values of travel distance were determined as 0.808 and 2.883, respectively, falling within the acceptable threshold values of less than 3 and 10, respectively. Based on these results, a parametric test (a one-way ANOVA test) rather than a non-parametric alternative was utilized to check if the travel distance scores of these three groups showed a significant difference (Kline 2011; Hair et al. 2019).

In conclusion, the study’s methodological approach incorporates robust statistical techniques to account for potential violations of normality and small sample sizes. Using the Mann–Whitney U and one-way ANOVA tests, the study ensures that the comparisons between traditional and AI-based order-picking methods are consistent and valid.

This study also examines the AI-based order-picking adoption in the case of Sienzi from a qualitative perspective. The data collection included a combination of interviews, direct observations conducted by two research team members at the warehouse site, seven meetings, and communications through WhatsApp and email, supplemented further by documents from Kairos. These data collection methods are particularly valuable for case studies and qualitative research (Barrett and Twycross 2018; Yin 2018).

The interview protocol was developed based on key themes from the pertinent literature, which guided the development of the interview questions. Key aspects of the warehouse’s order-picking activity were discussed in the interviews, including operational efficiency, system integration, employee training, and the overall impact on warehouse performance. The qualitative approach involved 11 semi-structured online and face-to-face interviews. Nine of these interviews were conducted individually, while two were group interviews. Participants included eight representatives from Sienzi, one from Select, and two from Kairos. This triadic approach facilitated a comprehensive understanding of diverse stakeholder perspectives. Face-to-face meetings took place in the headquarters of Sienzi and Select, and only meetings with Kairos’ representatives were conducted online for convenience. Prior to each interview, participants were informed about the study’s objectives, and participants' consent to record the interviews was obtained beforehand. The group interviews exclusively involved a mix of managerial and operational-level participants from Sienzi. Group Interview 1 comprised four managerial-level participants, and Group Interview 2 consisted of two operational-level employees. Details of the interview specifics and participants are presented in Table 1.

The interview questions were developed based on concepts from the literature (see Table 2). In total, 22 concepts formed the basis for all the interview questions. Of these, nine concepts targeted general inquiries from Sienzi, while the remaining 13 concepts explored specific topics drawn from past literature. These 13 concepts were examined in relation to the impact of the AI-based system on them.

One of the key 13 concepts is traveling time, which constitutes the largest proportion of the entire picking time (de Koster et al. 2007; Tajima et al. 2020). Our case presents the AI-based system specifically designed to affect the traveling time while optimizing traveled distances. Hence, this study selected traveling time for evaluation among the various time-consuming sub-activities. The next concept is pickers’ wellbeing, which involves various components such as workers’ health, motivation, and comfort (Larco et al. 2017; Richards 2018; Glock et al. 2019; Gajšek et al. 2021).

The third concept focuses on picking quality in terms of the number of damaged or incorrect orders (Kiefer and Novack 1999; Richards 2018). The fourth and fifth concepts center on operational and data flow control (Richards 2018) and documentation, which relates to the format and quality of checklists and printouts (Kiefer and Novack 1999; Richards 2018). The sixth concept refers to supply chain actors’ collaboration, which is manifested in the degree of partnership and cooperation (Richards 2018). The seventh concept addresses cost changes due to the arrival of the AI-based system, whether increasing or decreasing (Kiefer and Novack 1999; de Koster et al. 2007; Richards 2018). The eighth refers to the changes in the physical space occupation of the warehouse due to the adoption of the AI-based system (de Koster et al. 2007; Richards 2018). The ninth concept examines privacy, legal, and security concerns that may arise when using an AI-based system, such as data protection (Richards 2018). The tenth concept focuses on ease of use/training, which revolves around the readiness of workers to use the AI-based system (De Vries et al. 2016; Richards 2018; Pasparakis et al. 2023). The eleventh concept concerns system interoperability, which focuses on the degree of match between AI-based systems and the current technological infrastructure of the warehouse (Mahroof 2019). The twelfth concept considers items accessibility for the workers in the warehouse (Richards 2018). Finally, the thirteenth concept refers to the environmental impact that an AI-based system can generate in the warehousing process, such as changes in energy consumption (Đukić et al. 2010; Richards 2018; Bartolini et al. 2019).

We used the Gioia approach to conduct qualitative analyses (see Gioia et al. 2013). Through this approach, we developed a data structure (see Tables 5, 6, and 7) that made it easier to track the process of converting interview data into final themes. Drawing on Gioia et al. (2013) and consistent with Tables 5, 6, and 7, the data structure starts by developing 1st-order concepts from the raw interview data. Then, concepts filter into 2nd-order themes based on relevance and characteristics. Third, they emerge as aggregate dimensions of benefits, challenges, and critical success factors. A team of two authors undertook all three steps in developing the qualitative analysis and data structure to ensure a more reliable investigation. The development of the 1st-order concepts began with scanning the raw interview data, followed by marking and coding relevant parts of the texts based on the reviewed literature and research focus. Initially, each author independently scanned and marked the raw interview data. The outcomes were then compared and discussed by the research team members, resulting in the emergence of the final 1st-order concepts. The authors subsequently grouped these concepts into 2nd-order themes. A similar grouping process was used to develop the aggregate dimensions.

The transparency provided by Gioia’s data structure makes it a valuable methodological tool for conducting and presenting qualitative analyses, addressing the critique of data cherry picking often associated with qualitative methods (Gioia et al. 2013). Gioia et al. (2013) explain their intention behind the Gioia approach, noting that it aims to avoid cherry-picking quotations from reports, relying on clever explanations, and wrapping insights in appealing labels. The Gioia method provides a reasonable degree of visibility, enabling input tracing from raw data to the final output.

The traditional order-based single pick rounds consisted of 29 pick rounds, which were reduced to 25 batches in 25 rounds using order-batching optimization. Table 3 provides descriptive statistics over the baseline and the other two optimized order pickings at an aggregate level.

Table 4 compares travel distances and time in the traditional single-pick rounds, order-batching optimization, and order-batching optimization with an extra pick tunnel. Batch-picking optimization reduced the travel distance by 27.25% compared to the baseline single-pick rounds. The results of the order-batching examination show the importance of order batching in improving picking efficiency. An additional 13.91% reduced the travel distance with an extra pick tunnel. Moreover, the result of the simulation study shows the importance of and the relationship between order-batching optimization and cross aisles in improving routing flexibility and travel distance reduction. Regarding time, the batch-picking optimization gave a 22.82% reduction in travel time.

Implementing the system in the warehouse’s picking operations resulted in notable improvements in efficiency, specifically in terms of travel time and distance. To test the effectiveness of the AI-based system, Mann–Whitney U tests were conducted to compare the travel distances and times between the different picking methods. The choice of the Mann–Whitney U test was guided by the non-normal distribution of the data and the small sample sizes, as confirmed by the two-sample Kolmogorov–Smirnov test.

The median travel time difference between order-based pick rounds and order-batching optimization using AI was calculated as 00:19. Although a reduction in time was observed after the implementation, the change was not statistically significant (U = 252.50, p = 0.056). Hence, H₂ is not supported. However, a certain optimization level is achieved in time, despite not reaching acceptable statistical significance. The median travel distance difference between order-based pick rounds and order-batching optimization using AI is calculated as 37.91. AI implementation showed a significant reduction in terms of traveled distance (U = 177.00, p = 0.001). Hence, H_1a is supported.

In the ANOVA test, homogeneity in the variances of groups is expected, and it is checked using Levene’s test. If Levene’s test is non-significant (the p-value is greater than 0.05), the homogeneity of variance assumption is met (Kline 2011). In this case, the significance level of the F statistic produced by ANOVA checks the differences between groups (Field 2013). After finding the differences between groups, post-hoc multiple comparison tests can be used to determine which specific group differs significantly from the others. Bonferroni's and Tukey's tests both control the Type I error rate effectively but are considered conservative tests, meaning they may lack statistical power. Bonferroni's test is more powerful when the number of comparisons is small, while Tukey's test is more powerful when comparing a larger number of means. Bonferroni's test can be used effectively if the sample sizes of the groups are small and reasonably equal, and the homogeneity of variances is not violated (Field 2013). Therefore, Bonferroni's test is utilized for post-hoc tests to compare the groups. Bonferroni's test showed statistically significant differences in travel distance between group comparisons:

The analysis concluded that AI-based order-batching optimization with an additional pick tunnel (M = 182.35, SD = 31.96) and AI-based order-batching optimization (M = 211.81, SD = 33.88) resulted in statistically significant and lower travel distances compared to traditional single-pick rounds (M = 251.00, SD = 47.75). Therefore, hypothesis H1b is supported.

The results of the qualitative study, conducted using the Gioia approach, are illustrated in Tables 5, 6, and 7, highlighting the benefits, critical success factors, and challenges associated with adopting the AI-based system in the warehouse.

The study results reveal that the AI-based system provides several benefits for warehouse picking activities. One major benefit is the reduction of travel distance and time. Due to the optimal routing solutions offered by the AI-based system, pickers travel shorter distances and spend less time completing the same picking tasks, compared to the traditional approach without the AI-based system. Participants in the study acknowledged the benefits of optimized pick routes generated by the system. P1 states: “the system shortens the picking process a lot… as it shortens the road route of the vehicles”. P3 argues that: “The system made order picking faster. Things have gotten faster.” This paper’s quantitative results also corroborate the time and distance reduction.

Furthermore, the AI-based system has enabled improved documentation in the warehouse. Documentation is now more organized, with a reduction in the use of order list printouts and stationery. P1 emphasized that: “the system reduced documentation and consolidated locations.” Traditionally, each customer order had its own checklist and printouts, which Sienzi would compile to plan daily pickings. The AI-based system, however, combined orders and generated a checklist of batched items to pick. Compared to the traditional approach, the system produced significantly fewer checklists by batching single orders and reducing line space between suggested pick rounds.

Additionally, the AI-based system has simplified documentation. This reduction in complexity is partly due to a smaller number of checklists, as well as changes in the checklist format. Previously, checklists included only locations and products, leaving pickers to rely on their own judgment to determine the picking sequence. With the AI-based system, checklists now include suggested optimal routes for picking each item.

Another benefit is the positive environmental impact. The use of the AI-based system reduces energy consumption for the picking activity because the electric reach truck travels shorter distances due to batched picking and optimal routing. This approach replaces the previous order-by-order picking and manual routing methods. In this regard, P1 states that: “It has a positive impact (on environment) as it saves fuel and electricity……In short, I can say that it has a positive effect on fuel and electricity….” In addition, batched picking and optimal routing allow the truck to accomplish the same tasks with less use, which contributes to a longer lifespan for the truck and reduces the need for repair and replacement materials.

Concerning the environmental impact, Mr. Lijo mentions that: “there is saving of energy also… longer lifetime… to repair or replace the machine.” Moreover, P2 states that: “Saving time, picking orders faster and making the shipment faster affect both the customer and the warehouse in a positive way. Since the orders are picked faster, when we turn off the machine for 10 to 20 min, for example, there is no electricity consumption. As the time we do not use the machine increases, our consumption decreases.” This is consistent with the findings of De Giovanni (2021) and Gallo et al. (2023) regarding the positive environmental impact of AI on supply chains. Additionally, longer truck lifespans help reduce warehouse costs. Finally, reduced printout requirements in AI-based systems lower paper consumption, further benefiting the environment.

In some areas, empirical evidence has shown an increase in costs, which we have included in the challenges section of our study. However, in other areas, costs have decreased, which we have categorized as benefits. Specifically, the AI-based system in the warehouse has reduced time and distance in the picking process, resulting in lower energy costs for truck travel, reduced truck maintenance costs, and lower labor costs for this labor-intensive activity (Richards 2018). Additionally, improved documentation has led to lower stationery costs. The cost reduction benefit facilitated by the AI-based system and its systematic approach align with findings from previous studies (Choy et al. 2017; Pournader et al. 2021).

Furthermore, with optimal picking routes suggested in the checklist, pickers no longer need to consider the travel routes. Additionally, batch picking eliminates the need to revisit the same location multiple times for different orders. Together with reduced travel distances and time, these factors have contributed to a simpler order picking experience at the warehouse. In this regard, P3 states that: "it is better this way, and the complexity has been eliminated…we no longer search for the materials in the order one by one… The system made our work easier. We can easily find the materials on the shelf thanks to the system."

The higher order-picking simplicity has helped reduce errors and incorrect orders, leading to higher order-picking accuracy. P2 states that: “I can say that it affects positively, because when you do business in a non-systematic way, in old terms, in a chaotic way, there is a lot of room for error.” According to Setayesh et al. (2022), while pick errors can significantly impact operational success in manual picking, sufficient attention has not been given to enhancing pick quality. In this context, the AI-based system utilized in our case demonstrates fruitful results for improving picking accuracy. Our findings align with De Giovanni (2021), showing that AI implementation can reduce error rates and positively impact overall supply chain performance.

The final benefit is improved employee well-being. Completing tasks more quickly, accurately, and with simpler, more straightforward documentation has helped staff feel more comfortable. This improvement is particularly evident among pickers, who now experience less boredom from repeatedly visiting the same locations to pick items listed on different order lists. P1 emphasizes that: “the pickers feel more comfortable” and “… shortened picking time… satisfied the personnel.” Similarly, Grosse (2024) supports the idea that AI-based picking route optimization enhances employee well-being.

When the AI-based system was planned for integration into warehouse operations, a technological matching between the existing ERP system and the new AI-based system was necessary to establish the API. This necessity can be viewed as structural alignment, as suggested by Peng et al. (2021), involving architectural integration and interoperability between new and existing technologies. Select is responsible for integrating the ERP with the AI-based system. In this regard, Mr. Lijo states that: “the system warehouse is using its Warehouse Management System or ERP provided by Select, and to use the logic, they needed to use the API, or they needed to integrate API into their system. Otherwise, how do they use it? This is not a standalone solution, so they need to call the API and display it in their system for pickers to use the algorithm.”

Nevertheless, better communication between Kairos and Select could have reduced the time required for technological integration. During the project, Select experienced employee turnover, leading to a shortage of relevant personnel. Replacing these employees proved challenging, especially in finding suitable software engineers in Türkiye. As a result, Select struggled to maintain sufficient personnel for timely and effective communication during the technological integration. Although direct communication occurred at the business/managerial level, technical-level communication was largely lacking, which slowed the pace of integration– “… their engineer never talked to my engineer. I only talked to their manager.” (Mr. Lijo). Ultimately, the issue was resolved when proper communication took place. In this regard, Mr. Lijo explains: “…. they had to redo it, but then I guided them how to do it or how to use API properly, but, in any case, the second time, everything was OK.”

Operational alignment is another facet that complements structural alignment. It focuses on the flow of activities and processes involved (Maes et al. 2000; Peng et al. 2021). This alignment involves all three primary actors in the project—the AI-based system developer, the warehouse, and the ERP developer—emphasizing the importance of triadic collaboration. Furthermore, communication is a central element in achieving operational alignment (Luftman et al. 1999; Maes et al. 2000). In the long term, another essential aspect of operational alignment is ensuring the continuous, optimal functioning of the AI-based system in the warehouse, achievable through technological maintenance. Kairos has included maintenance service costs in its monthly subscription fee.

Despite its importance, operational alignment has encountered challenges related to changes in the documentation approach. “There was a process of reading and perceiving the documentation provided by the program. Because we switched to a different documentation from the documentation we were used to, that part was a bit difficult, if you recall…. Our flow of using the program has also changed. How we normally transcribe in the program differs from how we transcribe now. The answer given by the program is different.” (P1). Such challenges are, to an extent, natural, given that adaptation often accompanies change (Turienzo et al. 2024). Nevertheless, training is important for enhancing workforce usability of new technology (Büyüközkan and Göçer 2018). In this case, training supports understanding the new format of documents (checklists) generated by the AI-based system. The more effective the training, the easier it is to utilize the technology’s features (Davis et al. 1989; Zirar 2023). Conversely, as technology becomes easier to navigate, less intensive training is required. For this project, moderate training has been sufficient to enable effective use of the AI-based system. “In general, you need training for this. But the learning process was easy, there is nothing very challenging” (P3). Select was responsible for training Sienzi’s employees on using the system– specifically, on interpreting the new checklists generated by the AI-based system. This responsibility stemmed from Select and Sienzi's long-term relationship, their proximity within the same city, and Select’s role as the first to receive and communicate the new document format from Kairos after the setup process. Select conducted training sessions for Sienzi employees, who then further developed their skills through practice. Subsequently, new hires at Sienzi gained proficiency by learning from experienced colleagues. This reflects effective communication and triadic collaboration among the three parties to meet training needs and address the complexities introduced by changes in documentation.

Another critical factor in the success of the AI-based system in the warehouse pertains to strategic alignment, which focuses on ensuring that the adopted technology aligns with the broader goals and strategies of the business across different sectors (Peng et al. 2021). As cited in Gerow et al. (2014), Chandler (1962) defines strategy as determining an enterprise’s mission/goals and adopting actions to achieve them. In this context, strategic alignment addresses the match between adopted/adopting technology and the enterprise’s goals that it seeks to serve.

Henderson and Venkatraman (1999) contend that strategic alignment involves two domains: strategic fit and functional integration. Strategic fit includes business strategy and IT strategy—that is, how IT strategy can enhance or shape the business strategy. Functional integration refers to the importance of maintaining coherence between technology’s capability and the organizational settings. In our case, although the strategic fit was not a major issue, there was a lack of coherence between the business-wide setting and technology’s capability, manifested specifically in sorting complications.

Traditionally, Sienzi was picking items order by order, unloading them at the collecting area (i.e., where the distribution company picks up the items), then labeling the unloaded products for the distributor to identify them and put them in the transportation vehicle. The AI-based system provides Sienzi with the picking sequence based on the proximity of the stored items' locations, and the system combines/batches orders from different customers. With these changes in the picking logic, the traditional labeling approach involving product labeling in the collecting area was no longer effective. This was because products were picked in a mixed manner, and the picker in the collecting area could not recognize each product in the pool of mixed products. Consequently, with this traditional approach, time was wasted on sorting each product and labeling it correctly, causing the sorting complication. According to P1: “However, in projects with shipments to different points, after joint collection (batching), they (parcels) need to be separated during shipment… We lose the time we save on the shelf while sorting during the loading process to the car in picks with different shipment destinations.” The AI-based system’s capability for batching aimed to achieve time saving. However, it resulted in the need for additional time and more space in the sorting phase, leading to incoherence and complications with functional integration. This can be attributed to Sienzi’s poor pre-implementation business review of possible technology-driven alterations in the warehousing process. Another causal factor is related to the lack of joint planning, which is beneficial in achieving a clearer picture of the wide impact of the AI-based system on the activities and on the possible technological complementarity in tackling the unwanted sorting complication. Joint planning allows the perspectives of each side to be brought closer together, fostering collaboration and creating opportunities to enhance the configuration of the technology for a better fit with the business. In this regard, Mr. Lijo states that: “Software-wise, we can rearrange the sorting and that is certainly what I meant. We could do it before doing the batching, which would be slightly less improved…. So, this is the part I was saying upon the integration of process; if we did discuss how they would like to do it, including the possibilities that are in the system, we could have implemented it and to get a result. Right? And this was the missing part, but if you understand and design the system that is optimal for it….” The third element pertains to the lack of robust communication from Sienzi's side. Sienzi could have informed Kairos of the operational steps throughout the picking process, which potentially could have foreshadowed the sorting complication. Sienzi perceived the poor communication and other deficiencies as partly connected to the latest global pandemic. As the project coincided with the COVID-19 pandemic, Sienzi was hit with hectic work routines. First, the pandemic led storing companies to want to keep their stocks in the warehouse because their usual demand was affected, intensifying warehouse operations. Then, employees were contracting COVID-19, putting pressure on other staff to fulfill outstanding tasks. Moreover, they could not dismiss staff due to regulations that imposed financial restrictions on hiring new personnel. They were busy struggling to create an effective schedule and, hence, experienced a shortage of time. All these factors contributed to poor communication and pre-implementation weaknesses. The fourth attributing element is related to the limited understanding by Kairos of Sienzi’s operational dynamics. Kairos has customized a system to improve the picking process in Sienzi’s warehouse. This customization is associated with comprehending various contextual elements in Sienzi’s warehouse, such as the frequency and setup of customer orders, facility layout planning, and the existing technological environment. The operational procedure specific to Sienzi was among the topics that Kairos could have sufficiently understood, which would have led to foreseeing the sorting complication.

To address the issue, Sienzi later changed the labeling approach to reduce time wastage in the collecting area. In this new approach, Sienzi labeled the products while picking (the labels include the product order number). This adjustment ensured that products picked in batches arrived at the collecting area with labels already attached, thereby speeding up the sorting process. This can be included in business process re-engineering (Patrucco et al. 2020), which involves modifying the picking and sorting activities that have contributed to resolving the sorting complication.

Regarding re-engineering and adjusting the business for technological adoption, Sienzi had the possibility of adding an extra tunnel to the warehouse’s layout, which would have allowed system-driven route optimizations to provide higher yields. Nevertheless, it refused to do so. Two challenging elements are associated with this refusal: (i) costs; (ii) resistance to change.

Concerning the first element, there are costs associated with business adjustment that Sienzi could have incurred, reducing the attractiveness of making changes. The construction of a tunnel for route optimization costs both time and money. A tunnel, though well located, occupies space, and Sienzi argues that space is of critical importance. In some projects, it needs to maximize space utilization given the large number of products stored in the warehouse. It contends that, in those projects, losing storage space would cause economic loss. In addition, there are implementation and subscription fees associated with the AI-based system. These costs add to the overall list of expenses that need to be considered by the company.

Moreover, this refusal may be affected by the degree of resistance to change. The chaotic period of the pandemic and its impact on post-pandemic daily workloads caused staff members to experience fatigue and excessive stress. Consequently, they is an inclination to resist any alteration to their work routines, including initiatives such as building an additional tunnel on the warehouse site.

In the initial phase following the introduction of AI-based order picking, another instance of resistance emerged. Despite ultimately recognizing the benefits of the AI-based system for their well-being, initially pickers were reluctant to use it. This was attributed to the system's detailed instructions on the routing process and picking sequence, leading employees to feel that their freedom in picking was now constrained, whereas before they could decide on the traveling routes and picking sequences themselves. This reluctance among pickers hinders the adoption of AI systems for organizations due to certain fears such as loss of job, autonomy and control, which may threaten potential long-term benefits of such technologies (Zirar 2023).

Given a strong commitment and will at managerial level, many barriers can be overcome more effectively by allocating more resources. Mr. Lijo explains: “They did not look even in the basics. I mean, it’s very easy to see the way they are operating in the printout and so on…. it’s like the 90 s or 80 s. It’s very basic. So, there are many, many more things you could do, but it also comes with the cost of rearranging everything and stopping their operation and all. You heard the guy was saying he was even reluctant to put the tunnel because, you know, he loses the space there…. So it’s just a matter of thinking holistically and being able to accept the changes. So, I mean the optimization can do it in many ways. The structural change, or item location or even process change, there are many ways to do the thing.” The importance of considering costs associated with AI-based systems, as well as the role of managerial willingness, has been highlighted in the pertinent literature (see Merhi 2023).

Furthermore, there is the critical success factor of data privacy. As customer orders are now accessible to Kairos, there is a need for permission to share data and to consider privacy issues. There is also a confidentiality agreement with customers. In this regard, P1 explains: “We had to share some documentation to make the system usable. We had to let you see the products of various companies, we had to share some information. We did this with the permission of our companies.” In addition, data privacy-related matters are dealt with according to government and corporation regulations: “We work within the scope of the law on the use of personal data.” (P2). Despite the importance and challenging nature of privacy and security in the realm of Industry 4.0 and advanced technologies at supply chain level (Szozda 2017; Luthra and Mangla 2018), in our case, there seems to be no significant complications.

In summary, this study identifies several benefits of the AI-based system for improving the warehousing process. Additionally, a set of critical success factors, along with associated challenges, is identified. While this section has outlined the benefits, challenges, and critical success factors along with supporting arguments, the following section synthesizes these results into integrated insights by discussing their multifaceted components.

Utilizing the possibilities inherent in AI-driven solutions can be valuable to achieve operational excellence in the warehouses (see, Secundo et al. 2024). Several benefits identified in this study—travel distance and time reduction, improved documentation, positive environmental impact, cost reduction, higher order-picking simplicity, higher order-picking accuracy, and improved employees’ well-being– demonstrate the potential for AI-based new technologies to improve warehouse operations (Mahroof 2019; Ali et al. 2023; Cano et al. 2023; Pasparakis et al. 2023). The benefits identified in this study can transform the warehousing process through operational refinement and help overcome a set of common logistical challenges faced in the supply chain networks utilizing data-driven optimization (Turienzo et al. 2024). Among these benefits, travel distance and time reduction were further examined using a quantitative approach.

Although time reduction was non-significant, it can be inferred that time could be estimated in relation to the distance and speed. This non-significant time reduction result does not undermine the outcome because reductions in distance can eventually contribute to time savings with the help of other improvements. The findings on reductions in travel distance and travel time suggest that the order-picking efficiency can be improved by AI-based batching optimization. These findings align with prior research emphasizing the importance of order batching and using AI to enhance picking efficiency (Kuhn et al. 2021; Cergibozan and Tasan 2022; Zhou et al. 2023).

The qualitative results highlighted additional benefits not captured through quantitative measurements. This study suggests that certain advantages contributed to the development of other benefits. In addition to routing optimization, improved documentation emerged as a significant benefit, leading to reduced paperwork complexity and volume. The combined effects of improved documentation and reduced travel distance/time, achieved through batching, helped to increase the simplicity of order picking and enhanced employee well-being. As a result of more efficient routing, simplified picking processes, improved documentation, and enhanced employee well-being, picking accuracy increased, and errors decreased. Furthermore, the reduction in time, distance, and paperwork, alongside improved picking accuracy, had positive environmental implications. Paper usage decreased, and machines or trucks were utilized less frequently, leading to lower fuel consumption and longer machines lifetime. Finally, costs were reduced due to decreased energy consumption from reduced truck travel, lower truck maintenance from extended machine lifetime, decreased labor costs due to faster completion, and reduced stationary costs with less paperwork. Hence, it can be argued that these benefits are interrelated, aligning with previous literature (see, for example, Frederico et al. 2020; Dolatabad et al. 2022; Cano et al. 2023).

From a TOE theory perspective, it is evident that all AI-based system-derived benefits generally target both internal and external environments due to their interconnected nature, where one benefit can play a key role in the creation of another (Schwaeke et al.. 2025). However, in terms of the direct impact, cost reduction, improved documentation, higher order-picking simplicity, and enhanced employee well-being are primarily internal benefits because they largely affect the inside of the warehouse. These benefits focus on improving internal processes, reducing costs, and enhancing working conditions within the warehouse. Conversely, the positive environmental impact primarily pertains to the external context. This benefit aligns with overall sustainability goals and potential compliance with environmental laws and regulations established by governing bodies, and it highlights the broader external implications and consequences of AI-based systems. For benefits such as reduced travel distance and time, along with higher order-picking accuracy, it can be affirmed that these have effects in both the internal and external contexts. Internally, these benefits positively impact picking activity, performance, and efficiency within the warehouse. Simultaneously, they enhance the external context by enabling more reliable and higher-quality customer service.

In the TOE theory, the technological context refers to the characteristics and attributes of the technology itself (Wang et al. 2010; Abed 2020). For AI-based systems in Sienzi operations, the technological context includes elements such as the perceived benefits that the system brings to the warehouse. Leveraging AI advancements provides the warehouse with a range of benefits that positively impact various aspects of its operations (Turienzo et al. 2024). These benefits include cost reduction, improved documentation, enhanced employee well-being, reduced travel distance and time, simpler and more accurate order picking, and positive environmental impacts. In this case, all these benefits are made possible by advancements in AI technology. These findings align with Sodiya et al. (2024) regarding AI’s role in warehouse automation.

It is important to note that identified challenges in this study can be considered as critical success factors in the broader context of AI technology adoption (Marchet et al. 2015; Lee et al. 2018; Custodio and Machado 2020; Peng et al. 2021). If these challenges are effectively addressed, they can substantially enhance the opportunity to achieve the targeted outcome driven by the AI-based system. Challenges are a set of critical success factors that have faced complications throughout the process. Thus, we propose that strategic alignment, structural alignment, operational alignment, data privacy, technology usability, costs considerations, and degree of resistance or willingness to change are all critical success factors. However, strategic, structural, and operational alignments have faced specific barriers within their subsets. These barriers, along with cost considerations and resistance to change, account for challenges that have hindered the path toward optimal utilization of the AI-based system.

This study provided strong evidence for the critical role of strategic alignment in AI-based system adoption, particularly through strategic fit, functional integration, and business process re-engineering. Strategic fit centers on the implications of IT strategy for business strategy (Henderson and Venkatraman 1999) demonstrating, in our case, a reliable match between what the warehouse demanded from such an adoption and what the technology has delivered. The AI-based system successfully aligned with the warehouse's broader strategic goals. This alignment illustrates how technology's capabilities reinforce strategic objectives, which is particularly evident in the improvements related to order = picking efficiency and accuracy, as discussed previously. However, functional integration encountered notable complications. Functional integration refers to maintaining coherence between technology’s capability and the organizational setting (Henderson and Venkatraman 1999). In our case, the batching capability enabled by the AI-based system combined with Sienzi’s traditional labeling approach underpinned the incoherence between two closely connected activities of the warehousing process: picking and sorting. This incoherence was resolved when Sienzi adjusted its labeling approach.

In this paper, we associate such incoherence with four elements that hindered the functional integration:

Moving on to the next point, structural alignment involves achieving architectural integration (Peng et al. 2021), which in our case refers to the alignment between the architectures of the warehouse’s existing technologies and the AI-based system, particularly in terms of technological interoperability. In this study, such interoperability was a key goal, specifically between the AI-based system and the warehouse’s ERP system. Without interoperability and the establishment of the API linking them, the AI-based system would have been unable to deliver the promised performance. To address this need, communication between Kairos and Select became vital for developing a suitable API to establish the required connection between the technologies. However, due to internal challenges within Select, direct technical communication between these two actors was lacking, resulting in delays in achieving integration.

Communication also plays a crucial role in operational alignment. This type of alignment centers on the coordinated flow of operations to achieve success (Maes et al. 2000; Peng et al. 2021), which refers to maximum utilization of the AI-based system for Sienzi’s picking activity. To achieve operational alignment, effective communication and collaboration are essential between Sienzi, Kairos, and Select. In addition, technological maintenance is critical for sustaining operational alignment over time. The results indicate difficulties in understanding the new checklist generated by the AI-based system, highlighting a challenge stemming from changes in the documentation approach. This was revealed because the new checklists differed in format from Sienzi’s traditional checklists. To address this challenge, the triadic collaboration of Kairos, Sienzi, and Select was instrumental in providing the necessary training and education to help the pickers to successfully read and understand the new checklists.

In general, technology-business alignment focuses on sustaining a symbiotic relationship between business and technology to generate competitive benefits (Duffy 2002, as cited in De Haes and Van Grembergen 2009). It is marked in the literature as important in achieving business goals (Luftman 2004; Alaeddini et al. 2017), and this paper is in line with previous literature on the importance of alignment. In addition, Peng et al.’s (2021) categorization of alignment into strategic, operational, and structural elements aligns well with the empirical evidence of this research. An important component of alignment is communication (Maes et al. 2000; Luftman 2004). In our case, communication was a crucial factor in all three strategic, structural, and operational alignments. This supports Luftman et al. (1999), who state that all alignment elements are important, but none of them matter if effective communication is absent. In the case of structural and strategic alignment, communication has been impaired, leading to the emergence of challenges. However, to achieve operational alignment, proper communication helped in overcoming the relevant challenge.

Data privacy is the next critical success factor (Szozda 2017; Luthra and Mangla 2018; Sodiya et al. 2024). In this case, no issues related to data privacy were identified. This may be due to three data privacy concerns addressed by the actors involved. First, permission was obtained from the parties involved to share the generated data, and it appears that an established mechanism was in place for this purpose. Second, a confidentiality agreement was approved by partners, and a non-disclosure agreement (NDA) was signed. Third, privacy matters were managed in accordance with established regulations and laws.

A specific challenge concerned the issue of costs. Analysis and discussion of benefits revealed how the AI-based system contributed to cost reduction. Nevertheless, results indicate that certain costs arise, related to warehouse expenses, which can offset the cost savings achieved. First, using the AI-based system incurs subscription fees, which Sienzi must bear in addition to the implementation fee. For continuous use of the AI-based system, Sienzi has to pay a monthly subscription fee. Second, the warehouse incurs costs associated with adapting its operations to maximize the system’s benefits. These results highlight the importance of balancing initial and setup costs with achieved cost savings to create a cost-effective warehouse management strategy (Kalkha et al. 2024). As previously discussed, simulation results indicate that adding an extra tunnel in the warehouse can lead to a further reduction of 13.91% in travel distance. However, this tunnel was not implemented due to the additional costs and potential economic loss involved in configuring such a layout.

Another challenge noted in this study is resistance to change. Throughout the project, instances of resistance emerged regarding both the tunnel construction and the new approach of fully instructed routing. The resistance to the former arose from employee fatigue due to heavy workloads, making them reluctant to embrace change in their daily work routine. The resistance to the latter stemmed from an initial perception that the instructed routing approach limited their freedom in the picking process. Similarly, the previous literature has emphasized the impact of socio-technical factors on resistance to change (Hagström et al. 2023).

Finally, managerial commitment was identified as another critical success factor in this study. Management commitment is an important element that can significantly contribute to change management (Kumar et al. 2021). In this case, a firm commitment from Sienzi's management to undergo technological and process transformations in their warehouse was observed. Therefore, this element is not regarded as a challenge. However, in the case of tunnel construction, a more robust commitment might have motivated the management to conduct an inquiry. Such inquiry could have focused on identifying a suitable approach to construction, optimizing space utilization, and analyzing associated costs and benefits from both short- and long-term perspectives before deciding against building the tunnel.

From the perspective of TOE theory, strategic, operational, and structural alignments address both internal and external contexts due to the involvement of at least two parties. These alignments also target the technological context. Specifically, strategic alignment includes technological attributes that support business strategies, operational alignment involves elements such as technological maintenance, and structural alignment focuses on the interoperability of technologies, highlighting their technological characteristics.

Data privacy spans the internal, external, and technological contexts. Since data flow is inherent to the AI-based system, its relevant privacy considerations extend beyond the boundaries of a single party. Adjustment costs are specific to the internal context, depending on the unique characteristics of the warehouse. Implementation and subscription costs extend from the internal context into the external context. This is because the implementation and subscription costs require engagement with an external party—the AI-based system provider—and depend on its pricing model to provide the technology. While some costs relate to fees for utilizing the technology, they do not directly stem from the technology itself, placing them outside the technological context. Resistance to change also relates to the internal context because it mainly centers on the internal situation of the warehouse. Moreover, resistance to change relates to the technological context due to socio-technical motivators of resistance. Lastly, managerial commitment is relevant to the internal context. The main findings of the study discussed in this section are summarized in Fig. 2.

Full size image

By categorizing the challenges and critical success factors into technological, internal, and external contexts, these findings align with the TOE theory framework, illustrating how various factors impact the adoption and effective use of the AI-based system in Sienzi.