Heliyon 8 (2022) e10044

Contents lists available at ScienceDirect

Heliyon
journal homepage: www.cell.com/heliyon

Research article

Critical success factor based resource allocation in ERP implementation:

A nonlinear programming model
Ying Xie a,∗ , Colin Allen b , Mahmood Ali c
a
Faculty of Business and Law, Anglia Ruskin University, Bishop Hall Lane, Chelmsford, CM1 1SQ, UK
b
Faculty of Business, University of Greenwich, UK
c
University of Business and Technology, Jeddah 21361, Saudi Arabia

A R T I C L E I N F O A B S T R A C T

Keywords: This research examines how a Constrained Nonlinear programming model for ERP implementation (CNL_ERP)
ERP implementation can facilitate Small and medium sized enterprises (SMEs) to deploy resources to address the Critical Success
SMEs Factors (CSFs) in the pre-implementation phase, and to invest in them during implementation to increase the
Critical success factor
probability that the implementation will be successful. Applications of CNL_ERP in three case studies demonstrate
Resource allocation
that the average ERP implementation outcomes outperform the observed results. Using the Generalised Reduced
Nonlinear programming
Gradient Method, we developed an ERP implementation strategy realising resource allocation to CSFs. The
strategy provides a rich picture of where to concentrate effort in the initial, intermediate and final phases,
and is very helpful in enabling an SME to understand the progress of an ERP project and the resources needed. In
case there are changes in resources (such as budget, team performance), the model enables SMEs to rank CSFs,
and to adjust resources allocations accordingly to achieve the best ERP implementation performance.

1. Introduction depend heavily on the services and support offered by a Cloud vendor
to organise data and update software (Hashem et al., 2015). However,
An Enterprise Resource Planning (ERP) implementation project in- a Cloud based ERP solution does not remove the key requirements for a
cludes three phases (Hasibuan and Dantes, 2012): (1) pre-implemen- successful ERP implementation. The CSFs that can help an SME achieve
tation, (2) implementation, (3) post implementation. Preparing for the a successful Cloud based ERP implementation (Ahmad and Mehmood,
project in the pre-implementation phase is crucial to ensure successful 2016) are similar to those identified for on-premises ERP implementa-
implementation of ERP (Sun et al., 2016; Jagoda and Samaranayake, tion. Past research has attempted to examine the associations of CSFs
2017). In order to implement a successful ERP project an organisa- with ERP project success using quantitative or qualitative methods, such
tion will have to acquire adequate levels of employees’ skills, vendor as multicriteria decision making, interviews or surveys. However, em-
support and resources in the pre-implementation stage (Ahmadi et al., pirically testing the effectiveness of CSFs on ERP project success will
2015), and to deploy resources optimally to address critical success fac- contribute significantly to the existing body of knowledge (Kirmizi and
tors (CSFs) in the implementation stage (Ahmadzadeh et al., 2021). Kocaoglu, 2022).
ERP implementations in SMEs are especially vulnerable due to three In addition to understanding the effectiveness of CSFs on ERP project
challenges: 1) an ERP project is complex and large scale; 2) the sched- success, the ERP project manager needs to know which CSFs should
ule of the project is usually tight due to competitive pressures; 3) SMEs be prioritised at each stage and how to allocate limited resources to
have limited resources to devote to the implementation, and these re- address them (Sun et al., 2015; Kirmizi and Kocaoglu, 2022). The lack of
sources have limited or zero prior knowledge or experience relating to quantitative measurement of CSFs’ performance and their contributions
an ERP system. The emergence of Cloud based ERP systems has enabled to overall ERP implementation performance has led to a fragmented and
SMEs to experience the advantages of an ERP package while decreasing partial understanding of how to address the selected CSFs in order to
the upfront costs of computing infrastructure and required IT support achieve a successful implementation of, and performance improvement
(Fosso-Wamba et al., 2015). To implement a Cloud based ERP, SMEs from, ERP projects.

* Corresponding author.
E-mail address: [email protected] (Y. Xie).

https://doi.org/10.1016/j.heliyon.2022.e10044
Received 27 March 2021; Received in revised form 31 July 2021; Accepted 18 July 2022

2405-8440/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

SMEs are recommended to acquire resources to maximise the de- support, project management, change management, business process
gree of readiness prior to implementation (Ahmadi et al., 2015), and to reengineering (Vargas and Comuzzi, 2020; Malik and Khan, 2020), user
invest in them during implementation so as to improve the chances of training and education, clear objective setting, and interdepartmental
successful implementation (Saade and Nijher, 2016). Resource alloca- communication (Tarhini et al., 2015). The importance of, and associ-
tion has been widely discussed in other sectors, such as in communica- ations between, these CSF factors were further verified by Baykasoğlu
tions (Zhou et al., 2016) and edge computing (Liao et al., 2020), where and Gölcük (2017) using interpretive structural modelling and fuzzy
quantitative algorithms and frameworks were developed to allocate cognitive maps.
power supplies and computational tasks. However, literature pertain- The other significant development relates to attempting to associate
ing to resource allocation in ERP implementation is scarce, with a focus CSFs with the implementation stages. Drawing on real world case stud-
technical and social processes (Phillips and Bana e Costa, 2007) or key ies, Saade and Nijher (2016) consolidated a list of CSFs during the ERP
resources required (Chatti et al., 2021). While some sources identify re- implementation process and related these to the five ERP implementa-
source allocation as an important ERP adoption factor (Verdouw et al., tion stages. To manage the performance of CSFs in each ERP stage, Sun
2015), very few ERP studies determine or estimate quantitatively the et al. (2015) proposed performance assessment models and developed
resources required to achieve target ERP performance. KPIs for CSFs. The model developed by Sun et al. (2015) has the func-
This research attempts to fill the research gaps, by a) associating tionality to quantitatively measure ERP project performance at each
CSFs with ERP implementation performance using quantitative meth- stage and to identify remedial actions if the performance falls below ex-
ods, b) quantifying resource allocation to address CSFs in ERP imple- pectation; therefore, it could serve as a tool to decide where and when
mentation, and c) empirically testing and validating the effectiveness during the ERP lifecycle a CSF should be applied.
of investment in CSFs on ERP project success. As only a limited num- While there has been plenty of research exploring the CSFs for ERP
ber of SMEs have adequate resources to adequately address all the CSFs implementation, and associating CSFs with ERP implantation stages,
(Sun et al., 2015), this research aims to answer two research questions: research into how to address, resource or administer CSFs during ERP
(I) How can SMEs achieve target ERP implementation performance by implementation has been limited. By combining an empirical survey
appropriately allocating resources, i.e., time and budget, to address and mathematical modelling, we aim to provide both empirical and
CSFs? (II) In response to changes in resources available for ERP imple- scientific evidence of the direct influences of the chosen CSFs on ERP
mentation (such as budget, team performance), how can SMEs adjust implementation performance. Resource allocation is a key element of
resources allocated to address CSFs so as to optimise ERP implementa- the ERP implementation strategy (Parr and Shanks, 2000), and is the
tion performance? This study is based on a combination of analytical focus of this research. Since this research focuses on ERP implemen-
modelling and empirical case studies, demonstrating the practical ap- tation after ERP software selection and project planning, we do not
plication of CNL_ERP. consider factors in relation to ERP software selection, organisational
The paper is organised as follows. The relevant literature is re- environment, organisational experience or change management. We
viewed in Section 2. In Section 3, through a combination of modelling have chosen to examine how to address five CSFs in an ERP imple-
and empirical surveys, CNL_ERP is developed, combining both analyti- mentation project, including Top Management Support (TM), Users, IT
cal regression models and constrained nonlinear programming models. Infrastructure (IT), Project Management (PM), and Vendor Support (VS)
Section 4 shows the application of CNL_ERP via case studies. Finally, (see Table 1). Despite the terminological differences around CSF names
theoretical and managerial contributions, and directions for further re- that exist in the literature, the attributes considered under these five
search are presented in Section 5. CSFs represent a comprehensive list of factors that are identified as be-
ing directly associated with ERP implementation stages and having an
2. Literature review important influence on the success of ERP project delivery (Sun et al.,
2005). We chose these five CSFs for the following reasons.
The implementation stage of ERP has been widely studied, includ-
ing CSF identification, strategies and approaches for implementation, ∙ The five chosen CSFs are oriented towards the implementation
knowledge transfer and organisational ERP fit. While it is often argued stage (Vargas and Comuzzi, 2020; Gupta et al., 2018); therefore,
that the implementation of ERP is a continuous cycle of improvement, they fit well with the aim of this study.
the parameters of this investigation are limited to ERP implementa- ∙ These five CSFs have been consistently categorised and highlighted
tion post ERP software selection and project planning. This research as important factors for successful implementation of ERP projects
develops a tool to forecast the resources, i.e. project schedule and bud- (see references in Table 1).
get allocation (Nagpal et al., 2015), required for implementing ERP ∙ These CSFs have not previously been examined with regard to their
from the initial training until the desired ERP implementation perfor- association with key measures relating to project deliverables and
mance level is achieved. In this section, we review the CSFs of ERP constraints, including cost, time and contribution to the overall
implementation and various ERP performance measures, as well as the ERP implementation performance.
quantitative models developed for ERP implementation.

2.1. Critical success factors 2.2. ERP implementation performance

The discussion of CSFs is a predominant research stream in ERP lit- ERP implementation performance can be defined by multiple as-
erature, even in the era of Cloud computing, which has resulted in a pects, depending on when the performance is measured and who mea-
shift of ERP systems to Cloud platforms (Fosso-Wamba et al., 2015). sures it. The project manager’s key objective is to deliver the project
Cloud-based solutions remove the requirements to install IT hardware on time and within budget while, at the organisational and user lev-
on premises and to maintain an IT workforce in organisations, making els, the aim is to reap the projected operational benefits of the ERP
ERP implementation more affordable for SMEs. Recently, more research system (Kirmizi and Kocaoglu, 2022). The performance of implement-
has been carried out to understand the impact of CSFs on ERP im- ing ERP as an information system is usually measured at the end of
plementation performance, for both Cloud-based and on-premises ERP the go live stage, and based upon project delivery outcomes (Ram et
solutions (Alharthi et al., 2017; Ahn and Ahn, 2020; Gupta and Misra, al., 2013). Such performance is defined by multiple parameters, such as
2016). Traditional CSFs relating to both organisational and technical time, cost and functionality, to assess whether the expected objectives
aspects prove to significantly impact the successful implementation are being achieved through the implementation within the limitations
of Cloud-based ERP (Gupta et al., 2018), including top management (Lima et al., 2013; Sun et al., 2016). The main functionality expected

2
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

Table 1. Five Critical Success Factors considered in this research.

CSFs identified for ERP implementation CSF Attributes References that have identified the CSF

CSF 1 - Top Management support (TM) Leadership, participation and commitment of the senior level of management Tarhini et al., 2015; Malik and Khan, 2020

CSF 2 - Users Users’ perception, interest, commitment, participation, feedback, IT skills, Plaza et al., 2010; Baykasoğlu and Gölcük, 2017
team dynamics, in house training, familiarity with other team members and
external consultants

CSF 3 - IT infrastructure (IT) Hardware, software and IT architecture, databases of appropriate quality and Gupta et al., 2018
data migration capability

CSF 4 - Project Management (PM) Project team selection, team training, team competence, project tracking, Baykasoğlu and Gölcük, 2017; Gupta et al.,
business process reengineering 2018; Vargas and Comuzzi, 2020

CSF 5 - Vendor Support (VS) Vendor expertise relating to training, technical knowledge and support, Malik and Khan, 2020
maintenance, emergency management, updates, service responsiveness and
reliability

from an ERP implementation is to enhance integration of commercial as evaluating and setting critical success strategies for ERP projects (Yeh
processes in an organisation and, consequently, to deliver the projected and Xu, 2013). For example, Plaza and Rohlf (2008) investigated how
operational benefits, including improved efficiency, reduced production the training, learning and performance of the project team can min-
costs and maximised profits (Kirmizi and Kocaoglu, 2022). The percent- imise ERP project consultancy costs, and developed an analytical model
age of such functionality achieved by the ERP implementation is also to predict the project completion date. Based on their 2008 work, Plaza
an important evaluation criterion. et al. (2010) presented a comparative analysis of two types of learning
ERP implementation performance has been researched as a quali- curves and illustrated how they can be applied in four ERP implementa-
tative measurement or as a quantitative measurement, and measured tion projects. Although Plaza and Rohlf’s work enhanced the tradition-
subjectively, by either experts or users using measurement scales. For ally qualitative ERP research by developing quantitative models, their
example, Zareravasan and Mansouri (2016) quantitatively measured work has certain limitations: 1) only one CSF, project team progress,
the outcome of ERP projects using budget, time, and user expecta- is addressed; 2) the analytical models developed are only tested in the
tions. Sun et al. (2015) measured performance at each stage using CSF context of the case study organisations and not validated by statisti-
weighted KPI scores and identified remedial actions if the performance cal analysis; 3) analytical models cannot provide dynamic views on the
was below expectation. Plaza and Rohlf (2008) and Plaza et al. (2010) ERP implementation project processes.
defined performance as the rate of competition of a task (the number Complementing, and in contrast to, research that utilises deduc-
of modules configured, or the number of transactions completed), and tive research approaches to develop quantitative modelling, Sun et
modelled the performance as a mathematical function that is dependent al. (2005) used realistic data to quantify the measurements to be ad-
on time, training and learning. Sun et al. (2005) developed a quantita- dressed for CSFs during ERP implementation, including cost, schedule,
tive evaluation of overall ERP implementation performance in terms of and goal achievement. They also developed a simulation model to help
utilisation of the ERP system’s capabilities, and the organisation’s func- SMEs develop appropriate measurements to measure ERP implementa-
tionality requirement that is met by the ERP system. tion achievement. However, their model lacks key functionality relating
In this study, ERP implementation performance is defined from the to predicting the resources needed by SMEs and cannot help SMEs to
system implementation point of view and evaluated as to whether the plan resources in advance. As a result of restricted resource availability,
project is completed within time and budget limitations, and whether Sun et al. (2005)’s simulation model was limited with regard to both
the adopted ERP system helps the SME achieve the required level of validity and generality as: 1) only 6 SMEs were observed; 2) other data
effectiveness (Bhatt et al., 2021). A quantitative evaluation of the per- were generated using a data fitting method; and 3) it was assumed that
formance level is defined as the percentage of the organisation’s target variations between observed data and generated data were insignifi-
functional requirements met by the ERP implementation. The overall cant.
success of ERP implementation relies on the degree to which CSFs are Xie et al. (2014) developed an integrated decision support system
addressed during implementation; therefore, the evaluation of overall (DSS) for ERP implementation in SMEs, combining logistic regression
ERP implementation performance level, cost and time is broken down models, linear regression models, a nonlinear programming model, and
by CSFs. a simulation model, to predict ERP project implementation outcomes
and facilitate the allocation of resources. However, the validity of the
2.3. Quantitative models on ERP implementation DSS needs to be further tested in empirical studies to ascertain its prac-
tical use and benefits. The robustness of the model also needs to be
Many researchers investigated the direct association between CSFs tested and analysed.
and ERP implementation performance (Gupta et al., 2018; Malik and Zareravasan and Mansouri (2016) proposed a fuzzy cognitive map
Khan, 2020). Traditionally, qualitative research has been the most based dynamic model of ERP failure factors through project lifecycle
prevalent research method in studying such relationships. While the phases. Imitating human reasoning, this tool models uncertainty and
identified CSFs in case studies or surveys enable SMEs to develop a bet- related events and could be used to assess the joint influence of ERP im-
ter understanding of the CSFs’ impacts, the impacts extent is unclear, plementation failure factors on project outcomes. However, this tool is
limiting the ability of SMEs to make effective ERP implementations limited to the Iranian context and generalisation would be difficult. The
interventions based on the research. A scientific model is needed to model also strongly depends on experts’ subjective judgement on the in-
suggest how, when and which CSFs should be addressed during ERP im- terrelationships between factors; hence, both its use and its outcomes
plementation so that organisations can plan and execute ERP projects could vary significantly with different groups of experts.
that result in a more successful implementation (Vargas and Comuzzi, The above review indicates that mathematical programming and
2020). simulation have been valuable in providing insights into specific prob-
The use of operational research (OR) approaches to research ERP lems, facilitating organisational preparation for ERP implementation,
implementation has also received more attention over the last decade. and achieving success in ERP implementation; however, a mathemati-
OR models have the ability to evaluate existing concepts of ERP as well cal model that proves to be beneficial to firms in identifying required

3
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

Fig. 1. A logistic curve for an ERP implementation project.

resources and in developing an implementation strategy realising re- the vision behind the ERP implementation, the basis of the ERP system,
source allocation, has not been available. The current research aims to the required changes, and the predefined implementation strategy.
fill this gap in the literature and takes the view that the decision to al-
locate resources (time and budget) to address CSFs should be made by Assumption 2. Organisations have been given intensive training and
considering resource constraints with the aim of maximising ERP im- education to upgrade skills relevant to the ERP software and there
plementation performance. is support from the vendor. Due to this training, teamwork and col-
laboration have been achieved and organisations are ready for ERP
3. Mathematical models implementation; therefore, we can ignore the start-up effect in the ERP
project implementation.
Mathematical models were constructed to show the relationships be-
tween implementation cost and project duration, as well as the ERP Assumption 3. The dynamic ERP implementation environment can be
implementation performance level over the project duration. Three im- quantified and modelled using mathematical models at CSF level.
portant parameters were introduced in the models:
Assumption 4. The internal costs of ERP project implementation, in-
∙ Implementation cost: denoted by 𝐶, the cumulative cost of the cluding the one-off purchase of ERP software and hardware, overhead
overall ERP implementation project. costs, and system installation costs are not considered in the mathemat-
∙ Project duration: denoted by 𝑇 , the time elapsed from the start of ical model. The model only considers dynamic cost, which changes with
the initial training phase to the final go live phase, covering the the time spent to address each CSF.
configuration, testing, and conversion phases (see Fig. 1).
∙ Performance level: denoted by PF, the percentage of the organisa- 3.2. Modelling performance against time using the logistic curve
tion’s expected functional requirements met by the ERP implemen-
tation. The S-Curve is the one most commonly used project management
tools for cost estimation and productivity assessment (Konior and Szós-
The following notations are used to model parameters and variables: tak, 2020). The nature of an ERP implementation project results in the
progress growing rapidly during the training and configuration phases,
𝑀 total number of CSFs considered, more slowly in the testing and conversion phases and reaching an
𝑖 subscript of a CSF, 𝑖 = 1, 2, … , 𝑀 , asymptotic maximum when the project goes live (Fig. 1). The progress
𝑑𝑖 coefficient of the cost function, curve presented in Fig. 1 is a logistic curve, being similar to the S-Curve
𝑡𝑖 time spent on the CSF 𝑖 , but ignoring the start-up effect in the project planning stage. The logis-
cost𝑖 cost consumed by CSF 𝑖 , tic model is noted for its robustness and is frequently used to predict and
PF 𝑖 progress made by CSF 𝑖 , model the performance of an ERP project team (Plaza and Rohlf, 2008;
𝑝𝑖 performance threshold of CSF 𝑖 , Plaza et al., 2010), to measure project complexity (Dao et al., 2020),
𝑘𝑖 progressing curve coefficient of CSF 𝑖 , measuring progressing and to analyse productivity changes and financial implications of the
speed of CSF 𝑖 , introduction of new technology (Dardan et al., 2006). Plaza and Rohlf
𝑇 project duration of an ERP implementation, (2008) demonstrated that, for a project team working on a CSF, progress
follows a logistic curve. This is also the case for teams working on other
PF performance level achieved in an ERP implementation,
CSFs, vendor support, end users, IT infrastructure and top management
𝐶 cost consumed by an ERP implementation,
(Sun et al., 2005). Inside an organisation, teams addressing such CSFs
TL limitation on ERP project duration,
usually lack experience with, and knowledge of, the systems that they
CL limitation on ERP implementation cost (budget),
are implementing. Outwith the organisation, the ERP vendors’ team are
𝑔1 ERP project duration, imposed as constraint 1 on CNL_ERP,
usually relatively ignorant about, and lack experience of, the client. For
𝑔2 ERP implementation cost, imposed as constraint 2 on CNL_ERP
these reasons, the early impact on the ERP implementation performance
PI proximity index.
of a particular CSF is low, improving with time to a peak threshold of
performance at some point during the project.
3.1. Assumptions
We use the logistic curve in Fig. 1 to model the relationship be-
tween practice and performance. A critical predictor to measure the
In order to model a near to reality ERP implementation in relatively success of ERP implementation is claimed to be the system usage i.e.
simplified mathematical models, and to focus on the resource allocation system’s ability in performing tasks (Nwankpa and Roumani, 2014);
for ERP implementation, we introduce several assumptions: it is claimed that the higher system usage by the end-user increases
the chances of organisation’s achieving ERP implementation objectives
Assumption 1. Teams are fully prepared for ERP implementation and (Sun et al., 2005; Nwankpa, 2015). Xie et al. (2014) defined the per-
have received the necessary briefings and consultation to understand formance level as the percentage of the organisation’s target functional

4
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

𝑀
∑
requirement met by the ERP implementation. In this research, practice
s.t. 𝑔1 (𝑡1 , 𝑡2 , … , 𝑡𝑀 ) = 𝑡𝑖 ≤ 𝑇 𝐿
is modelled as project duration 𝑇 , while the overall ERP implementa- 𝑖=1
tion performance level PF is measured as the estimated percentage of 𝑀
∑
the company’s functional requirements that are met by the ERP imple- 𝑔2 (𝑡1 , 𝑡2 , … , 𝑡𝑀 ) = cost𝑖 (𝑡𝑖 ) ≤ CL
mentation. For example, the ERP system project manager estimated that 𝑖=1
only 50% of the company’s functional requirements are met by the ERP 𝑡𝑖 ≥ 0 for all 𝑖 = 1, 2, … 𝑀
implementation, which means PF = 50.
When addressing individual CSFs, practice is represented by time Define vector 𝐓 = (𝑡1 , 𝑡2 , … , 𝑡𝑀 )T and formula (5) can be rewritten in
spent on addressing CSF 𝑖 , denoted by 𝑡𝑖 , and performance means the matrix notation:
progress achieved by CSF 𝑖 , denoted by PF 𝑖 and measured as that CS- 𝑀 𝑀
∑ ∑
F’s percentage contribution to the overall ERP implementation perfor- Max PF(𝐓) = PF 𝑖 (𝑡𝑖 ) = 𝑝𝑖 ⋅ (1 − 𝑒−𝑘𝑖 𝑡𝑖 ) (5.a)
mance level. A Progress vs Time logistic curve is used to express the 𝑖=1 𝑖=1
relationship between the progress made by a project team against time 𝑀
to address a CSF 𝑖 and is formulated in (1): ∑
s.t. 𝑔1 (𝐓) = 𝑡𝑖 ≤ 𝑇 𝐿
( ) 𝑖=1
PF 𝑖 (𝑡𝑖 ) = 𝑝𝑖 ⋅ 1 − 𝑒−𝑘𝑖 𝑡𝑖 (1) ∑𝑀
𝑔2 (𝐓) = cost𝑖 (𝑡𝑖 ) ≤ CL
where 𝑝𝑖 is the performance threshold, or the maximum percentage con- 𝑖=1
tribution a CSF makes to the overall performance level. The progressing 𝐓≥0
coefficient 𝑘𝑖 relates directly to the rate of progress made by a team;
however, since the ERP project team is diverse by nature, and will vary The constrained nonlinear programming model in (5) or (5.a) is not
considerably with the context within which ERP is implemented (Yoon, explicitly solvable for symbolic solutions, but, if parameter values are
2009), 𝑘𝑖 is difficult to calculate accurately. The overall performance provided, can be solved numerically using optimisation tools such as
level of ERP implementation is calculated as: Excel’s Solver, Mathematica and CPlex. The optimisation tools imple-
ment different algorithms depending on which solver is used; we use
𝑀
∑ the Generalised Reduced Gradient (GRG) method that is a generalisa-
PF(𝑡1 , 𝑡2 , … , 𝑡𝑀 ) = PF 𝑖 (𝑡𝑖 ) (2)
tion of the Steepest Ascent (or Steepest Descent) method. GRG converts
𝑖=1
the constrained problem into an unconstrained one by using direct sub-
3.3. Modelling cost against time stitution and uses an iterative procedure to find an improved direction
for the objective function while satisfying the constraint equations; the
The overall ERP implementation performance level relies on the per- improved direction is determined by the reduced gradient. Discussion
formance contributed by the individual CSFs. Administration of any CSF of algorithms for nonlinear programming is beyond the scope of this
will usually require money, management effort and human resources. research, but the GRG method is to be used to analyse the model in
As such, the overall ERP implementation cost is calculated as the to- Section 3.4 and obtain implementation strategy in Section 4.4.
tal cost of improving the CSFs. Based on the authors’ observations and
research on ERP implementations (Sun et al., 2005), the overall ERP 3.5. Analysis of the constrained nonlinear programming model
implementation cost increases with the total time spent. A linear cost
function is therefore constructed for each CSF, showing the required The reduced gradient of objective function PF(𝐓) with respect to 𝐓
money against time spent to address the corresponding factor. is derived as ∇𝑟 𝑃 𝐹 (𝐓) in (6) (see Appendix A):
Adopting a uniform function to model the Cost vs Time relationship ( )
for all CSFs is used to simplify the functional form and focus on optimi- 𝜕PF 𝜕PF 𝜕PF
∇𝑟 PF(𝐓) = , ,…,
sation. Although some costs may be incurred when no time is spent on 𝜕𝑡1 𝜕𝑡2 𝜕𝑡𝑀
addressing CSFs, those costs are so low relative to the costs incurred in ( )
= 𝑝1 𝑘1 𝑒−𝑘1 𝑡1 , 𝑝2 𝑘2 𝑒−𝑘2 𝑡2 , … , 𝑝𝑀 𝑘𝑀 𝑒−𝑘𝑀 𝑡𝑀 (6)
spending time that they can effectively be regarded as zero (Xie et al.,
2014). The constant initial cost when 𝑡𝑖 = 0 is often omitted in the cost The extreme point 𝐓∗ = (𝑡∗1 , 𝑡∗2 , … , 𝑡∗𝑀 )T will be generated in the direc-
functions that model relationships between project schedules and costs tion of the gradient in such a way that the maximum PF(𝐓∗ ) is achieved
of ERP implementation, as evidenced in Sun et al. (2005) and Plaza and and constraints hold. It is obvious that the reduced gradient of ∇𝑟 PF(𝐓)
Rohlf (2008). Therefore, we assume that cost𝑖 (𝑡𝑖 ) = 0 when 𝑡𝑖 = 0, and changes with 𝐓. The gradient in (6) not only varies with the 𝑝𝑖 and 𝑘𝑖 ,
the functional model of the cost to address CSF 𝑖 is: but also with time 𝑡𝑖 allocated to CSF 𝑖 . A CSF with high performance
threshold 𝑝𝑖 , and fast learning or knowledge absorption speed which
Linear model cost𝑖 (𝑡𝑖 ) = 𝑑𝑖 ⋅ 𝑡𝑖 , 𝑡𝑖 > 0 (3) corresponds to a higher progressing coefficient 𝑘𝑖 , contributes to the
ERP implementation performance more quickly during the early part
The implementation cost of ERP is obtained as:
of the project duration. Time allocated to CSF 𝑖 in the next iteration is
𝑀
∑ proportional to the value of 𝑝𝑖 𝑘𝑖 𝑒−𝑘𝑖 𝑡𝑖 . The mathematical algorithm re-
𝐶(𝑡1 , 𝑡2 , … , 𝑡𝑀 ) = cost𝑖 (𝑡𝑖 ) (4) veals that the focus given to CSFs should be adjusted according to the
𝑖=1 amount of resources available, usually involving time and budget, and
the progressing coefficient 𝑘𝑖 .
3.4. Development of constrained nonlinear programming model for ERP
implementation
4. Empirical studies and results
A CNL_ERP model is constructed as a nonlinear programming model
defined by an objective and a set of constraints in (5). The objective A survey was conducted to collect empirical data including ERP
is to maximise the ERP implementation performance level, subject to project costs, schedules and performance levels from a group of SMEs.
limitations on budget and project duration: Statistical regression techniques were used to fit the empirical data to
the analytical models for ERP cost and performance at CSF level, as
𝑀
∑ 𝑀
∑ ( ) shown in equations (1)–(4). Case studies were then conducted to check
Max PF(𝑡1 , 𝑡2 , … , 𝑡𝑀 ) = PF 𝑖 (𝑡𝑖 ) = 𝑝𝑖 ⋅ 1 − 𝑒−𝑘𝑖 𝑡𝑖 (5)
the validity and effectiveness of the analytical models.
𝑖=1 𝑖=1

5
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

Table 2. Sampling criteria used in the survey.

Location UK or North America

Size 50-150 employees

ERP implementation Completion of at least one ERP project

CSFs All the five CSFs are addressed during the ERP implementation

Table 3. Coefficients and 𝑹 𝟐 of Cost and Progress regression models.

Parameters 𝑪𝑺𝑭 𝟏 -TM 𝑪𝑺𝑭 𝟐 -Users 𝑪𝑺𝑭 𝟑 -PM 𝑪𝑺𝑭 𝟒 -IT 𝑪𝑺𝑭 𝟓 -VS Average 𝑹𝟐

Linear Cost model 𝑑𝑖 659.92 656.28 719.66 1361 1770.7 0.75

𝑅2𝑖 0.91 0.61 0.68 0.77 0.79

Logistic Progress model 𝑘𝑖 0.045 0.163 0.040 0.076 0.143 0.77

𝑝𝑖 19.03 17.13 24.26 19.28 12.94

𝑅2𝑖 0.98 0.61 0.83 0.77 0.66

4.1. Survey results plementations and applied CNL_ERP to them. The case study selection
criteria are:
The initial questionnaire was developed based on the five CSFs iden-
tified in Table 1, focusing on the progress, cost and time associated 1) The selected organisations are SMEs who meet our sampling crite-
with each CSF during implementation. The one-off costs of software and ria in Table 2 and who participated in the survey.
hardware are not included in the cost to address IT infrastructure. Based 2) The selected organisations implemented ERP systems on a similar
on the criteria in Table 2, we used Thomson Data, small business asso- schedule, for example within 110 days or 120 days, or implemented
ciation websites, SAP user groups and ERP suppliers’ websites to choose them at similar costs, for example at a cost of $100,000. This is to
400 SMEs and conducted internet-based surveys with them. 60 valid re- enable direct comparisons of resource allocation and resultant ERP
sponses were received, giving a 15% response rate (see Appendix B). performance between the organisations.
Those 60 SMEs cover a wide range of industrial sectors, including Man-
ufacturing (28%), Information Technology (15%), Telecommunications The methodology presented in this section can be used by a project
(14%), Banking and Finance (10%), Utilities (9%), Education (2%), and manager to plan, prepare and deploy resources, and select training
others (23%). This ensures the sample is representative and the regres- and implementation strategies. Two of the authors were participant
sion model is relatively more generalisable. observers on projects 1 and 3 respectively and the research team
conducted post-implementation data collection on all three projects.
4.2. Analytical regression models for the observed data CNL_ERP is initially tested and evaluated using real data to demonstrate
the practical benefits of the model and to identify areas for improve-
Employing formulae (1) and (3), the values of 𝑑𝑖 , 𝑘𝑖 , 𝑝𝑖 and 𝑅2 were ment; testing with multiple case study companies generates more robust
obtained using least square methods for each CSF and presented in Ta- and precise results.
ble 3. 𝑅2 values for both the Linear Cost vs Time curve and the Logistic Observed data from each case are set up as a Baseline model, and
Progress vs Time curve are above 0.70, indicating a good level of fit. the case analysis involves two test scenarios as follows:
For each CSF, the Cost vs Time linear curve and Progress vs Time
exponential curve are formulated in equations (7)–(16): Scenario 1: obtaining the project outcomes by inputting the observed
𝐓 = (𝑡1 , 𝑡2 , … , 𝑡5 )T to the CNL_ERP, and releasing the constraints on
CSF 1 -TM: cost1 (𝑡1 ) = 659.92 ⋅ 𝑡1 (7)
planned project duration and budget;
( )
PF 1 (𝑡1 ) = 19.03 ⋅ 1 − 𝑒−0.045⋅𝑡1 (8) Scenario 2: calculating ideal solution 𝐓∗ = (𝑡∗1 , 𝑡∗2 , … , 𝑡∗5 )T , and associ-
ated project outcomes by maximising the project performance level
CSF 2 -Users: cost2 (𝑡2 ) = 656.28 ⋅ 𝑡2 (9)
whilst satisfying constraints on planned project duration and bud-
( )
PF 2 (𝑡2 ) = 17.13 ⋅ 1 − 𝑒−0.163⋅𝑡2 (10) get.

CSF 3 -PM: cost3 (𝑡3 ) = 719.66 ⋅ 𝑡3 (11)

Scenario 1 is used to test the validity and effectiveness of the
( )
PF 3 (𝑡3 ) = 24.26 ⋅ 1 − 𝑒−0.04⋅𝑡3 (12) CNL_ERP, checking as to whether the outputs from CNL_ERP are similar
to the observed results if the inputs 𝐓(𝑡1 , 𝑡2 , … , 𝑡5 )T , i.e. resources as-
CSF 4 -IT: cost4 (𝑡4 ) = 1361 ⋅ 𝑡4 (13)
signed to each CSF, are the same. Scenario 2 aims to compare resource
( )
PF 4 (𝑡4 ) = 19.28 ⋅ 1 − 𝑒−0.076⋅𝑡4 (14) allocations and associated ERP results from real life situations against
those recommended by CNL_ERP. Applying CNL_ERP to the case stud-
CSF 5 -VS: cost5 (𝑡5 ) = 1770.7 ⋅ 𝑡5 (15)
ies offers an opportunity to conduct preliminary tests and to identify
( )
PF 5 (𝑡5 ) = 12.94 ⋅ 1 − 𝑒−0.143⋅𝑡5 (16) the differences between the resources allocated by SMEs in real life sit-
uations and resources assigned by CNL_ERP, as well as the differences
4.3. Case studies between the resulting project outcomes. The case studies also evaluate
the analytical and practical aspects of CNL_ERP as a tool for predicting
4.3.1. Details of ERP implementation projects and allocating resources prior to implementation.
To demonstrate the practical benefits of the CNL_ERP model, and Due to confidentiality agreements and privacy, we refer to the case
evaluate its effectiveness, we collected data from three real ERP im- companies anonymously as Company 1, 2 and 3. Company 1 is a US

6
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

Table 4. Information about case companies.

Case Company 1 Case Company 2 Case Company 3

Participant’s Job Title MIS-Manager SQA-Analyst Net-Developer

Industry IT Leisure Industry Education

Location USA UK UK

No. of employees 118 220 240

No. of internal resources + external consultants 3+5 5+1 4+2

Implementation result Successful Successful Successful

Implementation completed on time? Yes Yes Yes

Completed within budget? Yes Yes Yes

Project duration (Days): 𝑇 120 107 110

Cost of implementation: 𝐶 $100,000 $90,000 $90,000

Project Performance Level PF 65% 70% 63%

Table 5. Comparison of observed results and results generated by CNL_ERP.

Case Test 𝑡1 𝑡2 𝑡3 𝑡4 𝑡5 𝑇 𝐶 𝑃𝐹 𝑃𝐼 Cost difference Performance difference

1 Baseline 30 40 30 10 10 120 $100,000 65% 0.69 0 0

Scenario 1 30 40 30 10 10 120 $100,000 68% 0.69 0 +3%

Scenario 2 33 16 39 17 9 114 $100,000 73% 1 0 +8%

2 Baseline 28 28 21 15 15 107 $90,000 70% 0.71 0 0

Scenario 1 28 28 21 15 15 107 $98,941 69% 0.71 +$8,941 −1%

Scenario 2 30 15 35 15 8 103 $90,000 70% 1 0 0

3 Baseline 20 40 20 10 20 110 $90,000 63% 0.62 0 0

Scenario 1 20 40 20 10 20 110 $102,866 64% 0.62 +$12,866 +1%

Scenario 2 30 15 35 15 8 103 $90,000 70% 1 0 +7%

based company designing and manufacturing switches, routers and Subject to different constraints in (18) and (19):
other networking equipment for clients in the governmental, corporate
∑
5
and educational sectors. They implemented three modules of SAP ERP Case 1: s.t. 𝑔1 (𝐓) = 𝑡𝑖 ≤ 120
within a planned budget of $100,000, including Financial Accounting, 𝑖=1
Materials Management and Production Planning. Company 2 is based in ∑5 (18)
the UK and is a supplier of booking and membership systems for univer- 𝑔2 (𝐓) = cost𝑖 (𝑡𝑖 ) ≤ 100, 000
𝑖=1
sities, leisure centres, clubs and health and fitness groups. The company
implemented two SAP ERP modules in four months, namely Sales and 𝐓≥0
Financial Accounting. Company 3 is also UK based and concentrates ∑
5

on supplying management software packages to schools, colleges and Case 2 and Case 3: s.t. 𝑔1 (𝐓) = 𝑡𝑖 ≤ 120
𝑖=1
universities; they implemented a heavily customised version of Priority
ERP software, with the core being the Sales and Financial Accounting ∑5 (19)
𝑔2 (𝐓) = cost𝑖 (𝑡𝑖 ) ≤ 90, 000
modules. Detailed information about the companies and their ERP im- 𝑖=1
plementations is shown in Table 4. 𝐓≥0
The following information was collected during interviews with the
participants from Case Companies 1-3, and recorded in the “Baseline” 4.3.2. Discussion of results
row in Table 5: (1) 𝑡𝑖 - time spent on the CSF 𝑖 , (2) 𝑇 - project duration, Scenario 1: As shown in Table 5, the observed 𝑡𝑖 from Scenario 1
(2) 𝐶 - implementation cost, (3) PF – performance level. is used as an input to Equation (17); the Performance level PF
By setting up a goal of maximising the ERP performance level, achieved in Scenario 1 is 68% for Case 1, 69% for Case 2, and
CNL_ERP was applied to make decisions on T = (𝑡1 , 𝑡2 , … , 𝑡5 )T for the 64% for Case 3, similar to the observed PF in the Baseline model,
Case Companies 1-3. The objective function is formulated in equation i.e. 65% for Case 1, 70% for Case 2, and 63% for Case 3. Inputting
(17) for the three case studies. Substituting (8), (10), (12), (14) and (16) observed data to CNL_ERP generates a similar implementation cost
to (17), the objective function becomes: as the observed cost. The results from the case studies verify the
validity and accuracy of the CNL_ERP model, showing that the in-
( ) ( ) tegrated model closely resembles the actual ERP implementation
Max PF(𝐓) = 19.03 ⋅ 1 − 𝑒−0.045⋅𝑡1 + 17.13 ⋅ 1 − 𝑒−0.163⋅𝑡2
performances. The small difference in costs could be explained as
( ) ( ) being due to the constructed analytical Cost vs Time model being
+ 24.26 ⋅ 1 − 𝑒−0.04⋅𝑡3 + 19.28 ⋅ 1 − 𝑒−0.076⋅𝑡4
obtained through the regression fitness approach and, therefore,
( )
+ 12.94 ⋅ 1 − 𝑒−0.143⋅𝑡5 (17) not producing the same results as the observations.

7
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

Scenario 2: Equation (17) is solved by Mathematica using the GRG

procedure. The calculated 𝐓∗ = (𝑡∗1 , 𝑡∗2 , … , 𝑡∗5 )T , implementation cost
𝐶 and ERP performance level PF in Scenario 2 are compared with
the observed results in the Baseline model in Table 5. Given 𝐓 is a
vector, we define a Proximity Index PI to compare the similarity of
the observed 𝐓 = (𝑡1 , 𝑡2 , … , 𝑡5 )T to the calculated 𝐓∗ = (𝑡∗1 , 𝑡∗2 , … , 𝑡∗5 )T .
We adopt the Similarity to Ideal Solution proposed by Hwang and
Yoon (1981) to calculate PI using the Technique of Order Prefer-
ence by Similarity to Ideal Solution (TOPSIS), as shown in (20):
( )
PI = 𝑆 − ∕ 𝑆 + + 𝑆 − (20)
where
√
( )( )T
𝑆+ = 𝐓 − 𝐓∗ 𝐓 − 𝐓∗ (21)
and
√
( )( )T
𝑆− = 𝐓 − 𝐓− 𝐓 − 𝐓− (22)
In equations (21) and (22), 𝐓∗ is the extreme point and used as the Fig. 2. Dimensional elements 𝜕PF
in reduced gradient ∇𝑟 PF.
𝜕𝑡𝑖
positive ideal solution, and 𝐓− = (0, 0, 0, 0, 0) is the negative ideal solu-
tion which minimises the performance level PF to be 0. 𝑆 + and 𝑆 − are
the separations from positive and negative ideal solutions, respectively. 4.4. ERP implementation strategy
The higher the proximity PI, the closer the observed 𝐓 = (𝑡1 , 𝑡2 , … , 𝑡5 )T
to the calculated 𝐓∗ = (𝑡∗1 , 𝑡∗2 , … , 𝑡∗5 )T . Once resources are allocated to CSFs prior to ERP implementation,
In Case 1, the observed T = (30, 40, 30, 10, 10)T is very different the next step is to develop an implementation strategy realising resource
from the calculated T∗ = (33, 16, 39, 17, 9)T , with the proximity indica- allocation to CSFs by finding out if CSFs should be addressed sequen-
tor PI being 0.69. PF = 65% in the Baseline, is lower than the PF = 73% tially or simultaneously, or a combination of both. SMEs tend to focus
achieved in Scenario 2. In the observed data, Case 1 allocated more time on one CSF and finish it before moving to the next; however, CSFs are
(40 days) to Users than to Top Management (30 days) as Users are more inter-related and should be considered in all phases of ERP implemen-
compliant and have more time to spend on implementation. It is also tation (Sun et al., 2005). CNL_ERP can provide guidance in developing
less expensive to address Users as a CSF. According to CNL_ERP, more implementation strategies by ensuring the pre-determined performance
time should be allocated to Top Management (33 days) than to Users level is achieved under budget and time limits.
(16 days). The MIS Manager in Case 1 explained that ‘… in SMEs, se- Using Case 1 as an example, an implementation strategy can be de-
nior managers do not usually have experience of implementing ERP projects veloped by maximising the performance level in different time periods
and also have many other responsibilities. …. In an SME, ERP implementa- while the budget is controlled to be under $100,000. Substituting val-
tion is a choice; to implement it, your manager has to be convinced of the ues of 𝑝𝑖 and 𝑘𝑖 to (6), the reduced gradient of PF is given in equation
benefit. The manager did not have great interest in ERP implementation, and (23), and dimensional elements 𝜕PF𝜕𝑡
are plotted in Fig. 2. From a math-
𝑖
he does not have time for it either … while users are more willing to spend ematical point of view, points 𝐷𝑛 (𝑡𝑖 , 𝜕PF
𝜕𝑡
) (𝑛 = 1, 2 … , 8) in Fig. 2 are
time learning the new software. The cost of having more users in the project is 𝑖
𝜕PF
much cheaper than having the upper management team involved.” Although where 𝜕𝑡𝑖
intersect, and the red arrows indicate how the search di-
a strong commitment from Top Management would be extremely valu- rection for 𝐓∗ changes when TT increases; while in practice, points 𝐷𝑖
able to the ERP implementation, Top Management in many SMEs often break the implementation period into 8 phases (𝐷5 occurs almost at the
cannot afford this kind of commitment, and instead would involve al- same time as 𝐷6 when TT = 47) and resource is allocated to CSF 𝑖 based
ternative resources such as Users. This conclusion is consistent with the on their priorities which are determined by 𝜕PF
𝜕𝑡
and corresponding con-
𝑖
conclusion drawn from Plaza and Rohlf (2008). straints. The numerical values of 𝐷𝑖 (𝑡𝑖 , 𝜕PF
𝜕𝑡
) and associated TT are listed
𝑖
Cases 2 and 3 have the same amount of resources in terms of time
in Table 6, based on which an implementation strategy is developed in
and budget available, therefore the calculated solutions are identical as
8 phases and resources are allocated to CSFs according to their priori-
T∗ = (30, 15, 35, 15, 8)T . Comparing the results from the two cases, we
ties (see Fig. 3). The priorities of CSFs do not stay the same but change
observed that:
along with the constraints imposed at each phase.

1) The observed T = (28, 28, 21, 15, 15)T in Case 2 is more similar to ∇𝑟 PF(𝐓)
the T∗ = (30, 15, 35, 15, 8)T recommended by CNL_ERP, with the ( )
𝜕PF 𝜕PF 𝜕PF
PI = 0.71. Time allocation in Case 1, T = (20, 40, 20, 10, 20)T , is = , ,…,
𝜕𝑡1 𝜕𝑡2 𝜕𝑡5
more distant from T∗ = (30, 15, 35, 15, 8)T , having the PI = 0.62. ( )
2) As a result of the closeness to the ideal solution, a higher perfor- = 0.86𝑒−0.045𝑡1 , 2.79𝑒−0.163𝑡2 , 0.97𝑒−0.04𝑡2 , 1.47𝑒−0.076𝑡2 , 1.85𝑒−0.143𝑡5
mance level is achieved in Case 2 with PF = 70% as opposed to (23)
PF = 63% obtained in Case 3.
3) The results presented in Table 5 confirm that allocating time as rec- Since project planning is assumed to be complete, an implemen-
ommended by CNL_ERP produces higher performance levels while tation strategy can be obtained as follows when the budget limit is
incurring lower costs. It should be noted that, generally, focusing $100,000 and time limit is 120 days: in Phase 1 (0∼12 days), once
efforts and resources on CSFs as recommended by CNL_ERP leads agreement and support have been obtained from TM, and the project
to better outcomes. team has been formed up in the planning phase, TM and PM do not
need to make much commitment to the project, while training should
However, the interview results from Case 1 indicate that the most be provided to Users and data migration should be started with IT and
effective approach is to combine expert judgement with the CNL_ERP VS; in Phase 2 (13-24 days), PM will start while the main focus is still
recommendation and make appropriate adjustments to the CNL_ERP on Users, IT and VS; in Phase 3 (25-32 days), TM starts making commit-
recommendation wherever appropriate. ment, for example, checking project progress and providing necessary

8
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

Table 6. Numerical solutions of 𝐷𝑖 and associated TT.

𝐷𝑛 𝐷1 𝐷2 𝐷3 𝐷4 𝐷5 𝐷6 𝐷7 𝐷8

𝐷𝑛 (𝑡𝑖 , 𝜕𝑃𝜕𝑡𝐹 ) (3.43, 1.13) (7.37, 0.84) (6.27, 0.75) (8.59, 0.69) (11.55, 0.61) (7.82, 0.60) (9.97, 0.60) (17.29, 0.39)
𝑖

𝑇 𝑇 (days) 12 24 32 38 47 48 54 78

Fig. 3. ERP implementation strategy developed by CNL_ERP.

support; in Phases 4, 5, and 6 (33-54 days), PM takes over VS, Users, The maximal ΔPF 𝑖 is obtained when Δ𝑘𝑖 → ∞, hence
and IT respectively, and should be allocated more time; once Users are
trained and data migration is nearly done, PM should be in charge of
lim ΔPF 𝑖 = 𝑝𝑖 ⋅ 𝑒−𝑘𝑖 𝑡𝑖 (26)
the whole project and monitor its progress until success occurs; finally, Δ𝑘𝑖 →∞
in Phases 7 and 8 (55-115 days), TM takes over Users and IT and be-
comes the second most important CSF, which should be addressed with The progressing coefficient 𝑘𝑖 is directly related to the rate of per-
an increased time resource in order to ensure that the implementation formance improvement. Changing 𝑘𝑖 causes a shift in the CSF’s progress
is successful and within budget. level, noted as ΔPF 𝑖 , which leads to a shift in the overall ERP implemen-
The strategy for implementing ERP was developed from mathemati- tation performance level. Applying (25) to Case 1, in Fig. 4 we depict
cal analysis of ERP implementation performance, and the MIS manager the impacts of progressing coefficient variation (Δ𝑘𝑖 = 0.01) on a CS-
in Case 1 explained that a real project may not be able to follow each F’s progress level change (ΔPF𝑖 ) as a function of 𝑡𝑖 . Fig. 4 shows that
phase as suggested here as each project is situational and has different an identical change in the progressing coefficient (Δ𝑘𝑖 = 0.01) has a dif-
characteristics. However, he advised that the plan obtained here does ferent impact on different CSFs, with the order of impact being, from
provide a rich picture of where to concentrate effort in the initial, in- strongest to weakest: CSF 3 (PM), CSF 1 (TM), CSF 4 (IT), CSF 2 (Users)
termediate and final phases, and is very helpful in enabling an SME to and CSF 5 (VS). At the same time point 𝑡𝑖 and with the same level Δ𝑘𝑖 ,
understand the progress of an ERP project and the resources needed, the CSF 𝑖 with the higher 𝑝𝑖 𝑒−𝑘𝑖 𝑡𝑖 achieves a higher ΔPF 𝑖 . As shown in
and therefore make appropriate arrangements or preparations. For ex- Fig. 4 and Table 7, when Δ𝑘𝑖 = 0.01, the maximum ΔPF 𝑖 is achieved at
ample, in initial phases 1 and 2, full time staff can be released from PM the point 𝐵𝑖 (𝑡𝑖 , ΔPF 𝑖 ). Among all the CSFs, CSF 3 achieves the highest
and TM teams while VS and Users need to be in place; in intermediate value of 𝐵3 = 1.99, while CSF 5 obtains the lowest 𝐵5 = 0.32. The im-
phases 3 and 4, some but not all staff from PM and TM need to work on pact of the progressing speed increment Δ𝑘5 on CSF 5 ’s progress level is
the project; and in final phases 5 to 8, demands on PM and TM become less significant in comparison with other CSFs. This shows that training
more intensive and staff accountable should be reserved for this period, provided to CSFs that have low potential (i.e., low 𝑝𝑖 𝑒−𝑘𝑖 𝑡𝑖 ) would be un-
while VS will not be needed and staff from IT and Users training can be productive in terms of improving the progress level or the overall ERP
reduced. implementation performance level. The results presented in Fig. 4 and
Table 7 confirm that the most cost-effective strategy is to offer train-
4.5. Impacts of progressing coefficient 𝑘𝑖
ing to CSFs that have a high potential in making contributions to the
overall implementation performance, i.e. CSF 3 .
The case studies in Section 4.3 provide helpful guidance when al-
We calculate the values of ΔPF 𝑖 when Δ𝑘𝑖 = 0.01 at the extreme
locating resources prior to ERP implementation. However, if extra re-
point T∗ = (33, 16, 39, 17, 9), as shown in the second column in Table 7.
source, such as budget, becomes available to provide ERP training,
which CSF (or the team that addresses the CSF) shall we make in- Compared with other CSFs, CSF 3 (PM) and CSF 1 (TM) are more sensi-
vestment in? Training increases the progress coefficient 𝑘𝑖 which is the tive to Δ𝑘𝑖 . Increasing the progressing speed by Δ𝑘𝑖 = 0.01 for every CSF
critical parameter affecting the overall project duration and implemen- leads to an increment in the overall performance level by ΔPF = 4.19.
tation performance. A change Δ𝑘𝑖 will cause change ΔPF𝑖 , as calculated As would be expected, the amount of resource and effort required to in-
in (24)-(26): crease the CSF progressing speed by Δ𝑘𝑖 = 0.01 varies from case to case.
SMEs could use CNL_ERP as a tool to predict the expected increment in
ΔPF 𝑖 = PF 𝑖 (𝑘𝑖 + Δ𝑘𝑖 ) − PF 𝑖 (𝑘𝑖 ) (24) the overall implementation performance level, and to identify where to
Substituting (1) to (24), invest in CSFs to increase processing speed. CNL_ERP can also be used
to judge if the required investment could be offset by the performance
( )
ΔPF 𝑖 = 𝑝𝑖 ⋅ 𝑒−𝑘𝑖 𝑡𝑖 ⋅ 1 − 𝑒−Δ𝑘𝑖 𝑡𝑖 (25) improvement, and to facilitate decision making.

9
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

Fig. 4. Changes of CSF’s progress ΔPF 𝑖 as a function of 𝑡𝑖 , when Δ𝑘𝑖 = 0.01.

Table 7. Impacts of 𝑘𝑖 on CSF 𝑖 ’s progress PF 𝑖 .

T∗ = (33, 16, 39, 17, 9) Δ𝑘𝑖 = 0.01
Δ𝑘𝑖 = 0.01

ΔPF 1 1.21 𝐵1 (20, 1.40* )

ΔPF 2 0.19 𝐵2 (6, 0.38* )

ΔPF 3 1.65 𝐵3 (22, 1.99* )

ΔPF 4 0.83 𝐵4 (12, 0.88* )

ΔPF 5 0.31 𝐵5 (7, 0.32* )

ΔPF 4.19
*
when Δ𝑘𝑖 = 0.01, the maximal ΔPF 𝑖 is achieved at
point 𝐵𝑖

5. Conclusions data collected from sample SMEs. Hence, it could be used in the pre-
implementation stage to support SMEs with similar characteristics to
5.1. Theoretical contributions determine the priorities of CSFs and acquire necessary resources; it
could facilitate decision making in SMEs about how much and when
Notwithstanding the fact that much of the ERP literature has identi- to invest in each CSF in the implementation stage. CNL_ERP can also
fied CSFs and their association to the implementation of ERP systems, be used as a tool to accurately measure the progress made and the
there is a lack of quantitative measurement of CSFs’ contributions to cost incurred for each CSF, and the progress of the overall ERP im-
overall ERP implementation performance. Such a research gap results plementation. Furthermore, as CNL_ERP explicitly incorporates several
in a fragmentary understanding of how much and when to invest in managerial decisions, it can be used as a tool to fine-tune the model’s
CSFs to achieve target ERP implementation performance. Specific con- behaviour, such as reaching maximum performance level targets set by
tributions from this paper to the ERP literature are: 1) quantitatively management while satisfying resource constraints.
associating CSFs with ERP implementation and assessing the perfor- This research also uses the generalised reduced gradient approach
mance contributed by each CSF; 2) offering a useful analytical tool to to help SMEs develop appropriate ERP implementation strategy, with a
accurately monitor progress made, and cost incurred, by each of the focus on resource allocation/planning decision at different stages where
CSFs against time; 3) empirically testing and validating the effectiveness resources vary.
of investment in CSFs on ERP project success; and 4) demonstrating the
potential of a nonlinear programming model as a method for planning, 5.3. Limitations and future work
acquiring and deploying resources in IT implementation.
Furthermore, a specific contribution made to the resource alloca- This paper presents an innovative quantitative model for ERP im-
tion literature is that this research provides a solid theoretical basis for plementation, namely CNL_ERP. Empirical studies show that CNL_ERP
studying CSF based resource allocations in projects to implement ERP closely resembles the actual ERP implementation performances and that
systems. focusing efforts and resources on CSFs as recommended by CNL_ERP
leads to better outcomes. Priorities of CSFs vary according to the re-
5.2. Managerial contributions sources (budget limitations) available, so focus on CSFs needs to be
adjusted accordingly. CNL_ERP works by giving preference to the lower
SMEs are significant actors in generating global economic growth. cost and faster progressing CSFs when the budget is low and shifting
The advent of Cloud-based ERP services offers them greater opportuni- preference to the medium cost and slower progressing factors, which
ties to streamline their business activities through ERP implementation. have the potential to achieve a high asymptotic maximum progress
In a similar manner to on-premises ERP solutions, Cloud-based ERP level, when increased budget is available. CNL_ERP switches prefer-
implementation requires adequate planning, resource acquisition and ences between CSFs in order to achieve the highest performance level
resource deployment to prepare an SME for the project. The quan- for the lowest cost.
titative model developed in this study, i.e., CNL_ERP, assesses CS- We also conclude that in cases where TM are very committed to the
Fs’ contributions to ERP implementation performance using empirical ERP implementation project (investing time and budget), or Time and

10
Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

Budget are invested to PM (project team training, improving compe- Define

tence), the ERP implementation performance will be enhanced.
The results of this research are: 1) The cost of ERP implementa- 𝐓 = (𝐘, 𝐙)
tion increases along the time horizon, whereas the performance level Let
reaches a maximum and then stagnates; 2) Priorities of CSFs vary ac-
cording to the resources available (budget limitations), so focus on CSFs 𝐘 be basic variables, 𝐙 be nonbasic variables, and 𝐠 = (𝑔1 , 𝑔2 )T
needs to be adjusted accordingly. We also conclude that, in cases where
TM are very committed to the ERP implementation project (investing The first partial derivatives of function PF(𝐓) with respect to 𝐘 and 𝐙
time and budget) or Time and Budget are invested to PM (project team is:
training, improving competence), the ERP implementation performance
will be enhanced. 𝜕PF(𝐘, 𝐙) = ∇𝐘 PF𝜕𝐘 + ∇𝐙 PF𝜕𝐙 (A.2)
Our analysis offers managers direct insights into CSF based resources The first partial derivatives of functions 𝑔1 (𝐓) and 𝑔2 (𝐓) with respect
(time and budget) when planning allocations, but it could be difficult to 𝐘 and 𝐙 are:
to generalise the quantitative model to SMEs who do not meet the sam-
pling criteria in Table 2. Future research will expand the sampling pool 𝜕𝐠(𝐘, 𝐙) = ∇𝐘 𝐠𝜕𝐘 + ∇𝐙 𝐠𝜕𝐙 = 0 (A.3)
to involve SMEs with diverse characteristics and refine the CNL_ERP
model using extra data sets. Further data collection and study is re- Define:
( )
quired to examine how the CSFs’ progressing coefficients are influenced ∇ 𝐘 𝑔1
by external consulting, staff training and staff allocation; CNL_ERP can Jacobian matrix 𝐉 = ∇𝐘 𝐠 = (A.4)
∇ 𝐘 𝑔2
identify which resources are required to reach the necessary progress-
ing speeds, thus enhancing and enabling ERP project planning and the ( )
development of training plans. ∇ 𝐙 𝑔1
Control matrix 𝐂 = ∇𝐙 𝐠 = (A.5)
∇ 𝐙 𝑔2
Declarations Substituting (A.4) and (A.5) to (A.3), we have

Author contribution statement 𝜕𝐠 = 𝐉𝜕𝐘 + 𝐂𝜕𝐙. (A.6)

Ying Xie: Conceived and designed the experiments; Analyzed and Given 𝐠(𝐓) = 0 and 𝐉 is non-singular (the components of the vector
interpreted the data; Contributed reagents, materials, analysis tools or 𝐘 must be selected such that 𝐉 is non-singular), it follows that
data; Wrote the paper.
Colin Allen: Conceived and designed the experiments; Analyzed 𝜕𝐘 = −𝐉−𝟏 𝐂𝜕𝐙 (A.7)
and interpreted the data; Wrote the paper. Substituting (A.7) to (A.2), the reduced gradient vector of PF with
Mahmood Ali: Performed the experiments; Contributed reagents, respect to the nonbasic variable 𝐙 is:
materials, analysis tools or data.
𝜕PF
∇𝑟 PF = = ∇𝐙 PF − ∇𝐘 PF𝐉−𝟏 𝐂 (A.8)
Funding statement 𝜕𝐙
Let 𝐘 = (𝑆1 , 𝑆2 )T , and 𝐙 = 𝐓 = (𝑡1 , 𝑡2 , … , 𝑡𝑀 )T , we obtain the reduced
This research did not receive any specific grant from funding agen- gradient vector of PF with respect to 𝐙 (or 𝐓):
cies in the public, commercial, or not-for-profit sectors. ( )
𝜕PF 𝜕PF
∇𝐘 PF = , = (0, 0) (A.9)
Data availability statement 𝜕𝑆1 𝜕𝑆2

Data included in article/supp.material/referenced in article. ( )

𝜕PF 𝜕PF 𝜕PF
∇𝐳 PF = , ,…,
𝜕𝑡1 𝜕𝑡2 𝜕𝑡𝑀
Declaration of interests statement ( )
= 𝑝1 𝑘1 𝑒−𝑘1 𝑡1 , 𝑝2 𝑘2 𝑒−𝑘2 𝑡2 , … , 𝑝𝑀 𝑘𝑀 𝑒−𝑘𝑀 𝑡𝑀 (A.10)
The authors declare no conflict of interest.
( 𝜕𝑔1 𝜕𝑔1 ) ( ) ( )
Additional information 𝜕𝑆1 𝜕𝑆2 1 0 1 0
𝐉= = , thus, 𝐉−𝟏 = (A.11)
𝜕𝑔2 𝜕𝑔2
0 1 0 1
No additional information is available for this paper. 𝜕𝑆1 𝜕𝑆2
( 𝜕𝑔1 𝜕𝑔1 𝜕𝑔1 ) ( )
𝜕𝑡1 𝜕𝑡2
,…, 𝜕𝑡𝑀 1 1, … , 1
Appendix A. Generalised reduced gradient 𝐂 = 𝜕𝑔 𝜕𝑔2 𝜕𝑔2 = (A.12)
2
,…, 𝑑1 𝑑2 , … , 𝑑𝑀
𝜕𝑡1 𝜕𝑡2 𝜕𝑡𝑀

Converting the inequality constraints 𝑔1 and 𝑔2 in (5.a) into equa- Substituting (A.9)-(A.12) to (A.8), the reduced gradient vector of PF
tions by adding nonnegative slack variables 𝑆1 and 𝑆2 , the nonlinear is:
optimisation problem becomes:
( )
𝑀 ∇𝑟 PF = 𝑝1 𝑘1 𝑒−𝑘1 𝑡1 , 𝑝2 𝑘2 𝑒−𝑘2 𝑡2 , … , 𝑝𝑀 𝑘𝑀 𝑒−𝑘𝑀 𝑡𝑀 (A.13)
∑
Max PF(𝐓) = 𝑝𝑖 ⋅ (1 − 𝑒−𝑘𝑖 𝑡𝑖 ) (A.1) Extreme point 𝐓∗
is obtained through a number of iterations where
𝑖=1
𝐓 is increased from 0 to 𝐓∗ in the search direction governed by ∇𝑟 PF.
𝑀
∑ Element 𝑡𝑖 of 𝐓 is increased in the order of the value of 𝑝𝑖 𝑘𝑖 𝑒−𝑘𝑖 𝑡𝑖 , subject
s.t. 𝑔1 (𝐓) = 𝑡𝑖 − TL + 𝑆1 = 0
𝑖=1
to constraints 𝑔1 and 𝑔2 are satisfied. The higher the value of 𝑝𝑖 𝑘𝑖 𝑒−𝑘𝑖 𝑡𝑖 ,
𝑀 the higher the priority hence 𝑡𝑖 in increased.
∑
𝑔2 (𝐓) = cost𝑖 (𝑡𝑖 ) − CL + 𝑆2 = 0
𝑖=1 Appendix B. Survey results
𝐓≥0

11
 Y. Xie, C. Allen and M. Ali
Company Criteria CSF1-TM CSF2-Users CSF3-PM CSF4-IT CSF5-VS Total Company Criteria CSF1-TM CSF2-Users CSF3-PM CSF4-IT CSF5-VS Total
1 Time (days) 30 40 30 10 10 120 31 Time (days) 7 28 21 14 14 84
Cost ($) 11,000 22,000 37,500 19,500 10,000 $100,000 Cost ($) 1,680 5,600 16,800 14,000 17,920 $56,000
Progress (%) 7 15 10 11 7 65 Progress (%) 3 15 13 13 13 55

2 Time (days) 1 4 3 3 3 14 32 Time (days) 10 20 30 20 20 100

Cost ($) 0 5,250 3,000 3,750 3,750 $15,750 Cost ($) 2,130 10,650 17,750 21,300 19,170 $71,000
Progress (%) 0 20 0 0 0 20 Progress (%) 3 9 11 6 7 35

3 Time (days) 18 9 36 108 9 180 33 Time (days) 10 21 28 15 6 80

Cost ($) 1,500 1,500 3,000 15,000 1,500 $22,500 Cost ($) 3,950 14,220 18,960 16,590 16,590 $70,310
Progress (%) 3 10 8 25 5 50 Progress (%) 7 27 23 16 9 80

4 Time (days) 20 5 40 30 5 100 34 Time (days) 14 25 35 50 26 150

Cost ($) 4,125 16,500 24,750 24,750 12,375 $82,500 Cost ($) 4,050 27,000 33,750 54,000 16,200 $135,000
Progress (%) 10 25 25 25 15 100 Progress (%) 5 20 10 15 10 60

5 Time (days) 10 50 20 20 20 120 35 Time (days) 10 20 25 20 5 80

Cost ($) 5,000 10,000 10,000 12,500 12,500 $50,000 Cost ($) 10,000 20,000 25,000 20,000 5,000 $80,000
Progress (%) 13 20 20 18 10 80 Progress (%) 10 20 25 20 5 80

6 Time (days) 8 20 16 30 16 90 36 Time (days) 5 12 18 10 5 50

Cost ($) 5,880 39,200 13,720 78,400 58,800 $196,000 Cost ($) 3,150 6,300 22,050 18,900 12,600 $63,000
Progress (%) 7 18 24 22 11 80.5 Progress (%) 1 6.5 5 5 2.5 20

7 Time (days) 30 30 30 60 30 180 37 Time (days) 14 28 30 24 14 110

Cost ($) 45,000 45,000 60,000 60,000 45,000 $255,000 Cost ($) 12,600 25,200 35,000 51,800 15,400 $140,000
Progress (%) 28 5 23 8 8 70 Progress (%) 3 23 23 21 12 80
12

8 Time (days) 30 60 90 90 30 300 38 Time (days) 0 14 20 10 6 50

Cost ($) 11,250 33,750 45,000 101,250 33,750 $225,000 Cost ($) 0 10,000 10,000 17,500 12,500 $50,000
Progress (%) 25 8 25 15 8 80 Progress (%) 0 0 0 0 0 0

9 Time (days) 30 130 60 100 40 360 39 Time (days) 10 17 25 27 11 90

Cost ($) 65,000 65,000 65,000 65,000 65,000 $325,000 Cost ($) 4,000 12,000 17,600 28,000 18,400 $80,000
Progress (%) 10 10 30 25 5 80 Progress (%) 5 30 18 28 10 90

10 Time (days) 18 18 18 18 18 90 40 Time (days) 14 35 42 56 33 180

Cost ($) 6,000 6,000 6,000 6,000 6,000 $30,000 Cost ($) 16,000 40,000 56,000 68,000 2,000 $182,000
Progress (%) 14 14 14 14 14 70 Progress (%) 3 13 10 10 5 40

11 Time (days) 20 20 30 40 10 120 41 Time (days) 0 28 30 35 27 120

Cost ($) 2,500 10,000 12,500 17,500 7,500 $50,000 Cost ($) 0 11,250 15,000 30,000 18,750 $75,000
Progress (%) 10 15 19 18 12 73 Progress (%) 0 25 25 25 15 90

12 Time (days) 20 40 20 10 20 110 42 Time (days) 14 40 49 63 44 210

Cost ($) 4,750 32,000 23,000 16,000 14,250 $90,000 Cost ($) 9,000 60,000 84,000 111,000 36,000 $300,000
Progress (%) 10 15 13 10 15 63 Progress (%) 4 16 15 11 15 60

13 Time (days) 30 60 70 70 40 270 43 Time (days) 5 12 15 18 10 60

Cost ($) 9,800 20,000 50,400 100,800 100,000 $281,000 Cost ($) 2,400 6,000 10,800 14,000 6,800 $40,000

Heliyon 8 (2022) e10044

Progress (%) 18 13 23 23 5 80 Progress (%) 2 8 8 7 7 30

14 Time (days) 2 20 10 50 18 100 44 Time (days) 9 21 19 11 10 70

Cost ($) 30,000 6,000 12,000 6,000 6,000 $60,000 Cost ($) 7,000 29,400 28,000 42,000 33,600 $140,000
Progress (%) 0 10 5 35 10 60 Progress (%) 5 8 9 10 9 40
 Y. Xie, C. Allen and M. Ali
Company Criteria CSF1-TM CSF2-Users CSF3-PM CSF4-IT CSF5-VS Total Company Criteria CSF1-TM CSF2-Users CSF3-PM CSF4-IT CSF5-VS Total
15 Time (days) 10 25 30 35 20 120 45 Time (days) 7 21 21 21 14 84
Cost ($) 8,000 16,000 20,000 24,000 12,000 $80,000 Cost ($) 2,920 11,680 14,600 25,550 18,250 $73,000
Progress (%) 14 13 23 13 17 80 Progress (%) 10 36 20 22 12 100

16 Time (days) 45 30 180 180 30 465 46 Time (days) 10 21 28 37 19 115

Cost ($) 40,000 80,000 100,000 80,000 20,000 $320,000 Cost ($) 3,300 51,150 39,600 54,450 16,500 $165,000
Progress (%) 10 20 25 20 5 80 Progress (%) 6 15 15 14 11 60

17 Time (days) 14 56 21 70 21 182 47 Time (days) 7 21 28 21 13 90

Cost ($) 15,000 75,000 50,000 175,000 85,000 $400,000 Cost ($) 1,350 24,300 39,150 40,500 29,700 $135,000
Progress (%) 7.5 10 10 13 10 50 Progress (%) 0 10 7 7 7 31

18 Time (days) 28 28 21 15 15 107 48 Time (days) 0 14 20 10 6 50

Cost ($) 20,000 30,000 10,000 20,000 10,000 $90,000 Cost ($) 0 12,720 11,660 15,900 12,720 $53,000
Progress (%) 15 15 15 15 10 70 Progress (%) 0 26 18 19 8 70

19 Time (days) 14 28 21 21 30 114 49 Time (days) 5 14 18 18 8 63

Cost ($) 18,000 45,000 27,000 63,000 27,000 $180,000 Cost ($) 810 16,200 21,870 26,730 15,390 $81,000
Progress (%) 15 20 10 18 8 70 Progress (%) 5 18 10 14 14 60

20 Time (days) 10 45 35 28 14 132 50 Time (days) 0 7 7 7 7 28

Cost ($) 20,000 60,000 20,000 60,000 40,000 $200,000 Cost ($) 0 6,150 10,250 14,350 10,250 $41,000
Progress (%) 15 25 10 13 8 70 Progress (%) 0 5 5 5 5 20

21 Time (days) 90 90 90 90 90 450 51 Time (days) 12 21 30 33 19 115

Cost ($) 50,000 50,000 100,000 100,000 200,000 $500,000 Cost ($) 8,850 53,100 44,250 53,100 17,700 $177,000
13

Progress (%) 21 15 21 17 17 90 Progress (%) 8 25 20 13 15 80

22 Time (days) 14 76 28 21 21 160 52 Time (days) 0 12 20 12 11 55

Cost ($) 9,000 81,000 24,000 150,000 36,000 $300,000 Cost ($) 0 11,850 15,800 31,600 19,750 $79,000
Progress (%) 21 18 19 17 17 90.75 Progress (%) 0 7 8 7 9 30

23 Time (days) 21 52 28 28 28 157 53 Time (days) 7 28 18 21 6 80

Cost ($) 11,250 45,000 13,500 123,750 34,875 $228,375 Cost ($) 2,760 15,640 23,000 27,600 23,000 $92,000
Progress (%) 21 21 21 20 17 100 Progress (%) 13 24 19 18 18 90

24 Time (days) 7 35 14 14 14 84 54 Time (days) 7 21 21 28 13 90

Cost ($) 2,550 29,700 6,800 40,000 5,950 $85,000 Cost ($) 0 10,710 15,750 22,050 14,490 $63,000
Progress (%) 19 18 20 18 15 90 Progress (%) 0 6 4 4 6 20

25 Time (days) 4 4 10 4 6 28 55 Time (days) 14 35 35 60 41 185

Cost ($) 800 6,000 6,000 5,200 2,000 $20,000 Cost ($) 7600 32300 32300 66500 51300 $190,000
Progress (%) 0 3 3 3 3 10 Progress (%) 5 20 15 20 17.5 77.5

26 Time (days) 7 21 18 20 14 80 56 Time (days) 12 32 35 42 24 145

Cost ($) 3,500 14,700 12,600 14,000 9,800 $54,600 Cost ($) 2,340 23,400 28,080 40,950 22,230 $117,000
Progress (%) 10 28 23 15 15 90 Progress (%) 7.5 18 20.5 20 14 80

Heliyon 8 (2022) e10044

27 Time (days) 0 30 35 30 25 120 57 Time (days) 4 17 14 12 9 56
Cost ($) 0 58,500 39,000 78,000 19,500 $195,000 Cost ($) 1,020 6,120 10,200 17,850 15,810 $51,000
Progress (%) 0 33 25 20 13 90.5 Progress (%) 3 8 7 7 5 30
(continued on next page)
 Y. Xie, C. Allen and M. Ali
Company Criteria CSF1-TM CSF2-Users CSF3-PM CSF4-IT CSF5-VS Total Company Criteria CSF1-TM CSF2-Users CSF3-PM CSF4-IT CSF5-VS Total
28 Time (days) 8 23 17 35 20 103 58 Time (days) 0 14 21 35 15 85
Cost ($) 3,000 36,250 24,650 50,750 30,450 $145,100 Cost ($) 0 10,800 15,840 21,600 23,760 $72,000
Progress (%) 4 17 14 11 5 50 Progress (%) 0 29 24 20 18 90

29 Time (days) 21 35 25 39 20 140 59 Time (days) 8 14 20 12 11 65

14

Cost ($) 16,800 42,000 31,500 94,500 42,000 $226,800 Cost ($) 1,740 15,660 22,620 25,230 21,750 $87,000
Progress (%) 3 21 14 29 14 80 Progress (%) 6 19 19 21 17 80

30 Time (days) 5 42 33 65 25 170 60 Time (days) 8 12 10 15 5 50

Cost ($) 2,100 16,100 21,000 21,700 9,100 $70,000 Cost ($) 5,200 12,350 14,950 17,550 14,950 $65,000
Progress (%) 7.5 32.5 19.5 22.5 8 90 Progress (%) 2 2 2 2 2 10

Heliyon 8 (2022) e10044

Y. Xie, C. Allen and M. Ali Heliyon 8 (2022) e10044

References Malik, M.O., Khan, N., 2020. Analysis of ERP implementation to develop a strategy for its
success in developing countries. Prod. Plan. Control 32 (12), 1020–1035.
Ahmad, N., Mehmood, R., 2016. Enterprise systems and performance of future city logis- Nagpal, S., Khatri, S.K., Kumar, A., 2015. Comparative study of ERP implementation
tics. Prod. Plan. Control 27 (6), 500–513. strategies. In: 2015 Long Island Systems, Applications and Technology, pp. 1–9.
Ahmadi, S., Yeh, C.H., Martin, R., Papageorgiou, E., 2015. Optimising ERP readiness Nwankpa, J.K., 2015. ERP system usage and benefit: a model of antecedents and out-
improvements under budgetary constraints. Int. J. Prod. Econ. 161, 105–115. comes. Comput. Hum. Behav. 45, 335–344.
Ahmadzadeh, A., Aboumasoudi, A.S., Shahin, A., Teimouri, H., 2021. Studying the critical Nwankpa, J., Roumani, Y., 2014. Understanding the link between organizational learn-
success factors of ERP in the banking sector: a DEMATEL approach. Int. J. Procur. ing capability and ERP system usage: an empirical examination. Comput. Hum. Be-
Manag. 14 (1), 126–145. hav. 33, 224–234.
Ahn, B., Ahn, H., 2020. Factors affecting intention to adopt cloud-based ERP from a Parr, A., Shanks, G., 2000. A model of ERP project implementation. J. Inf. Technol. 15,
comprehensive approach. Sustainability 12 (16), 6426. 289–303.
Alharthi, A., Alassafi, M.O., Walters, R.J., Wills, G.B., 2017. An exploratory study for Phillips, Bana e Costa, 2007. Transparent prioritisation, budgeting and resource allocation
investigating the critical success factors for cloud migration in the Saudi Arabian with multi-criteria decision analysis and decision conferencing. Ann. Oper. Res. 154,
higher education context. Telemat. Inform. 34 (2), 664–678. 51–68.
Baykasoğlu, A., Gölcük, İ., 2017. Development of a two-phase structural model for eval- Plaza, M., Rohlf, K., 2008. Learning and performance in ERP implementation projects:
uating ERP critical success factors along with a case study. Comput. Ind. Eng. 106, a learning curve model for analyzing and managing consulting costs. Int. J. Prod.
256–274. Econ. 115, 72–85.
Bhatt, N., Guru, S., Thanki, S., Sood, G., 2021. Analysing the factors affecting the selection Plaza, M., Ngwenyama, O., Rohlf, K., 2010. A comparative analysis of learning curves:
of ERP package: a fuzzy AHP approach. Inf. Syst. E-Bus. Manag. 19 (2), 641–682. implications for new technology implementation management. Eur. J. Oper. Res. 200,
Chatti, H., Radouche, T., Asfoura, E., 2021. Framework for the evaluation of the ERP 518–528.
implementation success: case study in SMEs. J. Manag. Inf. Dec. Sci. 24 (4), 1–25. Ram, J., Corkindale, D., Wu, M.L., 2013. Implementation critical success factors (CSFs)
Dao, B., Kermanshachi, S., Shane, J., Anderson, S., Damnjanovic, I., 2020. Developing a for ERP: do they contribute to implementation success and post-implementation per-
logistic regression model to measure project complexity. Archit. Eng. Design Manag., formance? Int. J. Prod. Econ. 144, 157–174.
1–15. Saade, R.G., Nijher, H., 2016. Critical success factors in enterprise resource planning im-
Dardan, S., Busch, D., Sward, D., 2006. An application of the learning curve and the plementation: a review of case studies. J. Enterp. Inf. Manag. 45 (6), 99–113.
nonconstant-growth dividend model: IT investment valuations at Intel Corporation. Sun, A.Y.T., Yazdani, A., Overend, J.D., 2005. Achievement assessment for enterprise
Decis. Support Syst. 41 (4), 688–697. resource planning (ERP) system implementations based on critical success factors
Fosso-Wamba, S., Akter, S., Coltman, T., Ngai, E.W.T., 2015. Guest editorial: informa- (CSFs). Int. J. Prod. Econ. 98, 189–203.
tion technology-enabled supply chain management. Produc. Plan. Control Manag. Sun, H., Ni, W., Lam, R., 2015. A step-by-step performance assessment and improvement
Operat. 26 (12), 933–944. method for ERP implementation: action case studies in Chinese companies. Comput.
Gupta, S., Misra, S.C., 2016. Compliance, network, security and the people related factors Ind. 68, 40–52.
in cloud ERP implementation. Int. J. Commun. Syst. 29 (8), 1395–1419. Sun, H., Ni, W., Lam, R., Ng, C.Y., 2016. A stage-by-stage assessment of enterprise re-
Gupta, S., Misra, S.C., Kock, N., Roubaud, D., 2018. Organizational, technological and source planning implementation: an empirical study from Hong Kong. J. Global
extrinsic factors in the implementation of cloud ERP in SMEs. J. Organ. Change Inform. Technol. Manag. 19 (2), 104–127.
Manag. 31 (1), 83–102. Tarhini, A., Ammar, H., Tarhini, T., e, Masa’deh R., 2015. Analysis of the critical success
Hashem, I.A.T., Yaqoob, I., Anuar, N.B., Mokhtar, S., Gani, A., Khan, S.U., 2015. The rise factors for enterprise resource planning implementation from stakeholders’ perspec-
of “big data” on cloud computing: review and open research issues. Inf. Syst. 47, tive: a systematic review. Int. Bus. Res. 8 (4), 25–40.
98–115. Vargas, M.A., Comuzzi, M., 2020. A multi-dimensional model of enterprise resource plan-
Hasibuan, Z.A., Dantes, G.R., 2012. Priority of key success factors (KSFS) on enterprise ning critical success factors. Enterp. Inf. Syst. 14 (1), 38–57.
resource planning (ERP) system implementation life cycle. J. Enterprise Resource Verdouw, C.N., Robbemond, R.M., Wolfert, J., 2015. ERP in agriculture: lessons learned
Planning Stud. 2012, 122627. from the Dutch horticulture. Comput. Electron. Agric. 114, 125–133.
Hwang, C.L., Yoon, K., 1981. Multiple Attribute Decision Making: Methods and Applica- Xie, Y., Allen, C.J., Ali, M., 2014. An integrated decision support system for ERP im-
tions. Springer-Verlag, Berlin/Heidelberg/New York. plementation in small and medium sized enterprises. J. Enterp. Inf. Manag. 27 (4),
Jagoda, K., Samaranayake, P., 2017. An integrated framework for ERP system implemen- 358–384.
tation. Int. J. Account. Inform. Manag. 25 (1), 91–109. Yeh, C.H., Xu, Y., 2013. Managing critical success strategies for an enterprise resource
Kirmizi, M., Kocaoglu, B., 2022. The influencing factors of enterprise resource planning planning project. Eur. J. Oper. Res. 230 (3), 604–614.
(ERP) readiness stage on enterprise resource planning project success: a project man- Yoon, C., 2009. The effects of organisational citizenship behaviors on ERP system success.
ager’s perspective. Kybernetes. Comput. Hum. Behav. 25 (2), 421–428.
Konior, J., Szóstak, M., 2020. The S-curve as a tool for planning and controlling of con- Zareravasan, A., Mansouri, T., 2016. A dynamic ERP critical failure factors modelling
struction process—case study. Appl. Sci. 10 (6), 2071. with FCM throughout project lifecycle phases. Prod. Plan. Control 27 (2), 65–82.
Liao, H., Zhou, Z., Zhao, X., Zhang, L., Mumtaz, S., Jolfaei, A., Ahmed, S.H., Bashir, Zhou, Z., Dong, M., Ota, K., Wang, G., Yang, L.T., 2016. Energy-efficient resource allo-
A.K., 2020. Learning-based context-aware resource allocation for edge-computing- cation for D2D communications underlaying cloud-RAN-based LTE-a networks. IEEE
empowered industrial IoT. IEEE Int. Things J. 7 (5), 4260–4277. Int. Things J. 3 (3), 428–438.
Lima, E.P.D., Costa, S.E.G.D., Angelis, J.J., Munik, J., 2013. Performance measurement
systems: a consensual analysis of their roles. Int. J. Prod. Econ. 146 (2), 524–542.

15