Requisites for applying adaptive control of safety lead time

Interesting because: To be effective in safeguarding production and delivery capability while remaining cost-efficient in today’s competitive situation, safety buffer sizes must swiftly adapt to variations in supply and demand. This is a fundamental manufacturing challenge, but no approach is effective until it is applied. Applying an adaptive control approach in practice requires managerial and organizational aspects or requisites that are not always in place in companies. An interactive, longitudinal research method enables new insights into the requisites for applying safety lead time, bridging the gap between theory and practice.

Theoretical value: Given the challenges in how safety buffers are commonly controlled, research gaps in adaptive control of safety buffers and a lack of practical studies using real-world data, this study aims to explore the requisites for applying adaptive control of safety lead time in manufacturing companies. The framework of requisites complements the literature that focuses on the control approach in itself. Extending the performance management literature to safety buffer control contributes to understanding and terminology in this area. A new combined metric – safety buffer utilization rate – with pedagogical properties to support the application is proposed.

Practical value: The framework, with provided examples, enables practitioners to understand how to apply adaptive control of safety lead time, increasing the potential real-world impact on delivery capability and cost. Interlinked requisites for understanding cause-and-effect relationships, metrics and measuring, balancing performance variables and organizing for adaptive control are presented. There are several steps managers can take to ensure that requisites are addressed both proactively and reactively.

This paper addresses the complex production planning process in manufacturing companies, which require decision models to make informed decisions (Farooq and O’Brien, 2015) regarding the control of safety buffers. These decisions are based on the size of a company’s additional resources for protection against stochastic variations, which are increasingly driven by the competitive environment with increased focus on delivery capability and cost-efficiency. Companies that can capture changing environmental conditions and make well-informed decisions, by reconfiguring their resources more effectively, maintain their competitive advantage (Keskin et al., 2022). To manage variations, a company needs the capability to sense and respond effectively to sustain high performance (Park et al., 2023). Zhu et al. (2024) demonstrate the importance of flexible stock in achieving optimal stock levels, thereby enhancing financial performance and productivity. The decided size of safety buffers affects important performance variables such as delivery capability (e.g. delivery precision or service level), capacity utilization and tied-up capital (Hedvall et al., 2023). The trade-off between protecting the flow and the cost of protection must be carefully managed (e.g. Stockman et al., 2021; Mattsson et al., 2020).

Safety buffers exist in various locations and types. A safety buffer may be located inbound, in process, or outbound. The safety buffer’s type of protective means may refer to material buffers like safety stock, capacity buffers like safety capacity and lead time buffers like safety lead time, as conceptualized by Hedvall et al. (2025). The size of safety buffers can be decided using calculations, judgments or a combination of both, as outlined by Jonsson and Mattsson (2016). Common in these cases is that variation and uncertainty in supply and/or demand are estimated and hedged by allocating resources to eliminate negative effects beyond acceptable levels, for example, in delivery precision. Delivery precision is here defined as the share of order lines delivered at confirmed delivery time and expressed as a percentage of total delivered order lines (Mattsson et al., 2020).

Both calculation and judgmental methods have challenges. Calculation methods are usually based on simplified assumptions, such as inventory withdrawals of products with normal distribution and stationary standard deviation, making them less suitable to apply in a complex reality (Mattsson et al., 2020). Many companies do not use calculations when dimensioning safety buffers, meaning that decision-makers judgmentally decide on its size (Jonsson and Mattsson, 2016). There is a component of manual decision-making in both judgmental and calculation methods, making decision-making individual-dependent and difficult to institutionalize (Calabretta et al., 2017). A challenge with judgmental methods is that they are loosely connected to planning environment conditions and data (Jonsson and Mattsson, 2008). Another challenge is the difficulty of observing cause-and-effect relationships between safety buffer size and performance (Hedvall et al., 2023). Also, both calculation and judgmental methods lack a clear connection to when and how to change the safety buffer size, making the safety buffers poorly adapted to the current state (Kahneman, 2011).

Normally, the effort needed for analysis and decision-making in the planning process, given the often-numerous articles, tends to limit control of safety buffers to certain product categories and to low frequency of adaptive decisions. This is partly why the frequency of re-planning within the production planning process has been studied, for example, by use of real-time data in MTO environments (Woschank et al., 2024). The process seldom has a mechanism for regularly revising the safety buffer size. These issues tend to make decisions on planning parameters and safety buffers static. Few incentives exist to re-dimension, often leading to unnecessarily high safety buffers (Hedvall et al., 2023). Hedvall et al. (2025) further revealed high complexity with several safety buffers in different locations and of various types in each manufacturing company studied, reinforcing the need for better control and dimensioning. Thus, there are several challenges and complexities to address, requiring decisions to be made within an appropriate process.

To overcome the challenges and complexities outlined above, an adaptive control approach to safety buffers (Mattsson et al., 2020) is applied. It is an approach to controlling problems in engineering, both in practice and in research, and is often applied to the control of complex manufacturing processes. Some examples also relate to the control of material flows, e.g. dynamic control of kanban cards (Xanthopoulos et al., 2018) and flow shop routing (Peng and McFarlane, 2004). Only a few studies on the adaptive control of safety buffers have been found in the literature, such as Mattsson et al. (2020). Research on dimensioning safety buffers is recently criticized for being far away from practice (Hedvall et al., 2023), as it often strives to advance calculation methods without examining their applicability (Jonsson and Mattsson, 2019). Surprisingly few studies have verified such calculations with real-world data (Gonçalves et al., 2020). This indicates a need to move beyond the conceptual idea of adaptive safety buffer control and its planning techniques, by examining requisites – practical aspects to consider – for effective adaptive control of safety buffers in manufacturing. This means that the paper’s focus is not on the adaptive control models as such. Instead, the study is positioned within the design of the production planning and control process, specifically the decision-making and updating of planning parameters. To the best of our understanding, no studies focus on identifying requisites for adaptive control of safety buffers, which would facilitate the development and application of such a process. This paper is limited to safety lead time, representing the additional time beyond the expected lead time to absorb variation and thereby safeguard delivery capability. In practice, manufacturing companies order materials or plan to start manufacturing earlier than otherwise expected (Hedvall et al., 2025). Safety lead times are much less studied than, for example, safety stocks. Therefore, the purpose of this study is to explore the requisites for applying adaptive control of safety lead time in manufacturing companies.

The purpose is addressed through a multiple-case study involving four companies. The paper contributes by describing their first years of effort in applying adaptive control to safety lead time. In relation to Hedvall et al. (2025), our study represents an application-oriented continuation that shifts attention to how one selected safety buffer type can be adaptively controlled. By deriving a framework of propositions outlining the requisites for such an application, our study extends the earlier framework of challenges in dimensioning safety buffers (Hedvall et al., 2023) in two ways. First, it narrows and deepens the focus to one seldom-studied type of safety buffer: safety lead time. Second, it narrows and deepens the focus to one seldom-studied dimensioning method: adaptive control.

The paper’s continuation is structured as follows. A frame of reference is introduced. The methodology is presented, followed by an empirical within-case analysis. A cross-case analysis is provided, generating propositions on requisites for applying adaptive control of safety lead time, organized into a framework. Reflections and contributions are discussed. Concluding remarks, limitations and suggestions for future research are presented.

This section describes adaptive control and its application to safety buffers and outlines four categories of requisites that represent the backbone of the study.

The general meaning of adaptive control is that it relies on a decision-maker who can modify behavior in response to changes in the process dynamics and the nature of the disturbances (Åström and Wittenmark, 2008). This approach is based on two key concepts. The first is “control,” which refers to the ability to handle different types of variation while maintaining output control. The second is “adaptive” and complementary to control, since it refers not to being constrained by earlier assumptions but to adapting to changing preconditions (Åström and Wittenmark, 2008). Adaptive control, therefore, enables the application of an iteratively updated control approach. According to Mattsson et al. (2020), adaptive control of safety buffers builds on a decision-maker comparing target and actual performance of the controlled system, as input to updating the safety buffer size.

By periodically measuring the selected performance variable of the safety buffer, actual performance provides feedback on how well the safety buffer protects against variations, thereby contributing to achieving target performance. When actual performance deviates from target performance, a decision is made on whether to adapt the safety buffer size, for example, by revising decision parameters. Mattsson et al. (2020) provide an example: if the differences between actual and target performance fall outside a specified tolerance range, the value of the decision parameter is revised. If the “number of days” calculation method is used for controlling safety stock, with a target performance for service level of 96% and actual performance measured to be 93%, the earlier decided number of days is increased and a larger safety stock size is dimensioned. In the subsequent period, a new comparison is performed, and the safety buffer size may be updated iteratively.

A framework should provide guidance. The beginning should be to pre-arrange categories in which requisites are expected to emerge, based on the literature (Steenhuis, 2015). Four categories of requisites for the practical application of adaptive control of safety buffers can be identified in the limited literature. Adaptive control builds upon understanding cause-and-effect relationships when the company adapts to changing conditions (causes), resulting in performance effects (e.g. Zhu et al., 2024; Keskin et al., 2022; Park et al., 2023). That can be when decisions are made, and the corresponding effect of the decision is fed back and studied. Adaptive control can leverage managers’ understanding of the mechanisms that improve companies’ delivery capability (Keskin et al., 2022).

To capture desired performance effects, performance metrics and measuring are needed. According to Forslund and Jonsson (2007), performance measurement and management include activities such as selecting performance variables, defining metrics and setting targets, all of which involve decision-making (Bourne and Bourne, 2023), and feedback where target performance is compared to actual performance. In the context of safety buffers, selected delivery capability performance variables can be service level or delivery precision, to internal and external customers, costs and tied-up capital (Hedvall et al., 2023). Defining metrics is a critical activity that includes deciding measurement object (e.g. orders, individual articles, production groups), time period for the measurement (e.g. hour, day, week) and measurement point for where the measurement should take place in the flow (Forslund and Jonsson, 2007). Accurately defined metrics enable adaptation to changing conditions (Åström and Wittenmark, 2008).

However, a higher safety buffer implies some sacrifice, like increased tied-up capital or prolonged production throughput time (Mattsson et al., 2020; Hedvall et al., 2023). This makes balancing performance variables important. To achieve robust operation, a holistic approach that considers trade-offs among performance variables and required resources is needed (Stockman et al., 2021). The effect of a too-small safety buffer typically differs from that of a too-large safety buffer. Therefore, it might be challenging to identify a single performance metric that captures both of these cases. Two performance variables can better inform the decision-maker about whether to adapt a parameter. The primary performance variable can, for example, be service level, while the secondary performance variables can be metrics on tied-up capital or space limitations, to balance the safety provided with its costs (Hedvall et al., 2023).

In many cases, the cause-and-effect relationships in adaptive control involve various organizational functions (Hedvall et al., 2023). As an example, outbound delivery capability is related to both purchasing and production, implying the need for cross-functional engagement. This emphasizes the importance of effectively organizing for control.

This study’s research methodology is presented in terms of research approach, case selection, data collection and analysis. Research quality and rigor (as described by, e.g. Näslund et al., 2010) are discussed in each section. An overview of methodological steps is provided in Figure 1.

Given the limited literature, this study inductively generates knowledge from empirical data, thereby pursuing theory-building ambitions (Bryman and Bell, 2022). A qualitative and interactive research approach (Svensson et al., 2015) is considered an expedient methodological choice for this exploratory study. As interaction means that companies have developed and tested adaptive control of safety buffers by making iterations in a coaching dialogue with the researchers, it contains elements closely related to action research, a specific form of interactive case studies (Näslund et al., 2010). Action research can simultaneously contribute to practice and research (Eltantawy et al., 2015), seeks to engage with complex contexts to discover practical solutions and fits well with applied topics (Stringer and Aragón, 2021), such as this study. The research approach was enabled by unique access to the companies and by reciprocal trust between companies and researchers (Eltantawy et al., 2015). It was carried out in a two-year research project, established through long-term participation in earlier interactive research projects with the same researchers and companies. Therefore, risks in interactive and action research, such as shallow participation, untrained researchers and short time (Näslund et al., 2010), could be avoided. The type of case study could further be categorized as planned (all cases were planned and selected early in the project), pragmatic case study with iterative components (data were collected and analyzed throughout the study, with insights leading to additional data collection), in line with Steenhuis (2015).

The study included three large (>249 employees) and one medium-sized manufacturing company from four different industries in Sweden. Such a broad set of cases was selected for its potential to fill conceptual categories (Steenhuis, 2015). All four companies (CA, CB, CC and CD) use a make-to-order (MTO) production strategy. CA, CC and CD have partly engineer-to-order (ETO) and CB has some products with make-to-stock (MTS) production strategy. Hence, the companies represent different types of flows. The researchers actively suggested which safety lead-time buffers the companies should include to cover complementary locations: inbound safety lead time at CA and CB, given uncertain inbound lead times from external suppliers and process safety lead time at CC and CD, given uncertain inbound lead times from internal processes. Respondents representing various company functions covered necessary competence (Stringer and Aragón, 2021), a form of data triangulation to increase validity and rigor (Farooq and O’Brien, 2015; Näslund et al., 2010). The companies were represented by experienced staff well-educated in control of safety buffers. Company characteristics and respondents’ job titles (rephrased to ensure ethical handling and anonymity, in line with Bryman and Bell, 2022) are:

1. CA is a large, project-based manufacturing company with high product complexity. Respondents are Planning Specialist, Purchasing Manager, Logistician, Senior Project Manager and Project Manager.

2. CB is large and has batch manufacturing with low product complexity. Respondents are Operations Manager, Logistics Manager and Controller.

3. CC is medium-sized and has project-based manufacturing with medium product complexity. Respondents are Supply Chain Manager and Production Planner.

4. CD is large and has project-based manufacturing with low product complexity. Respondents are Supply Chain Planner and Senior Logistician.

Data were collected through interviews and coaching dialogues with each company, at joint workshops and through extensive written field notes and reflections from the researchers. Informed consent was obtained from the participants (Stringer and Aragón, 2021), allowing them to validate and rephrase sensitive information, thereby ensuring ethical handling (Bryman and Bell, 2022). No ethics review board clearance was therefore necessary. The companies wrote their own records to remember details, place and time in the changes made, as Stringer and Aragón (2021) advised. These different sources of data enabled data triangulation to enhance validity (Farooq and O’Brien, 2015). Quantitative data in the form of measurement reports of, for example, safety buffer size and the primary performance variables were also collected. Lastly, investigator triangulation (Näslund et al., 2010) was applied, with four researchers, representing different educational backgrounds, genders and ages.

It was important to establish shared understanding and terminology to ensure construct validity and reliability (Näslund et al., 2010). This was done in two lectures. Iterative data collection during workshops 3–6 was based on a template (see Figure 1); hence, increased credibility was sought by collecting the same data across all cases and over time. Coaching from the researchers was necessary, as the companies had difficulty getting started, facing, e.g. a lack of measurement data, weaknesses in their ERP systems, or management not allowing changes. Such information requires deep access and trust to share with researchers.

The unit of analysis is safety lead time embedded within manufacturing companies. Four authors independently coded the raw data (i.e. first-order concepts) into second-order themes or requisites to achieve triangulation (Bryman and Bell, 2022). The analysis was conducted first with a within-case, then a cross-case approach incorporating pattern-matching elements (Bryman and Bell, 2022). For the researchers, it was important to stay close to the literature in the cross-case analysis to ensure both validity and reliability (Näslund et al., 2010). Good conditions for high internal validity (Bryman and Bell, 2022) were established by the robustness of adaptive control in establishing cause-and-effect relationships. The cross-case analysis included conceptualization of propositions regarding requisites identified in at least three cases. This approach has the potential to establish an empirically confirmed literature base (Svensson et al., 2015). A fifth researcher joined the article-writing phase. This reflexivity from an outsider’s perspective can potentially reduce insider bias, thereby increasing the validity and robustness of qualitative research (Finefter-Rosenbluh, 2017). The respondents approved the texts before publication.

This section explores how adaptive control was applied in each case, followed by how the requisites were addressed.

CA applied adaptive control of inbound safety lead time, with the Logistician being the main decision-maker. The inbound lead times ranged from 7 to 40 days. Based on a top-management decision, the inbound safety lead time was set to 20 days for most articles. Their buffer was limited to 135 article numbers from one supplier. CA used inbound delivery precision from suppliers as the primary performance variable to determine safety lead time.

At the outset, CA had no established relationship between inbound safety lead time and risk of stock-out to production. The purchasing department had recently become responsible for setting all lead times. As inaccurate safety lead times would adversely affect purchasing’s functional metrics, purchasing decided to increase the inbound safety lead time to 20 days. As stock-outs were rare, the logistics function perceived that the safety lead times were excessively long. Understanding the cause-and-effect relationship is developed by measuring buffer utilization rate as described below. When the project was completed, the Senior Project Manager stated, “We have built a totally different understanding of cause-and-effect that we now can prove.”

The increased safety lead time required the logistics function to demonstrate that inbound safety lead times were excessive. They therefore defined a new performance metric, called the safety buffer utilization rate, after discussions with the researchers. It was defined as Y/X, where X is the safety lead time of an article and Y is the utilized safety lead time for that article, calculated as the mean number of days of delay of articles arriving after the confirmed delivery date during a month. The first measurement showed that the mean utilization rate (Y/X) was 1.86% across the articles, confirming the perception of excessive safety buffering. In the subsequent measurement, Y/X was 7.4%, but in the context of the Russia-Ukraine war, which had led to a global materials shortage. CA also redefined the metric from a monthly to a weekly time period to avoid hiding variations, as they had previously aggregated the data too much. Not only were metrics adapted, but according to the Senior Project Manager, “we have also learned the importance of keeping parameters updated… for large article groups.” Also, CA learned that to create acceptance for new metrics, “we must be able to explain it well. Otherwise, the metric will be a shelf-warmer.”

The low safety buffer utilization rate indicated high delivery capability, but high costs and tied-up capital. “We have extremely high costs for lost sales, including delivery penalties; we cannot have late deliveries. Our management did, however, not see the cost for safety lead time, but the new metric has been their eye-opener”. CA struggled to establish guidelines for adapting the safety lead time based on utilization rates. They established a guideline stating that a safety buffer utilization rate of 0–20% should lead to reduced safety lead time, that 20–50% meant that the safety lead time could remain and that a utilization rate above 50% should be investigated to determine if increasing safety lead time was needed.

Increased cross-functional collaboration between logistics and purchasing was necessary, meaning challenging but fruitful discussions. “No single function had the knowledge to do it independently.” The safety buffer utilization metric highlighted the need to discuss reducing long safety lead times and served as a common basis. Aspects like determining the accurate safety lead time target were discussed.

CB investigated adaptive control for its inbound safety lead time, limited to MTS products, initially with the Operations Manager as the decision-maker and later also the Logistics Manager. At the start of the study, the safety lead time was two days, and the target performance for the primary performance variable, outbound delivery precision, was above 96%. The actual performance was 86%, which was largely accepted given the global situation, in which most suppliers had prolonged lead times.

CB believed that the inbound lead time had to be increased, but struggled to determine how much to increase the lead time, respectively, the safety lead time. Two days’ safety lead time were added. They did not know the causes of the low outbound delivery precision. It could be supplier-related, self-induced or planning-related, and the internally held truth was that suppliers, or the purchasing function were to blame. “The suppliers were protected by our safety buffers”. By the end of the project, after, e.g. the introduction of new metrics (see below), problem-solving is much more proactive, and the Operations Manager confessed: “Formerly, we just increased our buffers as soon as we had stock-out situations”.

CB recognized that the current measuring of outbound delivery precision did not facilitate understanding of cause-and-effect relationships. “We had an outbound performance variable for evaluating an inbound safety buffer.” Thus, a new measurement point was required at the planned start of each production order, as compared to the production plan, to define a new metric. As the current ERP system could not provide the necessary information, they initiated changes with their ERP system supplier. After ten months, it became possible to measure adherence to the production plan. Each production order that deviated from the production plan was assigned a reason code (e.g. lack of purchased material, lack of capacity or internal replanning), enabling deeper root-cause analysis. The results of this new metric surprised both managers, revealing that most causes of moved production orders were internal re-planning of production. According to the Logistics Manager, they made the mistake of not involving all affected employees as much as needed, resulting in low acceptance and adherence to entering reason codes in the ERP system.

Analysis of metrics and reason codes revealed that many articles had both an inbound safety lead time and an inbound safety stock. The Operations Manager emphasized: “We only focus on problems: problem suppliers, problems in production, and problems in delivery capability. We don’t look at what isn’t a problem, and maybe that’s what the problem is, that we have huge costs for buffering that cover all the problems.” At the end of the study, outbound delivery precision reached 96%. The Operations Manager explained, “Our telephones are now silent”. This situation, with potentially excessive delivery precision, has prompted the Operations Manager and Controller to consider changing the control model by iteratively increasing attention to a secondary performance variable, tied-up capital, to improve balance. “If we continuously monitor our costs, we may avoid safety buffer reduction campaigns. Buffer utilization rate is an intuitively good metric”.

The new metric and its insights led to the development of routines across several functions, including production planning, logistics and purchasing. “We had to discuss and solve our problems in production planning and purchasing, as the main cause of delays was our planning.”

At CC, the Supply Chain Manager was the decision-maker for the adaptive control of the process safety lead time. The performance variable was set as internal delivery precision from assembly. The actual performance was surprisingly low at 12%, compared to the target of 85%. The Supply Chain Manager initially assumed the size of their safety buffers would decrease, but realized an increase was necessary. They iterated from 2 to 5 days of safety lead time, which meant an earlier start of production. The increase was partially attributable to increased penalties for late customer deliveries.

Before the project, CC blamed the lack of material as the cause of production stops and low delivery performance. To establish a clearer cause-and-effect relationship between the size of the process safety lead time and delivery performance, a new metric was required (see below). According to the Supply Chain Manager, a key understanding was that “It was important that the researchers required us to map the current situation and report in all parts of their template”. This led to increased awareness, proactive problem-solving and several improvements. The Supply Chain Manager stated that the new metric enabled decisions based on actual data rather than perceptions, thereby improving cause-and-effect understanding.

CC began measuring every production-line stop and noted its cause, to learn about cause-and-effect. Although measured manually, this enabled the Production Planner to identify problem areas. “We learned that we almost never lack materials”. The monthly measurements were replaced with daily measurements, enabling proactive problem-solving rather than a reactive “wait and see” approach. CC realized that its ERP system was outdated and no longer supported its measurement requirements. The acquisition of a new ERP system was initiated.

The delivery precision from assembly increased from 12% to 50%, 70% and finally 97% after increasing and adapting process safety lead time and after introducing the new metric. The target was 95%, but actual performance above the target prompted discussions about reducing the process safety lead time from 5 to 3 days to account for the cost of tied-up capital (secondary performance variable).

The Supply Chain Manager initiated frequent cross-functional meetings (including purchasing, production and planning) to agree on how to manage the delivery precision from assembly. The introduction of a structured cross-functional collaboration stopped the “we versus them” behavior and has been “revolutionary”.

The Supply Chain Planner was the decision-maker for applying adaptive control to process safety lead time between materials kitting and welding. The selected performance variable was the service level to welding. The study was limited to seven in-house-produced articles that were merged into a welding kit. At the start, they had a safety lead time of 6 days, meaning that the kit waited for welding for 6 days if no delays occurred in producing any of the articles.

CD did not have a satisfactory understanding of the relationship between its safety lead time and service level to welding, but realized the urgent need for such an understanding. “There are concerns about low delivery capability, but we do not know what causes it,” according to the Senior Logistician. Limited power to make changes to the safety lead time without top management approval resulted in no changes being made. However, insights were gained, and new metrics were sought to provide the information needed. “We must develop this approach”.

As the ERP system could not provide any useful metric of the selected performance variable – service level to welding – a new metric had to be defined. “What we call cover time must build upon data from the ERP system”. Its target was 75%, but the actual performance was 58%, due to inbound material lead times of up to 80 weeks related to the Russia–Ukraine war. Setting up a new metric took a lot of time. The Supply Chain Planner emphasized, “It took a lot of time to define the right metrics that suit the company and are understandable” due to several changes to the definition and the need to manually convert values in Excel. “We also had a too low measurement frequency for capturing stock level variations; we should consider daily measuring instead of weekly.”

“Our management is scared to death of low delivery capability; we are, therefore, probably too fat.” CD realized that its new metric did not provide sufficient information for controlling the safety lead time. A safety buffer utilization rate metric was added based on discussions with the other companies. Finally, a metric was established in which the service-level probabilities for the seven articles in the kit were combined into a single metric for the welding kit. The safety buffers were, on average, utilized at 50% (20–80%), but the percentage increased over time. No targets were yet set, but the 50% utilization rate was intuitively satisfying. The buffer utilization rate for some articles ranged from 0% to 22%, indicating that safety lead times could be reduced.

Even though CD did not achieve much in its efforts to adaptively control safety lead time, it learned something about its organization. “There are concerns within many functions of the organization for low delivery capability,” and “the different functions fight, but we must be able to keep it all together.” Working cross-functionally will continue.

The deepened cross-case analysis is presented as nine propositions on requisites for practically applying adaptive control of safety lead time, related to the literature and arranged into the empirically confirmed categories of requisites into a framework.

The understanding of cause-and-effect relationships, as suggested by, e.g. Park et al. (2023) and Mattsson et al. (2020), was increased when companies applied adaptive control. The understanding was deepened and detailed into two types of cause-and-effect relationships: (1) the expected type between the dimensioned size of safety lead time and its performance effects (shown in earlier studies, e.g. Hedvall et al., 2023), but also the less expected type and (2) between causes generating the need for safety lead time and the dimensioned size. The latter is known from general quality management literature. All companies increased their understanding of type 1, though the calendar time required varied. Companies with inbound lead times expressed in days swiftly attained understanding, while one company with inbound lead times expressed in years needed much more time. This suggests that long lead times delay the expected attainment of cause-and-effect understanding. Altogether, the study reinforces the idea that adaptive control provides a clearer connection between dimensioning and performance and naturally incorporates cause-and-effect relationships. This feature is less prominent in common control of safety buffers (Hedvall et al., 2023), meaning the dimensioning is not adapted to the actual state (Kahneman, 2011).

Some cause-and-effect relationships of type 2 were enabled by additional routines for capturing cause codes for delays. Companies that started such routines discovered that most delays in the start of production orders were not due to a lack of material from suppliers, as they had assumed, but rather to internal causes. Hence, the organizational location of causes was surprising. Understanding cause-and-effect relationships has meant that they no longer respond reactively to perceived problems but also identify their causes, enabling proactive management. This finding was enabled by incorporating literature on performance management, which emphasizes identifying the causes of problems to achieve performance improvements (e.g. Bourne and Bourne, 2023; Forslund and Jonsson, 2007).

To acknowledge the cause-and-effect relationship, it is necessary to measure performance appropriately, in line with Bourne and Bourne (2023). A focus on measuring the primary performance variable – delivery capability – resulted in an increase in delivery capability. This reinforces the old saying, “what gets measured gets done” (e.g. Bourne and Bourne, 2023, p. 21). The companies selected and defined new metrics to prevent excessive safety lead time, a challenge due to the lack of such examples in the literature. Following the delivery service-based terminology and recommendations of Forslund and Jonsson (2007) when defining measurement object, measurement point and time period was valuable. The safety buffer utilization rate confirmed suspicions of excessive safety buffering, providing insight that the same delivery capability to production should be possible at a shorter lead time and lower cost (as unnecessary tied-up capital emerged during the unutilized safety lead time). This combined metric is also new to the literature. All companies soon introduced similar metrics, indicating the practical relevance and reported similar experiences with them.

Forslund and Jonsson (2007) discussed the details of defining metrics, such as time periods. In individual coaching with the companies, it was found that avoiding overly aggregated metrics was necessary. Metrics with long time periods (i.e. aggregated data, 6-month rolling averages) were considered to hide variation and reduce the ability to understand cause-and-effect relationships. This adds to the findings of Park et al. (2023) and Keskin et al. (2022) on the identification of variation that needs to be not only continuous but also swift.

Not surprisingly, all companies highlighted the importance of measuring the right things and in the right ways to obtain useful results. This implies accurately defining metrics, in terms of, e.g. measurement object and measurement point, as recommended by Forslund and Jonsson (2007). This performance management terminology added to the understanding of safety buffer control literature. It was seen as challenging to understand cause-and-effect relationships and a clear risk of distortion if the measurement point was not directly related to the safety buffer. With well-defined metrics, it becomes easier to see when a change in the safety lead time is needed (in line with Hedvall et al., 2023). Accurately defined metrics enable adaptation to changing conditions (Åström and Wittenmark, 2008). To exemplify from both inbound and process safety lead time, the inbound delivery capability of suppliers provides more useful information for controlling inbound safety lead time than the outbound delivery capability to customers, as the latter can also be affected by, for example, manufacturing performance. This realization led to changes in measurement points for both companies.

It has been important for companies to convey the necessary understanding to avoid sub-optimizations and prevent negative implications for the dimensioning of safety buffers (Hedvall et al., 2023). Companies had to iterate on the introduction of new metrics before everyone involved became sufficiently familiar with and accepted them. Examples were found in which everyone involved in measurement had initially not been informed of why and how the metric should be used, leading to incomplete data collection and the distortion of cause-and-effect relationships. This suggests that introducing new metrics is likely to take a long time.

After coming a long way and becoming more structured in applying adaptive control of safety buffers, companies realized that they focused on delivery capability at the expense of unnecessarily high tied-up capital, compromising the company’s competitive advantage (Keskin et al., 2022; Stockman et al., 2021). The need for additional metrics was discovered. The cost of tied-up capital was the most common secondary performance variable, a strategic, managerial restriction related to external stakeholders. A space limitation at a production unit was also identified as a secondary performance variable, being an operational, internal restriction that limited options for storing additional materials to support increased safety lead times. However, such secondary performance variables challenge the creation of a combined metric. These experiences highlight the need for not only using a primary performance variable but also to balance it with a secondary performance variable, extending the quantitative findings of Zhu et al. (2024) and Park et al. (2023) with qualitative, case-based data, to remain cost-effective.

The safety buffer utilization rate is a metric not identified in the literature. It therefore operationalizes and extends the suggestions of Zhu et al. (2024) and Park et al. (2023). In this study, applying such a metric revealed that safety buffers were larger than necessary. It was a valuable way to indicate the proper balance between delivery capability and cost for tied-up capital, captured in a single metric. The study has therefore concretized a central trade-off within logistics: the challenge of dimensioning safety buffers (Hedvall et al., 2023), which, in the context of controlling safety lead time, entails balancing delivery capability risk and cost risk.

Applying adaptive control of safety buffers can concern many parts of the flow and, hence, also several functions (Hedvall et al., 2023). New procedures of coding delay causes at planned production start meant identifying causes for variations in other functions than expected which, when mitigated, can reduce the need for safety buffers. The importance of cross-functional meetings involving logistics/supply chain management, planning, purchasing and production was recognized, over time leading to mutual understanding, less functional sub-optimization and increased collaboration. The gained understanding needs to be spread and anchored across the organizations. This shift resulted in organizational changes and new insights, such as finding common ground, and, in the end of the study, also involved sales, their forecasts and production plans, including decisions on manufacturing strategies.

A framework of propositions is presented in Table 1.

Given the limited literature in this area, this study has inductively generated knowledge from empirical data, thereby successively pursuing theory-building ambitions (Bryman and Bell, 2022). In what ways does a framework of requisites contribute to literature? Requisites themselves are not important, but acknowledging their implications for the planning process and decision-making in manufacturing companies is important for effectively dimensioning safety buffers, particularly safety lead time. As recognized by Keskin et al. (2022), competitive advantage can be achieved by making well-informed decisions that better capture and adapt to changing conditions. Change and variation are continuously part of the global landscape. Our study has shown one path to making well-informed decisions through adaptive safety buffer control. Flexible and optimized safety buffers, in line with the recent study of Zhu et al. (2024), can enhance financial performance and productivity, while balancing the cost of protection (e.g. Stockman et al., 2021). Our study extends the work of Park et al. (2023), showing a way to sense and respond to variations to sustain and enhance performance. The control approach in itself is not enough. It is also necessary to understand the requisites to reach these desired outcomes. Our systematic framework lays a foundation for that.

Our study builds on, e.g. Hedvall et al. (2023), who provided the main framework to relate to, showing common empirical challenges in the dimensioning of safety buffers across several types, and especially highlighting the weak understanding of cause-and-effect relationships and the difficulty of reducing excessive safety buffers. Hedvall et al. (2025) further developed the stream of literature on safety buffers by identifying and describing the selection of safety buffer types and sub-types in manufacturing companies. Our study extends and contributes to Hedvall et al. (2023) by specifying and focusing in depth on just one, seldom-studied type of safety buffer – safety lead time, and by focusing on just one, seldom-studied dimensioning method – the adaptive approach. Furthermore, it moves from identification and selection to application, specifying the practical requisites needed to apply adaptive safety buffer control. The literature on applying adaptive control of safety lead time is limited; one exception is Mattsson et al. (2020), which our study extends by acknowledging practical requisites for its application and implementation. Park et al. (2023), Keskin et al. (2022) and Zhu et al. (2024) stress adaptation, responsiveness and performance balance, where our study shows what these ideas require in practice. Understanding cause-and-effect relationships, metrics and measuring, balancing performance variables and organizing for control are important requisites when applying adaptive control of safety lead time. They are detailed and presented in a framework of propositions in Table 1, which is an initial step toward theory-building in the field of adaptive safety buffer control.

Another contribution to the literature is the combination of performance management literature with literature on safety buffer control, which is applied and has led to an expanded terminology and increased understanding. Metrics and measuring were identified as central requisites for adaptive control of safety lead time, as indicated by Mattsson et al. (2020). This was mirrored in the many propositions linked to metrics and measuring. Lastly, the development of a new combined metric, safety buffer utilization rate, that balances delivery capability risk and cost risk, is also new to the literature, to our knowledge, and was found to have appealing properties in revealing both too small and too large safety buffers. It can be calculated as the utilized number of days of the safety lead time divided by the safety lead time in days, aggregated into a meaningful measurement object, e.g. an article group or a welding kit, over a weekly or monthly time period, and expressed as a percentage. This means that our study extends the recent work of Zhu et al. (2024) by emphasizing that balancing delivery performance and cost can be operationalized using metrics.

We argue that a framework of requisites for applying adaptive safety lead time control makes several managerial contributions. It reflects the many challenges in how companies commonly approach safety buffers. Some of the major challenges identified in earlier research (Hedvall et al., 2023), such as difficulties in understanding cause-and-effect relationships, the lack of incentives to reduce safety buffers and the low updating frequency of planning parameters, could be addressed by applying an adaptive control approach. The identification of the buffer utilization rate metric, which balances performance variables such as delivery capability and cost into a single metric, was shown to have pedagogical merits and the potential to support managerial decision-making in the production planning process. This metric quickly spread between the studied companies, indicating its clear managerial relevance. Additionally, a cross-disciplinary approach to safety buffer control offers significant advantages from an overall performance perspective. In the long run, adaptive control can help reduce the gap between theory and its practical application (Hedvall et al., 2023; Gonçalves et al., 2020; Jonsson and Mattsson, 2019). The study also provided empirical examples of applying adaptive control of safety lead time to guide managers. There are several steps managers can take to ensure that requisites are addressed both proactively and reactively. The companies applied adaptive control of safety lead time in different ways, affecting the time required to complete each iteration. It is argued that their approach was influenced by, for example, different company characteristics such as company size, autonomy (i.e. freedom to make changes within the company), the product complexity under study and potentially also the respondents’ individual preferences. Another finding of managerial relevance is that the calendar time required to iterate and observe preliminary results can be extensive, especially when changes must be made to the ERP system to fully implement adaptive control. The interactive research approach has the potential to enhance the practical applicability of research results (Svensson et al., 2015; Eltantawy et al., 2015). The participating companies explained that the current research design enabled them to provide input to the study in a different way than when they were only interviewed, as they were used to in research projects. Finally, a central aspect of safety buffer control is planning parameter updating. Enabling this with an adaptive control approach has been valuable for managers in the study.

Given the complexities and challenges in dimensioning safety buffers, research gaps in knowledge on adaptive control of safety buffers and a lack of practical studies using real-world data, this study aimed to explore the requisites for applying adaptive control of safety lead time in manufacturing companies. A framework of requisites was developed.

The study’s limitations can inform further research. The propositions are empirically grounded but should be further validated on a larger population to take further steps toward theory-building, in line with Bryman and Bell (2022). An obvious continuation would be to test the propositions, reformulated as hypotheses, in a survey. One limitation is the focus on safety lead time. Further research can expand to study the requisites for other safety buffer types identified by Hedvall et al. (2025), examining whether the requisites found here for adaptive control of safety lead time also apply for material and capacity buffers. In addition, the effect of combining different safety buffers and the effect on requisites would be interesting to investigate. A more general survey-based investigation into the existence of various types of safety buffers in manufacturing companies would be a valuable continuation, expanding the case-based investigation of safety buffers by Hedvall et al. (2025). Another limitation is that, even if the companies represented different production strategies, industries and characteristics, they behaved surprisingly homogeneously in the study. Their differing lead time lengths were suggested to affect the speed of understanding cause-and-effect relationships, and the freedom to make decisions was noted to vary, which also affected the speed of understanding. As this study was limited to large- and medium-sized manufacturing companies, continued research on smaller manufacturers may identify other requisites and propositions.

The identification of different application approaches, influenced by various company characteristics, would also be interesting to study further. It was observed that, for example, the time needed for an iteration varied significantly with lead time length. Such an understanding would be relevant when implementing an adaptive safety buffer control process. Another interesting study would explore the use of control diagrams and control limits to manage, e.g. safety buffer utilization rate to enable better decision models. This could be built on the work by Mattsson et al. (2020), who discussed making changes when a primary performance variable is outside the set control limits. From a control perspective, it would also be essential to further study the dynamic properties of this approach, including potential issues with, for example, its dynamic stability. Finally, control of safety buffers should leverage available data or be highly data-driven. In such situations, machine learning can be used to develop an understanding of cause-and-effect relationships in a data-driven manner. Implementing digital tools and technologies is often tied to metrics and measurement, which has proven problematic (Kassem et al., 2024).