Big Data Demand Forecasting Regarding High Value/Highly Volatile Stock Keeping Units

Table of Content

List of Abbreviations

List of Figures

List of Tables

1 Introduction
1.1 Motivation and Background
1.2 Objective and Outline of this Paper

2 Acquisition of Big Data

3 Classification Concepts
3.1 VED Analysis
3.2 FSN Analysis
3.3 SDE Analysis
3.4 HML Analysis
3.6 XYZ Analysis
3.7 ABC - XYZ Analysis
3.8 Classification Synopsis

4 Demand Forecasting Methods
4.1 Qualitative Forecasting Methods
4.1.1 Delphi Method
4.1.2 Jury of Expert Opinion
4.1.3 Market Survey
4.1.4 Qualitative Forecasting Synopsis
4.2 Quantitative Forecasting Methods
4.2.1 Time Series Models
4.2.1.1 Moving Average
4.2.1.2 Exponential Smoothing
4.2.1.3 Box-Jenkins
4.2.1.4 Time-Series Forecasting Synopsis
4.2.2 Causal Models
4.2.2.1 Regression Models
4.2.2.2 Econometrics Models
4.2.2.3 ANN Models
4.2.2.4 Causal Forecasting Synopsis
4.4 Contrasting Juxtaposition of Forecasting Methods and Classification Concepts

5 Case Studies
5.1 Application of Forecasting/Classification Compatibility Matrix
5.2 Industrial Application

6 Conclusion
6.1 Contribution of this Study
6.2 Managerial Implication
6.3 Limitations and further Research

References

Assignment

Big Data Demand Forecasting regarding high value/highly volatile Stock Keeping Units

Sebastian Neumann

Abstract:

Predictive analytics, in the context of big data demand forecasting, is a revolutionary phenomenon in the modern world, and is expected to remain so in the foreseeable future (cf. Hassani and Silva, 2015, pp. 5-19 and Edwards, 2016). In this paper, the conditioning of big data, in the context of organizations that carry out a demand forecast, is differentiated into three parts. Accordingly, the procedure of data acquisition, data classification concepts, and forecasting methods are examined. The review of the literature has yielded an assessment method that supports demand planners to determine how big pools of data can be classified and utilized to carry out a forecast in a compatible manner. The case related application of the developed assessment method in organizations that forecast highly volatile stock keeping units with high monetary value was successful.

Keywords: Predictive Analytics, Big Data, Demand Forecasting, Statistics

Date of Submission: 02.08.2020

SCM08 Special Topics in Supply Chain Management

Big Data Demand Forecasting regarding high value/highly volatile Stock Keeping Units

Sebastian Neumann

Date of Submission: 02.08.2020

List of Abbreviations

Abbildung in dieser Leseprobe nicht enthalten

List of Figures

Figure 1:Research Design 1.0

Figure 2: HML Analysis based on cumulative unit cost

Figure 3: ABC Classification and Pareto Curve

Figure 4: XYZ SKUs Illustration of Demand Patterns

Figure 5: ABC-XYZ combined Classification

Figure 6: Categorization of Forecasting Models

Figure 7: First & fifth Iteration of Delphi Method

Figure 8: Research Design 2.0

Figure 9: Research Design 3.0

List of Tables

Table 1: Synopsis of Classification Concepts

Table 2: Synopsis of Qualitative Forecasting Methods

Table 3: Synopsis of Time-Series Forecasting Methods

Table 4: Synopsis of Causal Forecasting Methods

Table 5: Forecasting/Classification Compatibility Matrix

Table 6:BASF - Forecasting/Classification Compatibility Matrix

Table 7: Errasti et al. - Forecasting/Classification Compatibility Matrix

Table 8: Wavin and Noordstar - Forecasting/Classification Compatibility Matrix

Table 9: Nestlé - Forecasting/Classification Compatibility Matrix

Table 10: Conclusion - Forecasting/Classification Compatibility Matrix

1 Introduction

"Out of clutter, find simplicity.”

Albert Einstein

1.1 Motivation and Background

Inventory is of significant importance within any industry. Accordingly, managing it properly is vital for any organization regarding a financial and operational standpoint (Devarajan and Jayamohan, 2016, p. 563). Thus, inventory is essential for the provision of goods and services supplied to the customer, and it represents its financial investment for any company (Barlow, 1997, pp. 11-22). Without sufficient tools to control it, the non­availability of inventory comes with the cost of stock-outs, reordering costs, and additional transit costs (Devarajan and Jayamohan, 2016, p. 563). Simultaneously inventory can grow beyond economic limits, therefore tying up funds and increasing the cost of maintenance (Devarajan and Jayamohan, 2016, p. 563). Other authors state that managers have their difficulties in knowing when to order and how much to order, which is why inventory requires adequate attention for a business to stay competitive, flexible for the demand, and at low cost (Nemtajela and Mbohwa, 2017, pp. 699-706). The proper management of inventory is hence a necessity, to determine the item(s), time, quantity to indent, and amount of stock, for the reason to minimize storing and purchasing costs, without affecting distribution, production and functional effectiveness (Singh et al., 2014, pp. 80-82 and Das et al., 2012, pp.1-15). Therefore, the success of any organization is defined by how well it manages its inventory (cf. Baten and Kamil, 2009, pp. 183-187).

To manage any organization's inventory practices, they can examine their historical data and choose from a variety of inventory analyses to classify different stock keeping units (SKUs) into distinct categories (Melanie, 2018). The classification of SKUs is essential to a variety of departments within any supply chain organization (cf. Scholz-Reiter et al., 2012, pp. 445-451). Furthermore, organizations conduct demand forecasts, which is the process that is responsible for making informed business decisions regarding warehousing needs, inventory planning, and meeting customer expectations (cf. Lopienski, 2019). Therefore, the entire forecasting process itself can be separated into three separate sub-processes (cf. Regodic, 2017, pp. 29-31 and Kuusirinne, 2014). Accordingly, these are:

- Historical data acquisition [see. Chapter 2]. This first sub-process involves the collection of historical data about the previous sales and goods movement. The organization possesses many platforms to acquire big data.
- Data classification concepts [see. Chapter 3]. The second sub-process is about organizing the previously acquired data. Thus, organizations have an increasing number of SKUs in their inventory that must be classified to establish a system.
- Demand forecasting methods [see. Chapter 4]. Lastly, the final sub-process engages in the execution of the demand forecast itself. The forecast is carried out, utilizing different forecast methods.

Hence, organizations can only maintain long-term competitiveness and profitability if [...] and the associated forecasting processes are designed efficiently and effectively (Herrmann, 2011, p. 111). Since means of extracting historical data, classifying this data, and using it to conduct forecasts, goes hand in hand, absolute compatibility must be assumed.

1.2 Objective and Outline of this Paper

Abbildung in dieser Leseprobe nicht enthalten

Source: Own Illustration

Figure 1:Research Design 1.0.

This paper follows a predominant theoretical approach by consulting literature , where a multitude of different scientific sources; books, research papers, and websites have been consulted to identify correlations between the latter. Therefore, the objective of this paper is to identify the compatibility between classification concepts and forecasting methods. To give this paper objective-oriented reasoning, it is split into two parts. The first part of this paper deals with stating theoretical fundamentals to identify an assessment method to prove compatibility. The second part is dedicated to the application of this assessment method onto case studies, to receive practice related proof that compatibility is present in the business environment, The following paper is subdivided into a total of six chapters. Accordingly, after the introductory chapter has shed light on the subject, further work on the topic will be conducted as follows: Chapter two will provide insights regarding the relationship between big data and the objective of this paper, by providing information about the acquisition of planning data. Within chapter three, different classification concepts are introduced considering the context of compatibility with specific forecasting methods of chapter four, respectively. This predominantly conducted literature review results in an assessment method that is applied in chapter five, where identified insights are considered in the context of different case studies. Lastly, chapter six states a brief conclusion about essential results found out in this paper.

2 Acquisition of Big Data

Data has once begun as little bits of information available in mind, later then in language, text, books, and as files on tablets, phones, and personal computers. Digitalization has made it possible that through machine-to-machine (M2M) communication, data is developed, transferred, and stored. To handle these big data pools, ambitious tools and programs are necessary to ensure sufficient classification, analysis, and usage by humans (cf. Big Data Blog, 2020). Radtke and Litzel (2019) describe the terminology of “big data” as an immense pool of data that exists in organizations, as well as the internet. The identification of a unified definition for Big Data is difficult, however (Shi, 2014, pp. 6-11) presented a definition for academics, describing this phenomenon as a “collection of data with complexity, diversity, heterogeneity, and high potential value that are difficult to process and analyze in a reasonable time.” Bendel (2020) states that it involves, among other things, grid searches, (inter)dependency analysis, system and production control, as well as environment and trend researches. Technologies such as [...] and big data have become proficient at spotting trends in large data pools (cf. Gaskell, 2019).

“Predictive analytics”, a category of “data analytics” (analysis of big data pools (Rouse, 2020)), is aiming at making predictions about outcomes in the future based on historical data and analytics techniques. (cf. Edwards, 2016). Predictive analytics has gained in importance with the emergence of big-data systems (cf. Rouse, 2020). The field of predictive analytics has captured the support of a wide range of organizations, growing at a compound annual growth rate (CAGR) of around 21% between 2016 and 2022, according to a report issued by (cf. Zion Market Research, 2017).

Alicke (2003, p. 25) claims that for the new conception and operation of supply chains, historical data is utilized. This data is, thanks to the widespread use of enterprise resource planning (ERP) software, available in different formats on diverse systems (Alicke, 2003, p. 25). Further, data availability is often no longer a problem (Rouwette, 2010, p. 267). Data is accessible in different hierarchies (product, family, region) (cf. Stadtler, 2002, p. 343), and different temporal granularities (hourly, daily, weekly) (cf. Kepczynski, 2018, p. 72). On the one hand aggregation of data is essential for the application of forecasting methods (Wilson, 2019) as it provides greater security and reduces complexity (cf. Singh, 2016 and Alicke, 2003, p. 25). On the other hand, planning must take place at detailed product levels since it does not make sense to apply an inventory strategy to an aggregated product group (cf. Singh, 2016 and Alicke, 2003, p. 25). For simplicity reasons, this paper will continue to refer to the term “data” itself.

3 Classification Concepts

After the process of data acquisition has taken place in the context of big data, any organization must decide how it will continue the process. Usually, it is followed up by having a closer look and find a proper declaration for the data. Further, if companies began optimizing their data, retailers could increase margins by 60%, the average Fortune 1000 company, when utilizing data 10% more effectively, would gain $2 billion in added revenue and the U.S. healthcare system could save $200 billion per year (cf. Pal, 2016). Therefore, the gathered planning data will be analyzed, and the proper classification concept must be chosen.

Different purposes require the employment of different inventory classification methods (cf. Dhoka, 2016, p. 26). The classification of stock keeping units is essential to the manufacturing and logistics of organizations since it supports the stock management and helps with the realization of potentials (cf. Scholz-Reiter et al., 2012, pp. 445-451). Additionally, Bruckner et al. (1998) stated that a classification of SKUs can help with finding the material planning strategy of a variety of items. Furthermore, Bruckner et al. (1998) argue that many different possibilities to classify SKUs exist because of different objectives. One of the main techniques is the “ABC Analysis” where the items are ranked according to the annual turnover and the “XYZ Analysis” (analysis of the usage regularity) (cf. Scholz-Reiter et al., 2012, pp. 445-451). Others are known as VED-, FSN- , SDE- and HML-Analyses. To understand their implications, benefits, and limitations regarding the decision-making process, an introduction of these concepts is provided.

3.1 VED Analysis

The letters of the VED analysis stand for vital, essential, and desirable (Kumar, 2012), which is mostly utilized in spare parts management and not famously used while managing the inventory (Besta et al., 2012). However, there are some clear cases, where the criticality of goods is of importance. For instance in health institutions, products must be on stock, since some of the goods are essential to keep patients alive_(Günergören and Dagdeviren, 2017, pp. 11-17). Organizations can categorize their products just as they wish, but concerning criticality, it rarely makes sense for no health-related industries. Hence, in other industries, it often is applied to goods where for instance, the D and E parts cannot be manufactured without the V part, therefore the V part being the most critical for the manufacturing process (Panneerselvam, 2012, p. 310). SKUs that must be available at all times are in the V category. Items with a lower need for availability are within the E category and SKUs that may be available in the warehouse are included in the D group (Devnani, 2010, p. 203). The VED analysis is often used in combination with the ABC analysis, resulting in an ABC - VED Analysis (cf. Vaz, 2014, pp. 1-3).

3.2 FSN Analysis

In the manufacturing industry, not all items are required as frequently as others. Some materials are regularly needed, yet some others are needed less regular and some materials may have become obsolete and might not have been demanded for quite some time, which identifies as “dead-stock” (cf. Mitra et al., 2015, pp. 322-325). FSN stands for fast-moving, slow-moving, and non-moving SKUs (Chan, 2018).

To execute this analysis, the turnover ratio must be determined, since categories are defined by that value. Accordingly, the F category consists out of stock with a turnover ratio of more than three and constitutes around 10% and 15% of total inventory. The S category consists of stock that has a turnover ratio of one to three and is generally between 30% and 35% of total stock. The N category consists of stock that rarely moves with an inventory turnover ratio below one and between 60% and 65% of total stock (Orderhive, 2020).

According to Besta et al. (2012), this analysis fails when used in manufacturing, where raw materials are utilized for production, and manufactured SKUs will remain in inventory given a wrong assumption of consumption. Therefore, it is often used in combination with the XYZ analysis, simultaneously identifying the SKUs to be discarded, and the corresponding amount saved (cf. Devarajan and Jayamohan, 2016, p. 563). FSN analysis is mostly utilized to deal with obsolete SKUs whether spare parts are defined as raw materials or components. This analysis helps in the arrangement of stocks within stores, and their handling methods, including the distribution factor. Thus, the main objective of this analysis is to control obsolescence throughout the stock (cf. Devarajan and Jayamohan, 2016, p. 565), since non-moving inventory increases not only inventory carrying cost, but also the bottom line, including devaluing the invested money.

3.3 SDE Analysis

SDE analysis classifies inventory based on three different levels regarding the degree of availability. Accordingly, the letters stand for scarce, difficult, and easy (Chang, 2018). Scarce SKUs are usually products, parts, or materials that are imported, therefore taking longer to arrive, which equals in a more difficult availability. As “difficult” classified SKUs can be products that can be procured domestically but are some sort of rare, hence difficult to procure.

As “easy” classified SKUs are easy to come by and are usually attainable within short times (Melanie, 2018). This analysis mostly depends on how the vendors are managed, where strategic purchasing is of major importance, and inaccurate information can distort the analysis (Dhoka, 2016, p. 24). This analysis is controversial since Shawal (2020) states that this analysis provides proper guidelines for deciding the inventory policies, but another researcher argues that it can provide guidance, but should not be used to plan exact dates (Melanie, 2018).

Nevertheless, Brindha (2014, p. 8172) concludes that this analysis helps with purchasing policies. Accordingly, forward purchasing policies are adopted for scarce SKUs. For difficult SKUs, schedule-purchasing policies are adopted and for easy SKUs, contract­purchasing policies are adopted.

3.4 HML Analysis

Bhadiyadran et al. (2018, pp. 2387-2390) state that the letters of the HML analysis stand for high price, medium price, and low price. It cannot be used unless the SKUs have a noticeable impact on the total inventory, in the sense that SKUs with high value have very low transactions and in most cases are found in make to order (MTO) and pick to order (PTO) scenarios (cf. Dhoka, 2016, p. 24 and Singh, 2013). The main objective is to minimize the costs of inventory, labor, and material (Kumar et al., 2016, p. 521). The value of interest, within this analysis, is the unit price (cf. Hukum and Shrouty, 2019, p. 352). Therefore, Kumar et al. (2016, p. 521) argue that the HML analysis is similar to the ABC analysis since it is also based on the Pareto principle or the 80/20 rule, but instead of using the annual consumption value such as in ABC classification, cost per unit is utilized (Biswas, 2017, p. 37).

Abbildung in dieser Leseprobe nicht enthalten

Source: Own Illustration based on (Kumar et al., 2016, p. 525)

Figure 2: HML Analysis based on cumulative unit cost.

Figure 2 shows a data set, where SKUs are classified within the previously described H, M, and L groups. Thus, the X-axis shows the HML classification and the Y-axis shows the cumulative unit costs in percent. Accordingly, high-priced products are classified until 75%, medium-priced products are classified between 75% and 95% and low-priced products are classified between 95% and 100%. Hence, the most valuable SKUs require frequent replenishment (Biswas et al., 2017, pp. 37).

3.5 ABC Analysis

Jung (2006) states that if items are characterized by A-, B- and C-classes, the ideal Pareto principle is found with comparably few, but valuable A-SKUs, and a multitudinous of invaluable C-SKUs. Ravinder and Misra (2014, pp. 257-264) elaborate by stating that, the ABC classification is a well-established technique, for determining which SKUs should receive priority in the management of an organization's inventory. The periodic turnover is the characteristic, which is utilized to classify the SKUs and is defined as the cost of a unit and its consumption rate within a certain period (cf. Scholz-Reiter et al., 2012, pp. 445-451). Typically, as mentioned three classes are examined. Each consisting out of a certain number of percentages of all SKUs. Some rankings for A:B:C - SKUs are 80%:15%:5% (cf. Ivanov et al., 2017, pp. 8-13), 75%:20%:5% (cf. Vermol, 2020) and 60%:30%:10% (cf. Kuusirinne, 2014, p. 28), where the percentages stand for the accumulated consumption value (share of value). However, ranks are described as arbitrary (Stojanovic and Regodic, 2017, p. 36).

Abbildung in dieser Leseprobe nicht enthalten

Source: Own Illustration based on (Bellina, 2013).

Figure 3: ABC Classification and Pareto Curve.

Accordingly, as seen in figure 3, A-SKUs are considered to bring the most value or are considered as items that generate the greatest value. B-SKUs are generating less value or produce less value. Lastly, C-SKUs are items that bring the least value or are considered as items that generate the least amount of value (cf. SAP, n.d.). Therefore, 20% of the total number of SKUs are responsible for 80% of the SKUs worth (Pareto Principle). B-SKUs contribute around 15% and C-SKUs contribute 5% of the total share of value.

However, the ABC analysis is a classification method that exhibits many limitations. Those limits tend to exacerbate many pre-existing supply chain problems such as stockouts, overstocks, unreliability, and low productivity (cf. Ravinder, 2014). Ravinder (2014) argues that the current globalized hyper-responsive business environment must consider multiple criteria to guide the management of inventories. Keskin and Ozkan (2013) describe ABC analysis as a clustering problem in which SKUs that must be categorized are partitioned into three “fuzzy” clusters . Fuzzy clustering is the appropriate term to utilize since some SKUs can belong to not only one, but maybe two clusters (cf. Keskin and Ozkan, 2013).

Back in the early 1980s, the researchers in operations and inventory management recognized this fact and since then, have proposed numerous approaches to multi­criteria ABC classification (Flores and Whybark, 1987, pp. 79-85).

Additionally, for many organizations, the classification regarding SKUs based on an ABC analysis is not sufficient because the items are not differentiated enough (Logisticaudit, 2014). However, this paper solely considers the generic ABC classification technique in order not to extend the frame of this work further. Consequently, the ABC analysis has some limitations, which are overcome by introducing the XYZ analysis (Stojanovic and Regodic, 2017, p. 45).

3.6 XYZ Analysis

The XYZ analysis, hence, can be considered a secondary analysis of inventories that examines the average level of demand with the application of the demand variability criterion (cf. Stojanovic and Regodic, 2017, p. 45), or in other words, this analysis is utilized to identify SKUs and the variability/variance of their demand (cf. Stojanovic and Regodic, 2017, p. 36; CGMA, 2020 and Logisticaudit, 2014). T2informatik (2020) describes this analysis as a method within the spectrum of business administration or materials management, where goods are classified in terms of their regularity of turnover. Bulinski et al. (2013, pp. 89-96) continue by arguing that the XYZ classification is a modification of the ABC analysis, which provides the classification of a SKU based on the rate of their selling.

That translates to providing knowledge regarding decision making in stock management. The rate of SKUs selling is utilized to illustrate the variation in demand and depict the regularity of turnover (cf. t2informatik, 2020). The definition of the product groups and their ranking based on the following calculations are of major importance for this analysis (cf. Stojanovic and Regodic, 2017, p. 36) [see: Formula (1) and (2)]. The ranking with which the authors follow up with is suggested as X-SKUs classified from 0% to 10%, Y- SKUs from 10% to 25% and Z-SKUs are classified from 25% to infinity (~). Classification for XYZ SKUs is arbitrary as well, just as in the case of the ABC classification (cf. Stojanovic and Regodic, 2017).

Abbildung in dieser Leseprobe nicht enthalten

Figure 4: XYZ SKUs Illustration of Demand Patterns.

Figure 4 shows the movement patterns according to which the SKUs can be classified. X-SKUs are goods, with very little variation, which have a stable turnover time. Hence, demand can be forecasted reliably. Y-SKUs are goods with some variation. Given this, demand is not stable, which leads to the conclusion that demand can only be predicted to an extent. The reasons for demand instability are well-known factors such as seasonality, product lifecycles, competitor actions, or economic factors, as well as marketing promotions. Z-SKUs are goods with the most variation, where demand can alter strongly or occur sporadically. Nor trend or other predictable known factors can be accounted for these SKUs, making a reliable demand forecast almost impossible (cf. Stojanovic and Regodic, 2017, p. 36 and CGMA, 2020).

The coefficient of variation (CV) needs to be determined, which is calculated as the ratio of the standard deviation and average sales. The CV is a relative value based on the spread of the probability distribution (Stojanovic and Regodic, 2017, p. 36). However, as seen in formula (1) and (2), the sum of squares, variances, standard deviation (SD), and the coefficient of variation (CV) must be compiled, which can be tedious and prone to errors.

Abbildung in dieser Leseprobe nicht enthalten

Μ where σ is considered the standard deviation, n is considered the total number of observations, xt the individual values and ω the average value (cf. Dhoka, 2016, p. 24 and ncalculators, 2020).

Especially for organizations that have above 3000 SKUs spread over a large time frame, compiling the SD is exhausting (Dhoka, 2016, pp. 23-26). Given that this type of analysis must be conducted regularly for effective and efficient management of inventory (Dhoka, 2016, p. 25), it adds to the level of tediousness. The most essential drawback for the XYZ analysis is the classification of new products, since they are mostly classified as Z- SKUs, because of their not yet established demand pattern (Dhoka, 2016, p. 26). Furthermore, according to Dhoka (2016, pp. 23-26), there are no benchmarks nor industry standards, that define predictability of demand among SKUs. Lastly, this analysis can completely overlook seasonal SKUs (Dhoka, 2016, pp. 23-26).

3.7 ABC - XYZ Analysis

A symbiosis of the two previously described analyses results in the “ABC - XYZ Analysis” (Stojanovic and Regodic, 2017, p. 45). Purposefully, this analysis contributes to an optimal inventory level, which is considered a key condition concerning cost reduction of an organization (cf. Stojanovic and Regodic, 2017, p. 46). According to (Blaszczyk, 2005, p. 106), making a large selection of goods in sufficient quantities available to the end customer is considered the main task of all organizations. Organizations offer a large variety of goods, which is why the management calls for a division into goods of lower and higher strategic importance (Bulinski, 2013, p. 90). Brown (1982, p. 155) stated that such classification is performed since the determination of the purchase policy, production planning, and store management plays an important role in any organization.

Most commonly, this analysis is conducted by the aid of past consumption data and is considered of high importance for manufacturing strategies regarding supply and inventory control (Scholz-Reiter, 2012). By utilizing the ABC - XYZ analysis, an organization can identify products that might rupture or overstock and control this issue by defining a proper stock and security stock coverage and optimize production volumes (AbcSupplyChain LTD, 2020). Consequently, this analysis is a popular tool in supply chain management (SCM), which also builds upon the principle of Pareto, already described as the 80/20 rule, which expects that the minority of cases has a disproportional impact to the whole (Swamidass, 2000 and Kourentzes, 2016). Goldman (2010) elaborates, that as for the ABC part of this analysis in regard to SCM, it is assumed that 20% of the items establish 80% of the value of demand. In other words, “the important few versus the trivial many” (cf. Ravinder, 2014, pp. 257-263).

Accordingly, in SCM, the ABC - XYZ analysis is a method of assorting planning items (characteristic value combinations, SKU) based upon their worth (revenue or sales volume) and dynamics of consumption or sales (cf. SAP, n.d.). It is suggested that by classifying demand based on volume, importance is indicated, and the factor volatility is indicating forecastability (cf. Li, 2019 and Bulinski, 2013 pp. 89-96).

Abbildung in dieser Leseprobe nicht enthalten

Source: Own Illustration based on (Malik, 2019 and Nagtegaal, 2009, p.14).

Figure 5: ABC-XYZ combined Classification.

Figure 5 shows the symbiosis of the ABC and XYZ analyses from which the management can decide how the assortments must be managed (Malik, 2019). Different approaches on how to further define strategies for these classes are presented starting with the general individual perception of these classes, after which a more collective approach is described.

Accordingly, class AX consists of SKUs with great demand forecast accuracy given continuous consumption and a big share of the total value (cf. Stojanovic and Regodic, 2017, p. 37). The class AY consists of SKUs with a big share of the total value, but there might be some trends of seasonality, which limits the forecast accuracy (Baker, 2020). Class AZ consists also of SKUs with a high share of the total value, but a consumption from time to time, which results in little demand forecast accuracy (cf. Stojanovic and Regodic, 2017). Class BX consist of SKUs with a medium share of the total value, but constant demand, which results in great forecast accuracy (Baker, 2020).

The class BY includes SKUs with discontinuous consumption, the medium share of the total value, and therefore a medium degree of forecast accuracy (Stojanovic and Regodic, 2017, p. 37). CX is the class with a small share of the total value.

However, it has constant consumption which therefore results in great forecast accuracy (Baker, 2020). Stojanovic and Regodic (2017, p. 37) state that classes BZ, CY, and CZ can be neglected in terms of the impact regarding any organization's business operations. Furthermore, SKUs classified as AX, AY and AZ are mostly forecasted by the aid of an account manager and a sales forecast (Malik, 2019). SKUs that are BX, BY and CX are managed by statistical forecasting, and SKUs classified as BZ, CY and CZ do not receive much attention and require further discussion (Malik, 2019).

Another approach is to claim that the combinations AX, BX, CX, AY, BY are all suitable for just-in-time (JIT) procurement, where class BY can be managed via stock procurement strategies as well. Given that AZ and BZ are considered difficult to forecast, compromised strategies between expensive warehousing, high-cost ordering, and the risk of bottlenecks are utilized. CY and CZ shall be minimized, where a compromised approach is proposed as well (cf. t2informatik, 2020). There are more strategies for each combination (Noche, n.d.).

However, according to Bellina (2013, p. 28), literature does not provide unique insights regarding the relationship between an ABC analysis and the XYZ analysis. As described previously, A-classes consist of the greatest amount of total value and therefore Nahmias (2009) emphasizes the necessity of such classification regarding how great the attention must be in terms of inventory levels. Furthermore, Stock et al. (2000) argue that A-SKUs are most profitable for any enterprise and shall not be neglected to avoid shortages. Other authors consider that C-SKUs should receive maximum attention regarding service level, since facing stock-out costs is not appropriate even when these SKUs have low value (cf. Knod and Schonberger, 2001). Since the attention brought to the different SKU-classes, is not being unitary (Bellina, 2013, p. 29), the consequence of this mismatch (Teunter et al., 2010, pp. 343-352) exists because of missing factors such as the obsolescence rate, lead-times or replenishment costs (cf. Ramanathan, 2006, pp. 695-700; Zhou and Fan, 2007, pp. 1488-1491 and Chen and Li, 2008).

3.8 Classification Synopsis

This chapter provides a comprehensive conclusion about the introduced classification concepts, which is supported by table 1. This is of importance since first, it summarizes how classification concepts do or do not provide any benefit to the forecasting method, and secondly, it provides how their limitations are affecting the forecast. Furthermore, the decision-basis for choosing the proper forecasting method, based on the classified data is provided.

Table 1: Synopsis of Classification Concepts.

Abbildung in dieser Leseprobe nicht enthalten

Source: Own Illustration

VED analysis. Summarized, the classification with a VED analysis brings immense benefits for goods that can be evaluated according to their criticality. Combining it with an ABC concept yields a more significant basis for decision-making.

- As for the consideration in terms of forecasting, this classification method does provide insights regarding goods being critical, and therefore a certain amount must be on stock, leaving the demand certain, linear, and sufficient for every forecast method. Additionally, since a certain demand is given, no advanced forecasting method must be utilized.
- FSN analysis. The FSN analysis sets its emphasis on classifying the SKUs based on the turnover-ratio. Its goal is to identify obsolete products. To properly monitor the demand and classify SKUs accordingly, this method can display variability since it shows how often a certain product is consumed. If consumption rate changes, so does turnover rates, and demand (Orderhive, 2020). Leaving this analysis, with no choice but to display uncertain demand amongst those F-SKUs. In other words, an increasing or decreasing rate of consumption of a given SKU, is equal to an increasing or decreasing trend, defining the demand as uncertain and non-linear. However, demand can also stay certain and this analysis would be able to display a linear demand as well. Since, the FSN analysis can classify SKUs, ranging from certain to uncertain demand, some restrictions require consideration while choosing the forecasting method. Hence, if demand is uncertain and a trend is displayed within the demand, a sufficient forecasting method must be chosen.
- SDE analysis. The SDE analysis classifies SKUs regarding the scarcity. The main objective of this analysis is to define procurement policies regarding each of the classified SKUs. Therefore, setting the focus on the procurement of products and with it, the ability to ensure supply at the point in time where it is needed (Constantino et al., 2018, pp. 57-66). Given this, demand can be defined as certain and periodic, qualifying this classification approach for lesser advanced forecasting methods and no advanced forecasting method must be utilized. However, this classification lays the basis for inventory management principles such as the determination of optimal reordering points and optimal reordering amounts. It is questionable if executing an additional forecast is reasonable.
- HML analysis. The HML analysis is classifying SKUs concerning their worth and only provides information to some extent regarding the demand. Hence, one could argue that the most valuable SKUs require frequent replenishment, which translates into a certain, periodic demand. Therefore, qualifying this classification concept for lesser advanced forecasting methods. Just as for the SDE Analysis, inventory management principles are more than capable to sufficiently complement the HML classification concept and the execution of a forecast is questionable.
- ABC analysis. The classification of SKUs with the help of the ABC analysis is carried out regarding the share of value in percent and the number of SKUs in percent. Since it classifies products according to their worth, it shows similarities to the HML analysis. Therefore, again, one could argue that SKUs with more value require a more frequent replenishment than SKUs with less value. This translates into a certain, periodic demand, where inventory management principles would complement this classification sufficiently. Therefore, the utilization of less-advanced forecasting methods is questionable as well.
- XYZ analysis. The XYZ analysis classifies SKUs according to the variability/variance regarding their demand. Therefore, this classification concept sorts products in terms of the regularity of turnover. Hence, the variation in demand can be determined, whether it is certain and uncertain, providing valuable information about trends and seasonality. Depending on the outcome of the classification, sufficient forecasting methods can be chosen. With a certain, linear demand, less-advanced methods can be applied, and accordingly, if trends and seasonality are seen within the classification, advanced forecasting methods must be chosen.
- ABC - XYZ analysis. The ABC - XYZ analysis adds real value to the forecast regarding the classification/segmentation of data and contributes to driving better forecast accuracy once the optimal forecast method has been assigned (Malik, 2019). This classification method separates SKUs according to their variability and their monetary importance. Since the variability and therefore the volatility is determined, and products are classified accordingly, the demand, whether it is linear, shows a trend, or seasonality, can be forecasted properly. This classification method prepares the data set in a way, that however difficult the demand is, a sufficient forecasting method can be chosen.

[...]