Causes and Effects of Poor Demand 
Forecast Accuracy  
A Case Study in the Swedish Automotive Industry 
Master’s Thesis in the Master’s Programmes  
Management and Economics of Innovation & Supply Chain Management 
 

TEODOR MARTINSSON 
EDVIN SJÖQVIST 
 
 
 
 
 
Department of Technology Management and Economics 
Division of Supply and Operations Management 
CHALMERS UNIVERSITY OF TECHNOLOGY 
Gothenburg, Sweden 2019 
Report No. E 2019:075	  



	

 
 
 

MASTER’S THESIS E 2019:075 
 
 
 
 
 
 
 
 
 
 

Causes and Effects of Poor Demand Forecast 
Accuracy 

A Case Study in the Swedish Automotive Industry 
 
 

TEODOR MARTINSSON 
EDVIN SJÖQVIST 

 
 
 
 
 
 
 
 
 
 
 
 
 
 

Tutor, Chalmers: Patrik Jonsson 
Tutor, company: Johan Bystedt 

 
 
 
 
 
 
 

Department of Technology Management and Economics 
Division of Supply and Operations Management 
CHALMERS UNIVERSITY OF TECHNOLOGY 

Gothenburg, Sweden 2019  



	

 
 
 
 
 
 
 
 
 
 
Causes and Effects of Poor Demand Forecast Accuracy 
A Case Study in the Swedish Automotive Industry 

TEODOR MARTINSSON 

EDVIN SJÖQVIST 
 
 
© Teodor Martinsson & Edvin Sjöqvist, 2019. 
 
 
Master’s Thesis E 2019:075 

 
 
Department of Technology Management and Economics 
Division of Supply and Operations Management 
Chalmers University of Technology 
SE-412 96 Gothenburg, Sweden  
Telephone: + 46 (0)31-772 1000 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Chalmers digitaltryck 
Gothenburg, Sweden 2019



 

Abstract 
This study is a part of a FFI (Fordonsstrategisk Forskning och Innovation) project “Future of 
sharing schedule information in automotive industry supply chains using advanced data 
analytics”. The study is aimed at describing the current situation in terms of accuracy of 
demand forecasts sent from OEM companies to their suppliers within the Swedish automotive 
industry, identifying root causes for inaccuracies in demand forecasts and their effect on the 
suppliers. This study also aims to provide some guidance to future actions and initiatives for 
improvement of demand forecast accuracy. 
 
An extensive database of delivery schedules was used to identify current patterns in forecasting 
accuracy, utilising FAI (Forecast Accuracy Index) to analyse forecasting performance. The 
study employed a case methodology, studying three customers with a single supplier as the 
focal point as a basis to find root causes and effects of poor forecasting accuracy.  
 
The study found that current demand forecast accuracy was poor. Causes for poor performance 
were found both in sales forecasts, that were used to generate the MPS and subsequently 
component demand, and in the MRP systems of the customers. Inaccuracies in demand 
forecasts were found to mainly be dealt with through buffers of materials and finished 
components at the supplier. Improved forecasting accuracy is expected to allow suppliers to 
lower their inventory levels, resulting in cost savings across the entire supply chain. This study 
proposes evaluation of and changes to current MRP practices, closer integration of 
complementary data in the sales forecasting process and employment of machine learning 
algorithms in forecasting as promising areas for improving the accuracy of demand forecasts. 
 
Keywords: Supply planning, automotive industry, demand forecasts, forecasting accuracy, 
delivery schedule, FAI, MRP 
  



 

Acknowledgements 
This thesis was conducted during the spring of 2019 at Chalmers University of Technology, 
Gothenburg, Sweden. The thesis was written in collaboration with Meridion, as a part of the 
FFI (Fordonsstrategisk Forskning och Innovation) project “Future of sharing schedule 
information in automotive industry supply chains using advanced data analytics”.  
 
Firstly, we would like to extend our sincere gratitude to Johan Bystedt at Meridion for allowing 
us the opportunity to conduct this thesis in collaboration with them. His extensive knowledge 
of the automotive industry in Sweden, support and inspiration has been invaluable to this thesis. 
We would also like to thank Mikaela Holmer for her help with the Qlik application used for 
the quantitative data analysis in this thesis. 
 
Furthermore, we would like to thank the supplier studied for giving us access to the database 
of delivery schedules used as the primary quantitative data for this thesis. We also recognise 
the contribution of the persons at the supplier and their customers who allowed us insight into 
their forecasting and planning processes and extend our gratitude to them, this thesis would not 
have been possible without them. 
 
We would like to thank Patrik Jonsson, our supervisor at Chalmers, for his support during the 
process of writing this thesis. His input helped us greatly in delimiting and deciding on what to 
do next when we were confused or overwhelmed. 
 
Lastly, we are grateful to all of the employees at Meridion for their welcoming attitude and 
support, it has helped us greatly.  
 
 
 
Gothenburg, May 2019, 
Teodor Martinsson & Edvin Sjöqvist 
  



 

Table of Contents 
1. Introduction 1 

1.1 Introduction 1 
1.2 Aim 2 
1.3 Limitations 2 
1.4 Specification of issue under investigation 2 

2. Methodology 3 
2.1 Research approach 3 
2.2 Data collection 3 
2.3 Analysis 5 
2.4 Validity 6 
2.5 Limitations                                                                                                                                   7 

3. Theoretical framework 8 
3.1 Supply chain structure and information sharing 8 
3.2 Electronic Data Interchange 9 
3.3 Measurements for forecasting errors 10 
3.4 Forecasting 11 
3.5 Usage of forecasts in Production/logistics planning processes 13 

4. Case description 15 
4.1 Supplier studied 15 

4.1.1 Logistics and order process 16 
4.2 Customers studied 16 

4.2.1 OEM 1 16 
Material & information flow 17 

4.2.2 OEM 2 18 
Material & information flow 18 
Customer E 18 
Customer D 19 

Available data on forecasts 20 

5. Empirical results 21 
5.1 General patterns in FAI and WTS 21 

5.1.1 Patterns for the different component groups 21 
5.2 FAI statistics for major customers 25 

5.2.1 Customer D 26 
5.2.2 Customer E 27 
5.2.3 Customer H 28 

5.3 Interview data 29 
5.3.1 Interviews at OEM 1 29 

Material planner at OEM 1 29 
Volume and program manager 30 

5.3.2 Interviews at Customer H 31 
Assembly planning manager 31 



 

5.3.3 Interviews at OEM 2 31 
Interviews at the aftermarket organisation (Customer E) 31 

Material planner 31 
Forecasts 32 

Interviews at production plant (Customer D) 33 
Material planner 33 
Volume and Program Manager 33 

5.3.4 Interviews at the Supplier 34 
Interview 1 34 

Production and material planning 34 
Effects of variations 34 

Interview 2 35 
Effects of changes made by Customer E 36 
Managing variations 36 

Interview 3 36 
Interview 4 37 

6. Analysis & discussion 38 
6.1 Analysis of patterns in FAI and WTS 38 

6.1.1 General patterns in FAI 38 
6.1.2 Patterns in FAI for major customers 39 

Reasons and effects connected to poor FAI performance 40 
6.1.3 Patterns in WTS 41 

6.2 Potential sources and improvements to poor forecasting accuracy 42 
6.2.1 Sources of error and areas for improvement in sales forecasting 42 
6.2.2 Sources for errors and areas for improvement in processing and breakdown of sales 
forecasts 48 

Effects of batching in production, ordering and logistics 48 
MRP nervousness and volatility 51 
Order logic 55 

6.3 Effects on the Supplier as a result of unforeseen variations 57 

7. Conclusions 61 
7.1 Research question 1 61 
7.2 Research question 2 61 
7.3 Research question 3 62 

8. Recommendations 63 

 



1 

1. Introduction 

1.1 Introduction 
Information flow between manufacturers and their suppliers plays an important role in reducing 
uncertainties in the supply chain, which in turn can lead to operational and financial benefits 
(e.g. Gavirneni et al., 1999; Ryu et al., 2009; and Singh, 1996). It is therefore in the interest of 
all parties within the supply chain to ensure smooth and accurate flow of relevant information. 
One way of making this information transfer is electronic data interchange, EDI, which refers 
to the transmission of structured business data (UNECE, 2019). The message format used to 
transmit demand forecasts is called DELFOR. In 2012 Odette (Organisation for Data Exchange 
by Tele-Transmission in Europe) published a recommendation to track and improve quality of 
information regarding future demand. Two new metrics, FAI (Forecast Accuracy Index), and a 
WTS (Weighted Tracking Signal) were developed to help suppliers in assessing the quality of 
forecasts. (See appendix 1 for definitions of FAI and WTS). FAI measures the accuracy of the 
forecasts, and WTS tracks systematic bias in misestimates. Suppliers within the Swedish 
automotive industry have indicated that FAI scores are relatively low, with significant 
inaccuracies in demand forecasts resulting in uncertainties that make accurate planning of 
capacity and production more difficult. Lack of confidence in given demand forecasts is 
highlighted as one of the reasons suppliers build slack into their operations through temporal 
capacity or material buffers (Christopher & Lee, 2004). Building on this reasoning: If quality 
of demand forecasts improve confidence should follow, ultimately resulting in perceived need 
of costly slack and buffers being lower. While forecasting does nothing to reduce actual 
variations in end customer demand, accurate forecasting allows for companies to plan ahead, 
dealing with variations in a more reliable and cost-effective way. 
 
This study will explore forecasting accuracy across different time horizons, focusing on finding 
root causes of poor forecasting accuracy. These variations can have a number of different 
reasons, for example MRP ‘nervousness’ and bullwhip effects among others. Blackburn et al. 
(1985) argue that due to shifting demand from downstream processes, material requirements in 
the MRP system needs to be rescheduled. This creates nervousness in the system which can 
trigger upcoming orders to be sent upstreams. Variations given with short notice can prove 
problematic for suppliers as they cannot be fully compensated for and mitigated in planning. 
These variations require other measures, such as safety stocks to ensure high delivery precision 
to the customer. As significant safety stocks will prove expensive in the long run, it is of interest 
to investigate measures that the supplier can take to mitigate short term variations. Mackelprang 
& Malhotra (2015) found that suppliers that are able to utilise both production coordination 
with customers along with inventory and capacity buffers to mitigate the bullwhip effect 
experience a positive effect on their return on assets. This implies not only that suppliers to a 
significant extent are able to mitigate the bullwhip effect, but also that they could potentially 
gain a competitive edge by doing so. 
 



2 

To better be able to manage these variations, suppliers view it as important to understand where 
and why inaccuracies in communicated demand forecasts appear, and how specifically they 
impact the organisation. Finding ways to improve FAI scores on the customer end and 
developing a more sophisticated way of dealing with inaccuraccies on the supplier end would 
provide to add value to the supply chain by reducing costs from slack and buffers. 

1.2 Aim 
The aim of this study is to perform an in-depth case analysis of FAI data at a Tier 1 supplier 
within the Swedish automotive industry. Based on findings a cause and effect analysis will be 
performed to assess how the supplier is affected by the quality of forecasts and provide 
suggestions on ways to mitigate or reduce variation and inaccuracies in delivery schedules from 
the OEMs.  

1.3 Limitations 
To facilitate an in-depth investigation while keeping the workload manageable, the study is 
focused on the communication and demand forecast data between the supplier and three of their 
most important OEM customers (by revenue share). Furthermore, this study will not discuss 
aspects related to the design or relationships of the supply chain as a whole, but rather examine 
events occurring within the relationship of two business, or inside the organisational boundaries 
of these two entities operating within a larger supply chain. 

1.4 Specification of issue under investigation 
Considering the context and limitations of the study, three main research questions were 
formulated: 
 

1. What is the current level of forecasting accuracy in DELFOR messages? What 
patterns can be seen in FAI and WTS scores? 
 

2. Why do inaccuracies in demand forecasts in DELFOR messages appear, and what 
actions could organisations take to improve their forecasting accuracy? 
 

3. What are the effect on suppliers stemming from inaccurate forecasts in DELFOR 
messages? 

 
The study will be divided into two phases, the first phase will be quantitatively focused and 
seeks to answer the first research question. The second phase will seek to explain and provide 
suggestions based on the results of the first phase, answering the second and third research 
questions. 
 

  



3 

2. Methodology 
The methodology will cover our chosen approach to the subject matter and how each of the 
stages of the study was approached. 

2.1 Research approach 
This study employed a mixed approach, utilising both quantitative and qualitative methods 
during different phases of the project. Quantitative analysis was used to analyse and find 
patterns and inconsistencies in the Forecast Accuracy Index (FAI) and Weighted Tracking 
Signal (WTS) data. Based on the results of the quantitative analysis cause and effect correlations 
were explored using a qualitative approach. The purpose of the second phase study was to 
generate suggestions for improvement in regards to the issues identified. This study seeks to 
explore a specific instance of the issue and does not seek to establish general findings that are 
universally representative. According to the reasoning of Davies & Hughes (2014), a qualitative 
approach is then suitable. It was decided that as the study sought to find in-depth insights, and 
that access to people within the organisations were given to the researchers, semi-structured 
interviews would be the main source of qualitative data. 
 
Apart from initial informal interviews and discussions aimed at building a basic understanding 
of the organisations involved, interviews were conducted in two main rounds. One primary 
round based on findings in the quantitative analysis, and a secondary round based on findings 
of the combined analysis of quantitative data and the first round of interviews. The second round 
of interviews was used to further explore key findings from the first round of interviews and to 
address any discrepancies between the quantitative and qualitative data.  

2.2 Data collection 
The quantitative data used for this study was pre-existing and organised in a database, and 
therefore no quantitative data collection was performed as a part of this study. The data included 
delivery schedules sent to a supplier by all of their customers between January 2018 and January 
2019. In this study, delivery schedules that refers to the demand in the calendar year 2018 is 
analysed. No further quantitative data were not collected and analysed. Due to this, this section 
is mainly focused on the qualitative data collection.  
 
Qualitative data collection was done through semi-structured interviews. The study focused on 
a local supplier (henceforth “Supplier” or “the Supplier”) and three different supply chain 
segments they are active in. Interviews were conducted at the Supplier and three of their direct 
customers. Of the direct customers, one is an OEM, one is a company doing pre-assembly for 
an OEM and one is an aftermarket organisation. In addition to these three, employees at the 
OEM that one of the direct customers does pre-assembly for were interviewed. A more detailed 
description of each supply chain segment is provided under section 4. The different types of 
customer organisations and different role of the Supplier in each of the supply chain segment 
was expected to give a good overview of different aspects that impact the accuracy of demand 



4 

forecasts. At all customer organisations people involved in order planning and monitoring were 
interviewed. Additionally, at the manufacturing and assembling customers, production planners 
were also interviewed. At the Supplier, production planners for the different component groups 
sold in the three supply chain segments were interviewed, along with procurement personnel. 
At the customer organisations interviews were focused around research question 2, gathering 
data on how forecasts were created and used within their respective organisations to build an 
understanding of the forecasting process and potential sources of inaccuracies. Interviews at the 
Supplier focused on gathering data on how they used the demand forecasts sent by customers 
and if they had experienced any negative effects connected to poor forecasting accuracy from 
their customers. Figure 1 shows a summary of the interviews conducted, with the company, 
role of the interviewee and the main topic of the interview. 
 

 
Figure 1. Summary of interviews conducted for the study. 
 
In those cases questions arose after the interview session interviewees were contacted to provide 
complementary information. Interviewees were all from the examined organisations and 
selection was done using Snowball sampling. Snowball sampling includes asking interviewees 
for others who may possess valuable information (Easterby-Smith et al., 2015), and was deemed 
a good fit for the study as it sought to identify the effects of inaccurate demand forecasts and 
potential mitigating initiatives across the organisation in production, planning and purchasing. 
The different functions were expected to have different perspectives and views on the 
forecasting process, its value and its issues. Therefore, for the interviewees in customer 
organisations that were interviewed, an attempt was made to interview employees with similar 
areas of responsibility and expertise in all organisations to provide additional opportunity for 
comparison. The key methods and theories applied during interviews to ensure accuracy and 
validity in the collected data will be outlined below. 
 
As a first measure to ensure accurate transcription of interviewee answers, interviewees were 
asked to agree to an audio recording of the interview being made. Since the study did not seek 
to confirm or deny a pre-formulated hypothesis, but rather served as an exploratory case study, 
it was decided to follow an interviewing strategy proposed by Steinar Kvale. Kvale (1996) 
suggests that the interviewer should be mostly reactive in how they ask their questions, in order 
to allow room for the expertise and knowledge of the interviewee. Follow-up questions should 



5 

be asked based on answers to open questions to further build on interesting aspects of the 
answer. This allows for greater flexibility during the interview, building primarily on the 
knowledge and experience of the interviewee rather than the ideas and hypotheses of the 
interviewer. It does however place importance on the attentiveness of the interviewer to ensure 
that well formulated and relevant follow-up questions are asked. In this study, both researchers 
were present for all interviews in an attempt to encapsulate more perspectives with follow-up 
questions. 
 
While having more than one interviewer is of marginal value in a structured interview, 
Bechhofer, Elliott & McCrone (1984) argue that it could be useful in a more unstructured 
interview format, providing several benefits. Utilising multiple interviewers allows for one of 
the interviewers to focus more on note-taking and observation of the direction of the interview 
and topics yet to be cover. Additionally, the more passive interviewer could also step in should 
the main interviewer get stuck and become unable to continue a line of questioning or switch 
to a new one. It was therefore decided that for this study, both researchers would be present to 
improve fluidity of the discussion, provide additional input, and allow for more detailed notes 
and recollection of the interview. The authors also highlight that if done correctly, the interview 
can be made to feel more like a discussion, helping the interviewee to relax. 
 
The study sought to employ the technique of laddering up to connect general perceptions to 
specific occurrences and underlying reasoning. Laddering up is described as a technique where 
the interviewer, often by asking ‘why’ questions, tries to build a deeper understanding of the 
value base of the interviewee (Easterby-Smith et al., 2015). This was deemed appropriate to the 
study, as suggestions for changes provided would need to be anchored to the needs and wants 
of different parts of the organisation to provide a positive effect. If there is no understanding of 
the underlying values and perception of the employees, any change would risk being misguided 
or miscommunicated resulting in the value would not being understood or realised. 
 
 

2.3 Analysis 
The approach of the data analysis can be described as an iterative process, where collection and 
analysis of data takes place simultaneously. By analysing the collected data early in the research 
process, the results imply the direction of the next steps of the data collection process. (Bryman 
& Bell, 2011) 
 
As mentioned in earlier section, the process of the study was divided into two phases. The first 
phase aimed to provide answer to the first research question and the process is of quantitative 
nature. The second phase aimed to provide answer to the second and third research question 
and the process is of qualitative nature. 
 
In the first phase, the database of information regarding delivery schedules were analysed by 
using performance measures and figures created in Qlik Sense. The database contained all 



6 

delivery schedules the Supplier received from their customers. In order to analyse the data, 
certain choices were made based on different criterias. The first criteria is related to the 
components produced at the Supplier. Depending on which component type produced by the 
Supplier, components were aggregated in different groups. When the analysis of the component 
groups were completed, customers of the Supplier that purchase roughly more than 10 % of the 
total volume for each component group were identified. Delivery schedules of these customers 
were then analysed in order to find irregular patterns. When irregular patterns were identified, 
these were saved for further analysis in the following phase. Depending on the amount and type 
of identified patterns, it guided the further work. During the analysis, irregular patterns were 
identified in three different supply chains, which implied for further investigation. Each supply 
chain starts with the Supplier providing their customers with components. In two of the 
examined chains, the customer is in direct contact with the Supplier. In the third chain, one 
party is located between the Supplier and customer, performing pre-assembly for the customer.  
 
The analysis in the second phase was based on the results from the analysis performed in the 
first phase. To be able to find the causes and effects of the irregular patterns identified, the data 
was further analysed by conducting interviews with the parties involved in the examined supply 
chains. The data collected from the interviews were sorted and organised in a suitable way. 
Different codes and labels of relevant subtopics were used to facilitate the analysis depending 
of its relevance to the research. According to Easterby-Smith et al., (2015) it is of great 
importance to prepare the data in a way that support the future analysis process and achieve a 
rewarding analysis. Relevant parts from the collected data should be filed systematically and 
categorised in a consistent way. Bryman & Bell (2011) state that it is common that qualitative 
research generate high amounts of data, in forms of documents, transcripts and field notes and 
argue that coding of data facilitates understanding of the data and support the possibilities to 
manage high amounts of gathered data. 

2.4 Validity 
According to Bryman & Bell (2011) reliability can be described as a question whether the same 
results of a study could have been achieved if the study was carried out with the same approach 
and methodology. Reliability can be divided into two subsections; external and internal 
reliability. External reliability refers to the possibility that a study can be replicated, and whether 
the findings can be applied across a general context. The approach used in this study can lead 
to difficulties to recreate the same social context in a future state, due to fact that the study is 
limited to only a few specific supply chains. Bryman & Bell (2011) argue that this is one of the 
main issues in qualitative research related to reliability, since it is almost impossible to achieve 
the exact same social context as used in previous studies. To achieve higher reliability, the 
internal reliability, which is concerned with whether the observations made by the researchers 
match the theoretical ideas developed (Bryman & Bell, 2011) , was considered in the study. 
Bryman & Bell (2011) state that the members of the research team can interpret observations 
in different ways, which can affect the results of the study in a negative way. To mitigate these 
risks, the same persons that conducted the interviews analysed it as well. 
  



7 

Bryman & Bell (2011) defines validity as the ability to ensure that the measured concept 
actually reflect the concept that is supposed to be measured. It can be argued that the external 
validity within the study is limited, since only a few relationships of different supply chains is 
considered. Although, the similarities between supply chains in the same context will enable 
usage of the findings in the study. 
 
To verify our findings, they were presented and discussed at a workshop for the FFI project, 
with participants from suppliers and OEMs in the Swedish automotive industry. This helped 
verify the findings as well as provide additional input for some analysis. 
 

2.5 Limitations 
This study only featured forecast data sent to one supplier, which meant that the analysis was 
unable to make comparisons to the situations of other similar suppliers and provide an idea of 
whether this case is representative for the Swedish automotive industry as a whole.  
 
For the study, we did not have access to data regarding the accuracy of the sales forecast for 
any of the customer organisations. Therefore, component demand data was used as a proxy to 
discuss the impact of sales forecasting methods. This is not necessarily representative, as we 
showed that some of the processes between the sales forecast generation and the final 
component demand forecast can introduce significant variations and errors to the forecast. 
Additionally, the lack of access to sales forecast accuracy data meant that we could not draw 
any conclusions regarding the exact impact of the flaws in the processes transforming sales 
forecasts to component demand forecasts in terms of loss of forecasting accuracy. 
 
While the study had access to significant quantitative data for the case studied, the qualitative 
data collected was somewhat limited. Therefore, the connections and implications found should 
be seen as examples and indications of what could happen, and not as generalizable results that 
can be directly applied to a different case.  



8 

3. Theoretical framework 
Our theoretical framework will cover literature in the areas of planning, forecasting and 
information sharing, covering both the processes and ICT-technology used to support them. 

3.1 Supply chain structure and information sharing 
The supply chain can be defined as “...a set of relationships among suppliers, manufacturers, 
distributors, and retailers that facilitates the transformation of raw materials into final products.” 
(Beamon, 1998). The structure and production philosophy of the supply chain is dependent on 
the market that it serves, and Christopher (2016) discusses two main philosophies and when 
they are applicable; lean production and agile. The author argues that lean production is suitable 
when demand variation is low and lead-times long, as it utilises long term planning to optimise 
resource usage and create efficiency. Agile, on the other hand, is suitable when variation is 
high, but lead-times short allowing quick response to market variations. In cases where there is 
both variation and long lead-times a hybrid strategy of postponement is suggested, utilising a 
lean production philosophy up to a decoupling point from which production is done with an 
agile philosophy. One example of such an arrangement is found in the paint industry where a 
limited number of base colours produced according to lean philosophies can be mixed into a 
near infinite number of colours at the retailer.  
 
While these different approaches to supply chain strategy are suitable under different market 
conditions, Mason-Jones et al. (2000) argue that elimination of system-induced uncertainties is 
crucial to the success of all three. This would in turn mean that continuous and accurate 
information flow is important in improving the efficiency and stability of the supply chain. Lee 
& Hwang (2000) discuss different types of information shared between companies in a supply 
chain. The specific types of information discussed in their article are: Inventory level, sales 
data, order status for tracking/tracing, sales forecast and production/delivery schedule. In 
essence the purpose of sharing these data is to create transparency and creating a common 
understanding of current operations and how the near future might look. For this study, sales 
forecasts and production/delivery schedules are the most relevant types of sharing, as they are 
forward looking and facilitate coordination of activities.  
 
Lee & Hwang (2000) argue that these two types of forecasts can help both customer and 
supplier in planning their production using Quantity Flexibility contracts, where the customer 
has flexibility in deciding delivery quantity and the supplier can get a warning of shifts ahead 
of time through the forecasts. Huang, Hsieh & Farn (2011) call the phenomenon “rolling 
forecast”, where customers share and continually update their demand forecasts. Henceforth 
the phenomenon will be referred to as Quantity Flexibility contracts, or QF contracts. While 
seldom discussed in academia, the practice is popular in industries (Huang, Hsieh & Farn, 
2011). At an agreed point in time, forecasts in QF contracts become fixed orders. Lau, Xie & 
Zhao (2008) found that early order commitment (EOC) provide benefits in reducing inventory 
levels. Benefits from EOC and sharing of future forecasts are however mainly seen on the 
supplier end, whereas the forecasts and commitments are made by the customer. The value of 



9 

utilising QF contracts are directly contingent on the accuracy of forecasts provided by the 
customer, which can prove problematic if customers act more out of self-interest rather than 
fully collaboratively (Huang, Hsieh & Farn, 2011). 
 
Lee and Hwang (2000) highlight a potential issue with QF contracts, as there is an incentive for 
the customer to overstate future demand to ensure that they can get the quantity they need. This 
forces the supplier to carry additional inventory at their own expense as they will act on the 
forecasts available. Huang, Hsieh & Farn (2011) also identify a tendency for customers to adjust 
forecasts to protect their own interests, especially when the customer is more powerful than the 
supplier. 
 
Tsay and Lovejoy (1999) take a supply chain perspective on QF contracts, concluding that QF 
contracts in essence is about distributing flexibility and the inventory burden. Inflexible 
contracts result in the customer needing to carry additional inventory, and highly flexible 
contracts result in the supplier needing additional inventory to guarantee delivery precision. 
The authors also highlight that QF contracts require significant changes to the procurement 
process, potentially resulting in organisational resistance in changing current behaviours. They 
conclude that QF contracts constitute a trade-off between procurement prices and inventory 
costs, responsibilities often distributed to different groups in the organisation potentially 
resulting in sub-optimisation of the usage of QF contracts. Tsay (1999) studies the usage of QF 
contracts in manufacturer-retailer relationships based utilising cost parameters for both the 
manufacturer and retailer, concluding that QF contracts can be used to increase net system 
profit. However these gains were only demonstrated when there were shared beliefs, and no 
information asymmetry between the parties. 
 
As for the practicality of transferring data, Christopher (2016) mentions EDI, Electronic Data 
Interchange, as a way to create visibility and share demand information within the supply chain. 
Lee and Hwang (2000) use QF contracts as an example of when EDI usage is suitable, as these 
contracts are dependent on fast transfer of updated information for the supplier to be able to 
plan their production. 

3.2 Electronic Data Interchange 
EDI is a form of structured business communication for transmitting information in a 
standardised format (UNECE, 2019). In many supply chains EDI is crucial in ensuring smooth 
information sharing (Christopher, 2016). Transmitted information, such as demand 
information, is used in planning operations. A commonly used format is the UN/EDIFACT 
standard (UNECE, 2019), providing specifications for the structure and content of EDI 
messages to ensure correct transmission and interpretation of data. In order for electronic data 
interchange to be effective, a standard structure for messages is needed (UNECE, 2019). One 
framework standard for interorganisational data exchange is the UN/EDIFACT (United 
Nations/ Electronic Data Interchange for Administration, Commerce and Transport) (UNECE, 
2019). This study is mainly concerned with the usage of data transmitted by one of the message 
standards in the UN/EDIFACT, namely the DELFOR (Delivery Forecast) message. DELFOR 



10 

messages are used by customers to give a forecast of their future demand to their suppliers, so 
that they can plan accordingly. While EDI can be a powerful tool Lee and Hwang (2000) state 
that the existence of multiple standards can create flexibility issues as managing the usage of 
multiple standards can be difficult, should the company operate in multiple industries. There 
are challenges associated with the usage of EDI, but there are also potential benefits.  
 
Hart and Saunders (1997) identified mutual trust between the actors as an important factor in 
determining how extensive and successful EDI usage between them would be. They identified 
that in relationships where persuasive and collaborative approaches, rather than coercive 
approaches were used. Dearing (1990) argues that business can gain three types of benefits from 
the usage of EDI; direct, indirect and strategic. Direct benefits come from automating and 
making information transfer electronic and enabling better storage of this information. Indirect 
benefits come from enabling organisations to change the way that they operate, such as 
switching to JIT production. Strategic benefits come from the possibility to create closer ties 
with customers and suppliers, and long term collaboration and integration to create cost benefits 
on both ends of the relationship. 
 
To improve on EDI usage, metrics for meta-analysis of EDI messages have been proposed, one 
such metric is the FAI metric presented by Odette in 2012 (Odette, 2012). In a way, this metric 
can be seen as a way to facilitate transparency and communication between business partners 
to build the strategic benefits discussed by Dearing (1990). By using FAI and WTS data as a 
basis for discussion, companies could improve the quality of their communication of business 
data by jointly working to improve forecasts. 

3.3 Measurements for forecasting errors 
FAI is aimed at giving organisations a better understanding of the accuracy of forecasts, by 
giving possibilities to measure and compare the quality of the forecasted quantity to the actual 
required quantity (Odette, 2012). FAI functions as a 0 to 100% scale measuring the accuracy of 
delivery forecasts, with 100% indicating perfect forecast accuracy. The mathematical definition 
of FAI can be found in appendix 1, when calculating aggregate FAI for multiple 
components/customers, individual FAI scores are weighted by share of sales. The purpose of 
the metric is to be used in stabilising efforts to mitigate demand volatility and counteracting the 
‘bullwhip’ effect in the supply chain. FAI is also aimed at being a tool for feedback in S&OP 
and for improved transparency, communication and feedback in the customer-supplier 
relationship (Odette, 2013). To complement the FAI measurement a WTS was introduced to 
indicate whether inaccuracies in forecasts were due to over or underestimation of demand. The 
mathematical definition for the WTS can be found in appendix 1. WTS is a number between -
1 and 1, with 1 meaning that every misestimate is an overestimation and -1 meaning that every 
misestimate is an underestimation. A WTS of 0 means that there is no tendency towards either 
overestimations or underestimations. 
 
A similar measurement, Forecast-Qualitätskennzahl, has been established by the German 
‘Verband der Automobilindustrie’ (VDA). In their report, VDA (2008) defined a measurement 



11 

that measures the deviation of forecasts. In the same report guidelines for what constitutes good 
and poor performance. This will be presented in figure 2. The VDA eventually refined Forecast-
Qualitätskennzahl, publishing a report defining FAI and WTS (VDA, 2010).  Forecast-
Qualitätskennzahl, in turn, is similar to Mean Absolute Percentage Error (MAPE). MAPE, 
while a good measurement of forecasting errors in most cases, has a tendency to favour 
overestimations as underestimations are bounded by 100%, whereas overestimations are 
unbounded (Armstrong & Collopy, 1992). FAI, however, avoids this issue by limiting 
overestimations at 100%.  
 

 
Figure 2. performance levels for forecasts. Adapted and converted to FAI measurements from VDA (2008). 

3.4 Forecasting 
To be able to make decisions regarding future activities within a manufacturing organisation, 
information and assessment of the future state is needed. By evaluating the external factors that 
could influence future operations, qualified decisions can be ensured to a greater extent. Jonsson 
and Mattson (2009) define a forecast as an assessment of possible external factors that could 
affect a company, and the company cannot control themselves. Some external factors, such as 
market demand for certain products due to marketing and pricing are possible to impact, while 
other external factors like market conditions are difficult to influence. 
  
Jonsson and Mattson (2009) state that forecast methods used in demand planning are mainly 
qualitative and quantitative methods. Qualitative methods are primarily based on subjective 
estimations made by experienced workers. These can be anything from basic guesses to more 
qualified assessments. This method can be compared to quantitative methods, which are largely 
dependent on calculations. The calculations can originate from different types of historical data 
related to usage or sales of material. According to Jonsson and Mattson (2009) forecasts are, by 
default, always incorrect. The ability to conduct more accurate forecasts must be weighed 
against the required resources. Usage of resources are related to costs and needs to be lowered 
as much as possible. By comparing the value added by accurate forecasts to the costs it is 
possible to decide whether an organisation should invest more in creation of forecasts. 
Vollmann et al. (1988) argue that organisations should aim to reach as low cost per forecast as 
possible. Organisations needs to have simple and effective forecast methods to be able to lower 
the costs related to forecasts.  
 



12 

According to Jonsson and Mattson (2009) a company can use the master production scheduling 
to plan the sales and production operations. By defining the quantities that should be delivered, 
it is possible to plan in which period the quantities should be manufactured. To be able to deliver 
all required orders, it is needed to specify all components and materials included in the finished 
product. Jonsson and Mattson (2009) further state that material requirements planning can 
support the material planning by calculating when material requirements are expected to arise 
and orders are needed to be released.  
 
In a supplier perspective, customers send forecasts of longer periods of time which enable the 
supplier to plan their upcoming production processes. Further on, when more accurate demand 
information becomes available the customer demand can vary from the original plan. This leads 
to deviations and the supplier has to recalculate their production schedule. Order forecasts that 
changes with short intervals can make the supplier’s MRP system nervous, due to the 
difficulties to plan material requirements and production in a good way. (Li and Disney, 2017) 
According to Li and Disney (2017) the most effective method to cope with these kind of 
variations is to use some sort of buffer.  
 
Forecasting varies between different organisations. Rieg (2010) uses a conceptual framework 
of three independent variables that affect forecasting accuracy: 
 
● Statistical methods 
● Hardware & software 
● People, processes & organisation  

 
These three are then supplemented by a moderating variable; Uncertainty/volatility. Statistical 
methods is the choice of forecasting method(s) and hardware & software is the computer 
equipment used to assist the forecasting process. The impact of people, processes and 
organisation comes from among other factors training in the forecasting processes (Merrik et 
al., 2006). Gruber & Venter (2006) identified that learning effects do occur in forecasting, and 
that there are significant differences in how forecasting is conducted between different 
organisations. 
 
According to Armstrong (2001), forecasters often make significant mistakes in the forecasting 
process. Many of these mistakes are related to not utilising statistical methods correctly or 
basing forecasts on incorrect assumptions. The author underlines the importance for 
practitioners to safeguard the integrity of the process, and suggests checklists to ensure proper 
procedures as a way to do so. 
 
There are alternative, machine learning approaches to forecasting. In a meta-study, Zhang et al. 
(1998) evaluated Artificial Neural Networks (ANN) and featured a summary of articles 
evaluating the forecasting performance of ANNs compared to traditional statistical methods for 
various time series. The authors summarised the conclusions of 22 articles, with somewhat 
mixed results. 3 articles concluded ANNs performed worse or significantly worse, 8 concluded 
ANNs performed better or significantly better, 3 concluded ANNs were comparable and 8 



13 

concluded ANNs were situationally better and otherwise equal or slightly worse. Kang (1992) 
found that ANNs outperformed other methods when the time series featured seasonal or trend 
patterns. Kang (1992) also found that ANNs needed less data available to give accurate 
forecasts. 

3.5 Usage of forecasts in Production/logistics planning processes 
According to Lalami, Frein and Gayon (2017), the aim of production planning is to identify 
how many products that is needed to be produced, when they have to be produced and in which 
production plant the production should take place. Meyr (2004) has examined the supply chain 
planning in an automotive industry and defines the general flow of materials upstream in an 
automotive supply chain as convergent. Coordination difficulties is common in these type of 
supply chains, due to production capacity, manpower issues and incoming goods. To manage 
these difficulties an organisation needs to adapt not only an effective internal information flow, 
but also enable collaboration in terms of information sharing with other parties within the 
supply chain. Meyr (2004) states that planning processes differs between companies, and 
further mention forecast-driven and order-driven planning as two possible ways to manage the 
production planning. Forecast-driven planning is usually used in long- or mid-term decision, 
and order-driven planning is used in short-term decisions. Wei et al. (2017) describes three 
hierarchical levels within supply chain management (strategic, tactical, operational), where 
each level is related to different decision processes. Decisions at the strategic level is usually 
high level decisions, including decisions such as location of production plants or create new 
transport networks. At the tactical level, the process itself is defined by addressing cost control, 
customer satisfaction and risk management. These issues are connected to activities like 
production schedules, transportation and warehousing. The operational level refers to the 
regular day-to-day processes, and concern a more detailed plan of the work-in-progress, 
inbound and outbound logistics. 
  
Lalami, Frein and Gayon (2017) defines a planning process named rolling horizon planning. 
When this type of planning process is used, the production plan is updated periodically by 
referring to the most reliable information available. The reason to update the production plan is 
according to the authors, either because of a shift of planning horizon or uncertainty. Demand 
forecasts is usually used in production planning and by regularly reviewing the forecast, it 
enables the organisation to be reactive to forecast errors.  
 
Myrelid (2017) found that the quality of demand schedules shared by customers to their 
suppliers affect the production planning of the supplier. Among the determinants of quality 
used, was reliability which is the accuracy of the delivery schedule compared to final deliveries. 
Reliability is used in the same way that ‘accuracy’ is used for this study, and the study 
performed by Myrelid (2017) confirms the connection between customer demand schedules 
and supplier production planning. Karlsson & Ragnarsson (2014), a previous thesis within the 
same overarching project as this thesis, found that suppliers to a large degree rely on 
information sent by their customers, and are very limited in their own utilisation of an S&OP 
process. Some suppliers deem information sent by the customers as insufficient to base their 



14 

own long term planning on, citing difficulties in making accurate decision for machine 
acquisition and employee training due to short planning horizons. This implicates that in most 
cases, if the customer is unable to create accurate forecasts, suppliers will be heavily affected.  

  



15 

4. Case description 
The subject studied is forecast accuracy in demand forecast (DELFOR) messages sent from the 
customers of a selected supplier to the supplier. The development of demand forecasts is done 
according to the general process described below at all three of the customers studied. 
 
The raw data used to generate the forecast is collected from sales forecasts. These forecasts are 
then run through the MRP system, where production and/or inventory policies, lead times, unit 
loads and other factors are taken into account. Through this process, the initial sales forecast is 
broken down and processed to generate a delivery forecast which is subsequently sent to the 
supplier. These forecasts will contain variations in demand volumes. For the purpose of this 
study, variations is used to describe the differences in final order volumes, either as temporary 
increases or decreases, or long term more trend-like changes. Important to note is that this study 
does not examine variations in end customer demand, but only the variations in demand from 
the customer organisation to the supplier. In the process described above, variations can stem 
from two different sources. Firstly, variations can be a product of changes to the end customer 
demand, whereby the total production volumes of the supply chain as a whole change. 
Secondly, variations in end customer demand can be amplified or reduced internally, through 
the planning process to give the final variation. Examples of factors that may affect the variation 
are production batching, order batching, inventories, safety stock and MRP nervousness. The 
purpose of the forecast is to predict variations ahead of time. 
 
Forecasting accuracy is the ability to predict these variations ahead of time, and subsequently 
indicate correct demand volumes to the Supplier as far ahead of time as possible. In this study, 
possibilities to affect end customer demand will not be considered, but only the possibility to 
predict them through sales forecasts, and how these are then processed and used to generate 
demand forecasts sent to the Supplier. To facilitate for the reader, this study differentiate 
between the definition of component and product. Component is used to refer to components 
created and delivered by the Supplier or pre-assembly companies. When components are 
received by the OEM and put together it creates a product.  

4.1 Supplier studied 
The supplier studied for the purposes of this study is located in Sweden, supplying metal 
components in the automotive industry. The Supplier is a wholly owned subsidiary of a major 
OEM, operating independently under a different brand. Over the past years, the Supplier had 
an average yearly revenue in the area of SEK 1.8B. The Supplier is a tier 1 and tier 2 supplier 
in the automotive industry delivering primarily to OEMs with some deliveries to companies 
doing pre-assembly for OEMs. Components used by the OEM studied in this case are sent to a 
pre-assembly company. The Supplier has over 400 employees operating at their plant. They 
supply components to several major OEMs in the automotive industry, mainly within the 
European Union. Roughly 70% of revenues are from within Sweden, 28% from the rest of the 
EU, and 2% from countries outside of the EU. The company product line consists of three main 
component groups; C10, D10 and H10. 



16 

 
The Supplier delivers components for both cars and heavy vehicles. Around 55% of revenues 
are from the heavy vehicles industry and the rest is from the car industry. A few large OEM 
constitute a majority of their revenues, whereas the supplier is just one of many for the OEMs 
power in the relationships is heavily skewed in favour of the OEMs. Most customers only buy 
components from one component group, resulting in limited overlap in the customer bases of 
the different component groups. There is some overlap in the customer bases of component 
groups C10 and D10 but none with H10  

4.1.1 Logistics and order process 
The supplier is using an arrangement similar to the Quantity Flexibility contracts described in 
the theoretical framework. Customers continuously send demand forecasts outlining their needs 
for the future through a delivery schedule. Once a delivery date is close, the schedule is “frozen” 
and the quantity cited in the delivery schedule acts as a formal order. The structure of these 
arrangements varies between customers. Some customers have a defined latest date to make 
changes ahead of delivery after which the order is frozen. Others have no such period, with the 
latest delivery schedule always acting as the order, this gives the customer more flexibility at 
the cost of uncertainty for the supplier. Customers generally send demands aggregated on a 
monthly level for demands far into the future, weekly demands at a medium time horizon and 
daily demands for the near future. For the customers studied, daily demands are sent for roughly 
the next two weeks. 
 
The supplier attempts to offset some of this uncertainty through safety stocks. They keep safety 
stocks of finished components depending on weekly volumes. They also keep safety stocks of 
materials equalling no more than five days’ worth of production. Stocks for screws and similar 
small components are larger as the costs associated with keeping them in the inventory is low. 
There are no interim storages for work components apart from a small buffer before final 
assembly (roughly 6 hours). 
 
Lead times for inbound logistics vary from one day for some generic goods to three months for 
some special components. The plans sent to their suppliers are based on the forecasts received 
from customers and cover the same planning horizon (roughly one year). 

4.2 Customers studied 

4.2.1 OEM 1 
OEM 1 studied is a multinational, major producer of heavy vehicles. This study focuses on one 
of their sites in Sweden. Production flow at this site is JIT type, with sequential delivery of most 
major components, among those are components from a pre-assembly company between OEM 
1 and the supplier studied. 



17 

Material & information flow 

Figure 3 outlines the flow of information and materials between OEM 1, the pre-assembly and 
the supplier in the supply chain studies. OEM 1 sends information about needs both in the form 
of DELJIT messages detailing the components, sequence and time for delivery, and DELFOR 
messages forecasting future demand. The pre-assembly company then makes a decision for 
their procurement based on the composite information from DELJIT and DELFOR messages 
from the OEM. This is then sent to the supplier in the form of a DELFOR message. Deliveries 
from the supplier to pre-assembly are not sequential and usually in daily batches. Both the 
supplier and pre-assembly company carry safety stocks to guard against issues in logistics or 
production, as penalty fees for failing final delivery to the OEM are high.  
 

Figure 3. Material and information flows. Information flow(s) denoted by gray arrows, and material flow(s) by 
black arrows. 

There is no direct formal flow of information between OEM 1 and the supplier. As both the 
pre-assembly company and the supplier were previously owned by OEM 1, many of the supplier 
contracts of both companies are held by OEM 1. Because of this, there is significant 
communication between the organisations on other matters, but in terms of production and 
material planning only EDI messages are sent. Figure 3 shows an overview of important 
production- and demand related material and information flows. 
 
The total quantity of tied up capital in inventories and production throughout the segment of 
amounts to roughly 25 days’ worth of production. This results in stability for OEM 1, but can 
conceal issues with production and communication in the supply chain. For example, at the 
time of the study there had recently been a disruption between the Supplier and one of their 
subcontractors due to a contract-related issue. Deliveries to the Supplier were limited or 
completely stopped for a period of several weeks, however the effect of this never reached OEM 
1 in the form of missed or delayed deliveries.  
 



18 

 
Figure 4. Inventories, buffers and materials bound in the segment of the supply chain studied. 
 
There is a total of around 24 working days’ worth of production in inventories for this segment 
of the supply chain. Of this, roughly 19 working days’ worth of production is kept at the 
Supplier. Based on an approximation of 260 working days in a year, the inventory turnover rate 
of the Supplier for this component group is roughly 13.7. Based on financial data from the 2017 
annual report of the Supplier, their overall inventory turn rate was 12.2 for FY2017. 

4.2.2 OEM 2 
For this study, interviews in the aftermarket service organisation, Customer E, as well as one 
of their main production facilities of a second OEM, Customer D, were conducted. OEM 2 is 
a global car manufacturer. 

Material & information flow 

Both Customer E and Customer D uses EDI messages in UN/EDIFACT format, primarily 
DELFOR, when communicating demand information with their suppliers. There are however 
some differences between the two in how they communicate with their suppliers, and in the 
nature of the supply chain, as one is a production supply chain and one is an aftermarket supply 
chain. Some of the differences and attributes of each supply chain segment studied will be 
presented below. 

Customer E 

Customer E is the aftermarket organisation of OEM 2, supplying spare parts globally for cars 
produced by OEM 2. In total they supply over 50 local offices/warehouses from their main 
warehouse, which is where goods from the Supplier are sent. The Supplier sends daily deliveries 
to the main warehouse, mostly following the DELFOR. However, Customer E and the Supplier 
have regular contact by phone and e-mail and if needed they forego the DELFOR messages, 
being willing to make changes and adaptations if either party deems it necessary. The Supplier 
delivers goods to Customer E every day, however, quantities are generally higher on Mondays 
and taper off over the course of the week. Deliveries from Customer E to each of the local 
offices and warehouses are generally less frequent, and based on inventories at each warehouse 



19 

using EOQ with trigger points and safety stocks. Figure 5 shows an overview of the flow of 
information and materials/goods. 
 

 
Figure 5. Material and information flows. Information flow(s) denoted by gray arrows, and material flow(s) by 
black arrows. 
 
Figure 6 shows average inventories within the supply chain segment. Inventory levels are given 
as number of (working)days’ worth of average demand for one of the high-volume components 
sent from the Supplier to Customer E. In total, there is approximately 32 working days’ worth 
of materials and components in this segment of the supply chain at any given time. Of this, 
roughly 25 working days’ worth of production is at the supplier. Based on an approximation of 
260 working days per year, the inventory turn rate of the Supplier is 10.4 for this component 
group. 

 
Figure 6. Inventories, buffers and materials tied up in production within the supply chain segment. 

Customer D 

Customer D is one of the main production facilities of OEM 2, and purchases components 
directly from the Supplier. There is no communication between Customer D and the Supplier 
outside of EDI messages and transactions, unless there are significant issues and most issues 
are solved by referring to the contract terms. There is little to no joint problems solving or 
adaptation outside of the contract terms unless it is the last resort. Goods from the Supplier to 
Customer D are delivered on a daily basis, according to DELFOR messages sent by Customer 
D. The Supplier is given 48 hours of frozen orders. Goods are batched and non-sequential. In 
total, 9 components and variants are delivered by the Supplier to Customer D, with 2 



20 

components amounting to roughly ⅓ of the total quantity each. Figure 7 gives an overview of 
the main modes of communication and deliveries. 
 

 
Figure 7. Material and information flows. Information flow(s) denoted by gray arrows, and material flow(s) by 
black arrows. 
 
Figure 8 shows inventories and capital tied up in production for the segment of the supply chain 
studied. Quantities are given in days’ worth of average daily quantities. The numbers given are 
for the component(s) making up the largest portion of total quantity, as these flows are more 
stable over time as production batching and unit loads makes a smaller impact on quantities in 
inventories at any given time, as well as these components having the biggest impact on the 
total capital bound. The figure gives an idea of where there are buffers, and where the supply 
chain is most susceptible to disruptions. 
 

 
Figure 8. Inventories, buffers and materials tied up in production within the supply chain segment. 
 
In total, there is normally 27 working days’ worth of components in this segment of the 
supply chain at any given time, mainly at the Supplier. 

Available data on forecasts 
For the purposes of this study, access to a database containing all DELFOR messages received 
by the supplier from 1/1/2018 until 31/1/2019 was given. The data was accessed through a 
frontend application which allows aggregation of forecasts based on customer, delivery address, 
component, demand week and when the forecast was received. Separate to this database, 
information on component groups and translations of customer ids was available.   



21 

5. Empirical results 
The empirical results will feature the general patterns found in the data to give a comprehensive 
overview of the current situation in terms of forecasting accuracy. A summary of data for each 
of the studied customers will also be given. Some of the more specific and in-depth data for the 
major customers will be given directly in the discussion for convenience reasons, as it is 
frequently referenced. In addition to the general data, summaries of key data and insights from 
each of the interviews conducted will be given in the empirical results.  

5.1 General patterns in FAI and WTS 
Below a selection of graphs constructed from aggregated data regarding the main component 
groups C10, H10 and D10 will be shown. Component group H10 consists of components 
delivered to the heavy vehicle industry, whereas groups C10 and D10 are two different types 
of components delivered to car manufacturers. Patterns found in these data, as well as 
implications and conclusions to be drawn from them, will be discussed in detail under the 
analysis chapter. We will characterise forecasting accuracy as the FAI score of the forecast, and 
forecasting inaccuracies as the difference between the FAI score and 100%. FAI and WTS 
scores are calculated for different time “lags”. The lag is the time from the date the forecast was 
sent to the Supplier to the date of the forecasted demand, e.g. for a delivery schedule sent in 
calendar week 32, the forecast at 8 weeks lag would be the forecasted demand for calendar 
week 40. Some calculations are made as a weighted FAI for several different lags. A FAI score 
labelled with lag (2/4/6) is a weighted calculation of the FAI scores of 2 week, 4 week and 6 
weeks lags, respectively. In all cases where weighted FAI scores are used, three different lags 
are used and they are all weighted equally and the number is the average of the different time 
lags. 

5.1.1 Patterns for the different component groups 
Figure 9 shows a comparison of FAI scores for each component group for for different lags, 
ranging from pure 2 week delivery schedules to schedules 6-10 weeks ahead of time. In the 
data, we can see that FAI scores for group H10 are more than 9 percentage points higher for the 
2 week short term prognosis. However, differences in FAI scores become smaller for longer 
planning horizons. 
 



22 

 
Figure98. FAI scores for each component group, for different lags. Lags are given in weeks. 
 
Data for the WTS connected to the FAI scores in figure 9 can be found in figure 10. The WTS 
scores for component groups C10 and D10 are significantly above 0, revealing that customers 
ordering these components consistently overestimate their demand in delivery schedules. For 
group H10 WTS scores are close to 0. Customers ordering these components also have no 
tendency to consistently over- or underestimate their demand. 
 

 
Figure 10. WTS numbers for each component group, for the same lags as used in figure 8. Lags are given in weeks. 
 



23 

Figures 11, 12 and 13, show the FAI scores for each component group from delivery schedules 
during the year 2018. The graphs start in week 8, as six weeks of data is required to make the 
calculations since the longest lag is 6 weeks. None of the component groups show any clear 
patterns for how the FAI has changed over the course of the year. There are varying levels of 
volatility, with H10 being the most stable, and C10 being the most volatile. All three component 
groups have significant changes in FAI between the weeks 28 and 31, which are traditionally 
vacation weeks for factory workers in Sweden. This could explain the difficulty in forecasting 
production volumes for these weeks, subsequently leading to lower FAI scores. 
 

 
Figure 11. FAI scores for group H10 during 2018, using 2, 4, and 6 week lags. Scores are given for each calendar 
week from 8 to 52 (6 Week FAI can be calculated at Week 8 at the earliest 
 

 
Figure 12. FAI scores for group D10 during 2018, using 2, 4, and 6 week lags. 
 



24 

 
Figure 13. FAI scores for group C10 during 2018, using 2, 4, and 6 week lags. 
 
Figure 14 shows a summary of FAI scores with standard deviations for each component group 
with different lag settings. The standard deviation here becomes a measurement for the 
consistency in accuracy of forecasts, with a low standard deviation meaning that forecasts are 
roughly as accurate every week. We measured this to capture the stability and consistency of 
forecasting methods employed. While comparisons of the standard deviations between different 
component groups can be influenced by the volatility of demand for the final product, the 
standard deviation still gives a rough idea as to how stable and consistent the forecasting of 
customers is. The average FAI scores and standard deviations further highlight the patterns seen 
in figures 11, 12 & 13, with forecasts for H10 generally being more accurate as well as more 
consistent in their level of accuracy. 
 

 
Figure 14. Summary of average FAI scores with standard deviations. 



25 

5.2 FAI statistics for major customers 
Figure 15 shows the same information as figure 13 but only for the most significant customers 
of each group (roughly 10% or more of total volume for that component group). Apart from 
Customer B and Customer E, all major customers had equal or better FAI scores compared to 
the average of their respective component group. Also worth noting is that differences between 
the component groups are smaller when only major customers for each group are considered. 
Customer H is the company doing pre-assembly for the OEM studied, and Customer G is a 
company in the same corporate group as the OEM studied. Comparing the standard deviations 
for FAI scores of major customers purchasing from the same component groups can give an 
idea of how stable the forecasting methods for each customer are. 

 
Figure 15. Summary of FAI scores for major customers of each component group.  



26 

The two customers with the worst FAI performance were Customer B and Customer E. 
Customer B was expected to have worse performance, as they are not as mature in their usage 
of ERP systems and EDI communication as the other customers. In addition to this, transports 
to Customer B are very long distance transports, resulting in batching and less frequent 
transports. Changing the delivery date of one of these batches one day forward or backward, 
for example, will have a large effect on FAI. This is shown by the standard deviation, as some 
forecasts are fully accurate, whereas others are very inaccurate. This study focused primarily 
on the supply chain segments connecting the Supplier to OEM 1 and OEM 2, and as such 
Customer D, Customer E and Customer H are of particular interest. A more detailed 
presentation of their respective forecasting accuracy, as well as any unique patterns for each 
will be presented below. 

5.2.1 Customer D 
Customer D has a high forecasting accuracy when compared to other customers of the Supplier, 
particularly for long forecasting horizons. Figure 16 summarises the forecast accuracy of 
Customer D for different forecast horizons ranging from 4 to 26 weeks. The low forecasting 
accuracy for 8 week forecasts is likely due to when Customer D splits their demand from 
monthly to weekly. Depending on the weekday, the demand for same weekday 8 weeks ahead 
may be given with different levels of aggregation, which will confuse the algorithm used by the 
software when calculating accuracy. For example, some forecasted demands will appear as 
monthly demands and some as weekly, but both will be treated as weekly demands by the 
algorithm and compared to weekly reference demand when calculating FAI. The 8-week 
accuracy is likely due to calculation difficulties, as bucketing is somewhat inconsistent at this 
point, and should not be considered as too serious as in practice planners can clearly distinguish 
between weekly and monthly demand numbers. 
 

 
                       Figure 16. forecasting accuracy of Customer D for different forecast horizons. 
Customer D has a comparatively high forecasting accuracy for the components sent most 
frequently from the Supplier. Particularly for forecasting horizons beyond 12 weeks, where 



27 

many other customer quickly fall off in terms of forecast accuracy, Customer D manages to 
maintain a high level of accuracy. 

5.2.2 Customer E 
Customer E exhibits a very unique pattern compared to the other customers, particularly in the 
fact that there are significant changes to their performance across the period studied. 
 

 
Figure 17. 2 week FAI for customer E for 2018. 
 
As can be seen in figure 17, 2 week FAI for Customer E follows completely different patterns 
before and after week 29, which is during the Swedish industrial vacation period. Up to and 
including week 29, Customer E had an average FAI of 60.7%. From week 30 and onwards their 
average FAI was 96.0%. The difference is remarkable, and in the underlying data it appears to 
be almost momentaneous. Interestingly, Customer E had a lower ratio of changes within 2 
weeks of delivery (number of orders changed less than 2 weeks before delivery compared to 
the total number of orders) than the median for all customers. Their changes were however 
more significant in terms of volume. A few orders appear to have been modified by a factor of 
5, mere days before delivery. In addition to lowering the frequency of late changes, their 
magnitude was also lowered. The nature of their misestimates also changed. Prior to week 30, 
almost all misestimates were overestimations. After week 30 WTS scores were more balanced, 
with a slight tendency towards underestimations, as can be seen in figure 18. 
 



28 

 
Figure 18. WTS numbers for Customer E over the course of 2018. 
 
Looking at the forecasting accuracy of Customer E for different planning horizons, from 4 
weeks up to 26 weeks, reveals some interesting patterns to be considered. The forecasting 
accuracies for the different horizons are given in figure 18. The accuracy of forecasts appear to 
be roughly the same, regardless of the planning horizon. Customer E does not appear to have 
more reliable information regarding demand 4 weeks ahead than they do for demand 26 weeks 
ahead. 
 

 
                       Figure 19. forecasting accuracy of Customer E for different forecast horizons. 
 

5.2.3 Customer H 
Customer H has a comparatively high forecasting accuracy for horizons between 4 and 10 
weeks. For longer horizons their forecasting accuracy decreases rather quickly to level out at 



29 

roughly the same accuracy as most other customers. Figure 20 summarises forecasting accuracy 
for Customer H for different planning horizons. Similar to Customer D, Customer H also has 
one planning horizon with very low accuracy. This is also due to changes in demand buckets 
18 weeks ahead of delivery, similar to the dip in accuracy at the 8 week horizon for Customer 
D. 
 

 
                       Figure 20. Forecasting accuracy of Customer H for different forecast horizons. 

5.3 Interview data 
Interviews were conducted at OEM 1 and OEM 2. Interviews were done at two different 
organisations within OEM 2. Interviews at OEM 2 were done both at the aftermarket 
organisation as well as at one of their main production plants. The relationships and the nature 
of material flow and demand profiles were expected to be different between the two OEMs, but 
also between the two different internal organisations within OEM 2. 

5.3.1 Interviews at OEM 1 

Material planner at OEM 1 

As an exploratory interview at OEM 1, a material planner, from here on referred to as the 
Planner, was interviewed to better understand the information, and material, flow between 
OEM 1 and Customer H. The Planner is the main contact person for questions and concerns 
from Customer H, and forwards or otherwise addresses concerns that Customer H raises. In 
addition to the formal information flow of DELFOR and DELJIT messages, the planner also 
maintains continuous contact with Customer H. Orders, and sequencing, are frozen 3 weeks 
ahead of time, and any changes beyond that point have to be done manually and approved by 
Customer H before the order can be changed. OEM 1 has very limited information on the 
current situation in production at the Supplier and only has extensive information one step 
backwards in the supply chain. 
 



30 

Outside of the 3 weeks of frozen production schedule, OEM 1 also sends forecasts outlining 
approximate needs for the coming 1 months. Changes that are made within the freeze period 
are due to changes in the specifications from the end customer. The components that are part of 
the flow from the Supplier to OEM 1 through Customer H are rarely subject to change in 
specification. Almost all changes to this flow within the freeze periods are due to cancellations 
of end customer orders, which are rarer. The Planner did however state that there had previously 
been some issues with certain end customers completely changing orders a few weeks before 
delivery to the extent where it did affect the flow from Customer H, and in turn from the 
Supplier.  
 
The modules delivered from Customer H to OEM 1, are large and inefficient to store. As such, 
at most one day’s worth of production can be stored at the OEM 1. Therefore, any stop in 
production will lead to deliveries being postponed at Customer H until production can catch up, 
which can lead to carry-on effects. There is also a possibility for the logistics firm to store some 
modules at their hubs in case of a major stoppage. There has however been no major production 
stoppages in recent history.  
 
If Customer H has any questions regarding delivery schedules they contact the Planner, who 
either forwards the question to the relevant function and then sends the answer back to Customer 
H, or established direct contact between Customer H and the relevant function for more 
complex issues. Any issues regarding changes to delivery schedules also pass through the 
Planner. If Customer H has any issues with changes in delivery schedules, they are mostly 
concerned with changes made with short notice, and rarely anything concerning forecasts. 

Volume and program manager 

To improve insight into the actual forecasting process, two employees working with production 
program management were interviewed to build an understanding of the forecasting process 
and potential sources of issues further upstream in the supply chain. 
 
OEM 1 has sales offices in each geographical area where they are present. The sales 
organisation creates volume forecasts for each of the two main product lines for every market. 
The forecast follows a cyclical process where initial estimates are collected, and then discussed 
in terms of reasonability before to arrive at a final forecast that is sent to the planning department 
which updates the total production program two times per month. The sales department collects 
data from their customers as a part of their process to build forecasts. Information about major 
infrastructure or industrial developments is also incorporated as this will increase demand of 
heavy vehicles. The collected data is compiled to give an idea of the potential development of 
demand going forward, and is put into relation to historical sales figures. In the end, however, 
data pertaining to future demand is qualitative in nature and the forecasts are heavily reliant on 
experience and the personal judgement of the forecasting personnel. OEM 1 has started looking 
into how they can utilise more quantitative data in their forecasts, such as GDP and other 
indicators of economic performance of each market to predict future demand. Other companies 
within the same corporate group have been using such measurements successfully in their 
forecasting process, improving accuracy of forecasts. 



31 

 
Due to the complexity of the product structure and its variants, forecasts for each variants are 
created by automatically breaking down total volume forecasts for each country using historical 
data on distribution of each variant. Based on the production program, each factory then breaks 
down the production into material requirements, taking into account lead times, unit loads, 
inventory levels among other factors to create demand forecasts for each individual supplier. 
 
The size and complexity of the company, its markets and product range makes it virtually 
impossible for anybody to have a comprehensive picture of everything.  

5.3.2 Interviews at Customer H 

Assembly planning manager  

Delivery schedules are sent from OEM 1 to Customer H one time per day via EDI. The 
assembly planning manager analyse the plan each morning to be able to plan the daily 
production. He believes that the quality of the delivery schedules sent from OEM 1 is good. 
Customer H can experience some issues if the mix of requested components is changed, due to 
long lead time from some of their suppliers. They got the ability to handle small changes if they 
are small and not to close date of delivery. The assembly planning manager has daily contact 
with the material planner at OEM 1 to mitigate potential issues in the material flow. The 
material flow to OEM 1 is perceived as generally being stable, with some exceptions. OEM 1 
manufactures some vehicles that are sent in “lego boxes” from Sweden to certain geographical 
markets. Orders for “lego boxes” from the international factories are batched before they are 
sent to the Swedish production plant. This subsequently results in an equivalent batch order for 
components being sent further down the supply chain. This can result in some difficulties for 
Customer H, however they are not seen as critical. 
 
Delivery forecasts sent to the suppliers are created by the ERP system by breaking down the 
delivery schedule received from OEM 1 into component level. Lead times are added to the 
forecasts and sent to the supplier. Creation of forecasts are only based on the order info received 
from OEM 1. The assembly planning manager mentioned that the accuracy of the forecasts are 
not measured by Customer H.  
 

5.3.3 Interviews at OEM 2 

Interviews at the aftermarket organisation (Customer E) 

Material planner 

The planner mentioned that the issue of poor stability in prognoses and delivery schedules had 
been raised by the Supplier for some time, and Customer E themselves had also realised that 
their lack of stability had caused problems for their suppliers. As a solution to this the freeze 
time for orders was prolonged from 1 week to 2 weeks. This significantly improved the quality 



32 

of short term delivery schedules. The material planner perceived that the effects on their own 
organisation as a result of this change had been minimal. The change was possible to implement 
overnight and the material planners have not seen any direct effects on their daily work as a 
result. 
 
Apart from the delivery schedules, the material planner maintains regular contact with the 
Supplier. In case of a variation in demand or other disturbances in the flow, the material planner 
and the main planner for the flow to Customer E at the Supplier worked together to come up 
with a solution to handle the situation.  

Forecasts 

The forecast planner mentioned that forecasts used at Customer E are created from consolidated 
data from the 50 local offices. Each local office calculates a sales forecast based on the historic 
sales from the specific local office. All forecasts are aggregated at the central warehouse to 
calculate the total amount required of each component. The warehouse in Gothenburg is the 
central warehouse, responsible for resupplying local warehouses around the world. 
  
Customer E uses two different types of orders in their organisation to manage the inventory at 
their warehouses. The first type is procurement orders. Each of the largest warehouses is 
responsible for supplying some components to every local warehouse around the world. 
Allocation is based on the sourcing strategy, single- or multiple sourcing, and the geographical 
location of the supplier. The forecast planner states that locally based procurement is needed to 
be able to maintain a good collaboration with the suppliers. The aftermarket organisation 
usually purchase components from the same suppliers as the OEM factory does. 
  
The other type of orders are refill-orders. They are used to replenish the inventory of 
components at the central warehouses and local offices. The team that are responsible for the 
refill-orders are located in Gothenburg and manage the material flow between the central 
warehouses and the local offices. Each local office uses a standardised way to calculate a 
forecast based on their own sales data the last 12 months, weighted on the last couple of months. 
Each distribution centres uses re-order point to plan future replenishment. By considering safety 
stock, lead time and EOQ, it is possible to identify when specific materials are required. The 
replenishment team access the forecast data through the organisation’s global ERP-system. The 
needs from the distribution centres are consolidated at the central warehouse and added to the 
delivery schedule sent to the supplier. If the required amount of material are changed at any 
local office, the delivery schedule are instantly changed. 
  
The forecast planner mentioned sales-campaigns, seasonality and trends as external differences 
that affect the process when forecasts are created. If a sales-campaign takes place, the purchaser 
override the forecast data from the distribution centres and creates a forecast based on their own 
experience.  
 
There are no explicit measures taken to ensure proactivity when forecasts are made. Normally, 
forecasts are made based on weighted sales figures from previous months and years. The 



33 

interviewee expressed a desire to develop their forecasting to incorporate measurements such 
as the number of cars in use, service intervals and average lifetime of each part to better foresee 
future changes in demand. Currently their approach is perceived as reactive, managing 
variations in demand once they appear rather than preparing in advance. 

Interviews at production plant (Customer D) 

Material planner 

Compared to the aftermarket organisation, the production plant has significantly shorter freeze 
times for orders, with 24 hours guaranteed in the contracts, but 48 hours generally being given. 
This is to increase flexibility for the OEM, allowing them to quickly reacting to any change in 
demand by adjusting their production. Internally, the production sequence is only frozen for a 
few hours. The delivery schedules sent to suppliers is based entirely on actual customer orders 
for the coming 2 months, approximately. The full delivery schedule encompasses 250 
production days, 50 weeks, with large parts being based on prognosis. The prognoses are 
updated monthly, and are built based on a mix of historical data, along with adjustment for 
current, and probable future market trends. Final orders for all materials and components are 
always based on actual production requirements for each day, rounded up to nearest unit load. 
They do not use EOQ (Economic Order Quantity) to dimension their orders. Customer D keeps 
some safety stocks of materials and components, with volumes being based on experience and 
the nature of the material or component. For example, fragile components that could break 
during transport or handling are generally kept in higher volumes. 
 
Unlike the aftermarket organisation, production does not send any complementary information 
to their suppliers. Their stance is that, ideally, the supply chain should not feature any 
information flow outside of information flow. This is naturally not the case, but it is up to the 
suppliers to contact the materials planning department if they want to raise any issues. There is 
no proactive communication from Customer D, operating with a no news is good news mindset. 
In general, Customer D sticks closely to the terms agreed to in the contract expecting both 
parties to fulfil their commitments without interventions by the other party. As long as 
variations in volumes are within what was stipulated when the contract was drawn, they do not 
send any information apart from the delivery schedules. If the prognosis indicates that they will 
go above the agreed capacity in the contract they work with the affected suppliers to draw a 
new agreement to ensure fulfilment of their needs.  

Volume and Program Manager 

The sales organisation of Customer D is divided into several geographical regions. The division 
of geographical regions are based on each markets expected sales. In high-volume markets each 
country represents one region and in low-volume markets each region is based on typical 
characteristics, such as technical preferences. The local offices in each region acquire 
production slots, based on their forecast of potential sales. If they are unable to fill their assigned 
production slots, it is possible to collaborate with other local offices within a region and 
exchange production slots with one another.  
 



34 

Each specific local office creates a forecast of their potential sales based on historical sales 
together with market trend analysis and economic situation in their sales region. To ensure that 
fluctuations and trends in the market can be fulfilled, adjustments of the forecasts are made 
continuously and the forecasts are updated twice per month. The forecasts are consolidated in 
each market and communicated to the Volume and Program Manager. He creates an overall 
weekly production plan of which quantities of which product model each factory needs to 
produce. 
  
The overall production plan does not manage the complexity of product structures and 
customisation. Instead, the total volume of the overall production plan is automatically broken 
down into separate needs of specific components. Depending on unit loads, inventories and lead 
times, the required amount of each component is transferred to delivery schedules sent to the 
Suppliers.  

5.3.4 Interviews at the Supplier 
To further explore effects resulting from variations in demand forecasts interviews were 
conducted at the Supplier. 

Interview 1 

The first interview focused on general processes around planning and production, with some 
aspects of demand variation being discussed. 

Production and material planning 

Production schedules are generated by MRP and released to production every day. Data 
regarding order information is updated only once each day to ensure that data is kept in synch 
between the information system, logistics and production. All production plans are aggregated 
each day and the total production volume of each component is calculated and planned. The 
Supplier asks at least 4 days of frozen plans from each customer, but would like to have longer 
freeze times as lead times from suppliers can be long for some components. Currently, 
production lines are currently running at around 80% of maximum capacity, which is lower 
than what it had been previously, where production was overloaded due to backorders. For 
component group H10, some variants share the same initial production steps, only being 
differentiated in the final stages. A few variants are significantly different from all others and 
have their own production steps. The possibility to postpone differentiation for some variants 
helps in managing variations as the total throughput time can be upwards of one week. For 
components that are differentiated later on in production, changes to the product mix can be 
made within the last week. 

Effects of variations 

The interviewee perceives that short-term variations have a larger effect than long term 
variations in their organisation. Variations more than 2 weeks ahead can generally be dealt with 
through replanning of production and material orders to subcontractors, or by running extra 
shifts to build a buffer ahead of time. For variations communicated shortly before delivery, 



35 

replanning might not be enough, or too slow, and variations need to be mitigated by safety 
stocks. 
 
Most customers have their plans frozen 4 days ahead of delivery, which can result in late 
changes to production schedules and product mix. Furthermore, the subcontractors ask for 
longer freeze times. Therefore, when a late change to a customer order arrives, material flow 
from the subcontractor is already frozen. Due to this the Supplier generally orders more than 
the believe necessary to ensure that customer order fulfilment. As a result, the Supplier has high 
inventory levels resulting in additional cost. Additionally, most of the contracts with the 
subcontractors are held by the OEM, resulting in arrangements sometimes not fitting the 
Supplier. Most prominent is the fact that the Supplier has very limited flexibility and power in 
these relations on their own, which can leave the Supplier powerless when it comes to creating 
solutions to manage short term demand variations. 
 
To guard against variations in customer orders and disruptions in production the Supplier aims 
to finish production of customer orders ahead of delivery, keeping finished components in 
storage until delivery. For high-volume components, orders are ready around 3 days ahead of 
delivery for some high-volume components. For lower volume components safety stocks are 
kept instead as demand is irregular and production batches are larger than order volumes. In 
general, the Supplier keeps large inventories of materials as well as finished components to 
guard against variations. At the time of the study there had recently been a week where 
Customer E had ordered roughly twice their average weekly volume, which had to be 
“absorbed” by their inventory. Furthermore, the Supplier perceives that it is necessary to keep 
large inventories to ensure that they can supply customers even if the product mix of the demand 
changes. 
 
Forecasts are perceived to be roughly equal in quality between most customers and most 
customers also send complementary information regarding likely short-term upswings in 
volume due to, for example, promotional periods or new product launches. This is perceived as 
very helpful in preparing and planning for these spikes in demand.  
 
As production is at 80% of max capacity currently, and has been even higher previously, the 
Supplier is sensitive to significant upswings in demand. As a result of this, and the previously 
mentioned long lead times from subcontractors, the Supplier keeps a significant inventory of 
materials and finished components. Inventory levels vary during the year, but are generally 
above 10% of total assets in terms of value. Most of the value bound in inventory is in the form 
of finished components. 

Interview 2 

The second interview focused more on variations and their effects on the Supplier, putting extra 
emphasis on the changes in forecasting accuracy from Customer E. The interview was 
conducted with a production planner for component group D10. 



36 

Effects of changes made by Customer E 

The change made by Customer E to prolong the freeze period from 1 to 2 weeks ahead of 
delivery has made variations much smaller and the flow more easily managed during the 
autumn and winter. So far, the change has made day-to-day planning easier and has allowed for 
better stability in production and inventory. There are however still some concerns regarding 
forecasts from Customer E. As Customer E is a global aftermarket organisation, with over 50 
local offices around the world, each with their own demand, the Supplier often receives 
information about significant spikes in demand quite late. If a local office decides on a 
promotional period for a certain component this is only communicated to headquarters in the 
form of requests for material. Because of this, from the Suppliers perspective these demand 
spikes often appear out of nowhere with little to no explanation as to what is happening and the 
long term effects on volume. By comparison, Customer D, one of the production plants of the 
same OEM as Customer E, was able to provide better information on future increases in 
demand. This was perceived to be due to the nature of end customer demand, with Customer D 
producing to fulfil a relatively known demand, and Customer E keeping replacement parts. 

Managing variations 

As the supplier closes production for 4 weeks during the summer due to vacations, maintenance 
and renovations they need to have built up a buffer of finished components ahead of time. This 
buffer is built up during the spring, and results in a heavy strain on production, making them 
susceptible to upswings in demand. While they will have components in their inventory to meet 
the immediate demand, building a sufficient buffer for when they close production could prove 
difficult and expensive. The Supplier may have to order overtime production or forego weekly 
maintenance routines to meet production requirements. In addition to this general sensitivity to 
upswings in demand, the Supplier is particularly sensitive to variations the last 3 weeks before 
they close production. Materials are purchased from foundries in Germany and transported by 
train, with materials taking around 2 weeks to reach the Supplier. Changes during the last three 
weeks before closing production cannot be met by ordering more material by train, but would 
have to be sent by express delivery, which creates additional costs for the Supplier. Some 
variations might be too late to account for at all before production closes, which could create a 
risk of inventory levels not being able to cover the 4 weeks of shut down production. 
 
An additional aspect of the train transport is that batch volumes are either a half or a full cart. 
For certain low-volume components this could be equivalent of months of production. 
Cancelled orders of low volume components can therefore result in the Supplier having 
significant volumes of material in their inventory without any real demand in the near future. 
Additionally, for component group D10, the smallest production batch is around 2000 units, 
whereas the unit load to customer is just over 100. This results in large inventories of low 
volume components, which make up a significant portion of inventory costs but not of revenues. 

Interview 3 

To identify any differences in the effects on production for component groups D10 and H10, 
an interview with a production planner of component group H10 was conducted. 



37 

 
Similarly to production planning of component group D10, the production of component group 
H10 is generally not impacted significantly by variations in customer demand and changes to 
customer orders. In almost all cases, there are enough finished components in the inventory to 
cover for any changes to customer orders. The extra components are taken from the inventory, 
and that volume is then added to the next production batch of that specific product. While this 
does affect the production schedule slightly as batch sizes may increase soon before production 
start, it does not affect downtime and therefore has no significant effect on production costs. 
Customer demand volumes are perceived as generally being stable over time, however orders 
are sometimes moved forward or backwards in time, resulting in variations. This does not result 
in any significant issues for production planning, as inventories are able to absorb the variations 
allowing production to be stable. 

Interview 4 

To build a better understanding of how variations in customer demand impacts the Supplier 
outside of production, an interview was conducted in the purchasing department. 
 
The main materials for each component are sourced from smelteries, both from within Sweden 
and from the rest of Europe. Currently, smelteries in Germany account for the largest share of 
material. The smelteries have different geographical locations, resulting in different lead times. 
However, lead times from the smelteries delivering the largest quantity are roughly one week. 
The supplier has not experienced any issues with the smelteries in terms of delayed deliveries 
or contract disputes. As the Supplier is relatively small in terms of revenue compared to both 
their main customers and the smelteries, there could be a risk of having no flexibility or leverage 
in either direction. The interviewee believes that the fact that they are a wholly owned 
subsidiary of a major OEM has helped them in being given better service and leverage than 
what their share of the revenue of the smelteries would otherwise warrant. The interviewee has 
not experienced that they need to adapt in any way to manage the power relationship with the 
smelteries. 
 
The supplier has not had to order any significant number of express transports from the 
smelteries as a result of material shortage in their own production. As a whole, variations in 
customer demand is not perceived as impacting the order process in such a way where additional 
costs are accrued due to late changes. 

  



38 

6. Analysis & discussion 
The discussion will feature three different segments, each focussed mainly on one research 
question. 6.1 will cover research question 1, 6.2 covers research question 2, and 6.3 covers 
research question 3. 

6.1 Analysis of patterns in FAI and WTS 

6.1.1 General patterns in FAI 
As described in the empirical results, customers of group H10 had significantly better FAI 
scores for short planning horizons compared to customers of component groups C10 and D10. 
A potential explanation for this is that the customers are active in different industries. The OEM 
studied within the heavy vehicle industry has their production frozen for a 2 week period, 
meaning that no changes to production planning will be done within this period. However, this 
information s