Content uploaded by Eric K.H. Leung
Author content
All content in this area was uploaded by Eric K.H. Leung on May 15, 2019
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tprs20
International Journal of Production Research
ISSN: 0020-7543 (Print) 1366-588X (Online) Journal homepage: https://www.tandfonline.com/loi/tprs20
A B2B flexible pricing decision support system for
managing the request for quotation process under
e-commerce business environment
K.H. Leung, C.C. Luk, K.L. Choy, H.Y. Lam & Carman K.M. Lee
To cite this article: K.H. Leung, C.C. Luk, K.L. Choy, H.Y. Lam & Carman K.M. Lee (2019): A B2B
flexible pricing decision support system for managing the request for quotation process under e-
commerce business environment, International Journal of Production Research
To link to this article: https://doi.org/10.1080/00207543.2019.1566674
Published online: 22 Jan 2019.
Submit your article to this journal
View Crossmark data
International Journal of Production Research, 2019
https://doi.org/10.1080/00207543.2019.1566674
A B2B flexible pricing decision support system for managing the request for quotation process
under e-commerce business environment
K.H. Leung, C.C. Luk, K.L. Choy∗, H.Y. Lam and Carman K.M. Lee
Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
(Received 29 September 2018; accepted 22 December 2018)
In the era of digitalisation, e-commerce retail sites have become decisive channels for reaching millions of potential cus-
tomers worldwide. Digital marketing strategies are formulated by the marketing teams in order to increase the traffic on
their e-commerce sites, thereby boosting the sales of the products. With the massive amount of data available from the
cloud, which were conventionally made with a high degree of intuition based on decision makers’ knowledge and experi-
ence, can now be supported with the application of artificial intelligence techniques. This paper introduces a novel approach
in applying the fuzzy association rule mining approach and the fuzzy logic technique, for discovering the factors influencing
the pricing decision of products launched in e-commerce retail site, and in formulating flexible, dynamic pricing strategies
for each product launched in an e-commerce site. A pricing decision support system for B2B e-commerce retail businesses,
namely Smart-Quo, is developed and implemented in a Hong Kong-based B2B e-commerce retail company. A six-month
pilot run reveals a significant improvement in terms of the efficiency and effectiveness in making pricing decisions on each
product. The case study demonstrates the feasibility and potential benefits of applying artificial intelligence techniques in
marketing management in today’s digital age.
Keywords: data mining; artificial intelligence; market intelligence; fuzzy association rule mining; decision support system;
e-commerce
1. Introduction
The traditional business-to-business (B2B) Request For Quotation (RFQ) process prerequisites the order specifications by
the buyers. These specifications include, but are not limited to, the type of product or service, the quality and quantity
demanded, the delivery terms, and the payment terms. Upon receiving an RFQ from a buyer, the supplier then undergoes
a series of decision-making activities, such as determining whether the specified requirements of quality and quantity of
the requested product or service can be fulfilled, and the price to be offered to the buyer. Once the supplier responds, the
buyer compares the received offers proposed by multiple suppliers and decides which supplier should win the contract. This
conventional and standard RFQ process is time-consuming and costly (Teich et al. 2004), and consequently prolongs the
amount of time needed by the supplier in handling the subsequent business activities on receiving the quotation request. The
problem of inefficiency in the RFQ process among buyers and suppliers under the B2B business environment is attributed
to the number of complex decisions to be made by both parties, which, very often, requires a considerable amount of time
for decision-makers to manually make a decision based on their past experience and expertise. While human knowledge and
past experience play an inevitable role in the decision-making activities under the conventional RFQ management process,
human knowledge and experience can only be shared through internal meetings and discussions among the decision-makers.
Without any repository of knowledge, the RFQ cycle time cannot be reduced.
On the supplier side, deciding the price of the requested product or service is one of the most crucial decisions to be
made under the RFQ management process. Such decision hampers not only the chance of winning the order, but also the
profit and loss (P&L) of the company. Adjusting to an appropriate price so as to remain competitive while ensuring the
B2B order is profitable requires a series of considerations and tradeoffs. The pricing decision-making process becomes even
more complex under today’s B2B business environment, as more factors can be involved in making a pricing decision in the
B2B e-commerce business environment, such as the buyers’ purchasing behaviour, purchasing frequency and credibility.
In the era of digitalisation, e-commerce retail sites have become decisive channels to reach millions of potential customers
worldwide. This phenomenon not only applies in the business-to-customer (B2C) environment, where individual customers
source and purchase their favourite items via online shopping sites, such as Amazon and Alibaba’s Tmall, but also in the
*Corresponding author. Email: kl.choy@polyu.edu.hk
© 2019 Informa UK Limited, trading as Taylor & Francis Group
2K.H. Leung et al.
Figure 1. The adoption of Smart-Quo for providing pricing decision support under the RFQ process.
B2B environment, where buyers can reach suppliers through the B2B e-commerce marketplace, requesting a quotation by
adding the requested products into the shopping carts of the e-commerce websites, and then simply clicking a ‘Request for
quotation’ button to request the suppliers to provide a quotation accordingly. With the massive amount of data available from
the cloud under the B2B e-commerce business environment, the click rate of each product’s web page, and the consumers
buying patterns, can be extracted for supporting decision-making through the application of artificial intelligence (AI)
techniques so as to mine essential, hidden relationships and knowledge.
The pricing of products and services sold over Internet channels is becoming more dynamic. A typical example is the
fluctuation of the prices of air tickets and hotel rooms. In part, this is due to the increasing use of auction models in business
and consumer markets for selling commodities, excess inventories, and used merchandise (Kannan and Kopalle 2001). The
RFQ process is one of the most common operations carried out by every purchasing and supply manager on a daily basis so
as to obtain orders by offering competitive pricing, and it is a fact that such pricing involves decisive decision making that
needs careful consideration. This paper proposes and integrates a fuzzy association rule mining approach, for developing
a pricing decision support system to assist a supplier in making a wise pricing decision for a B2B e-commerce order, as
illustrated in Figure 1. The system, namely, Smart-Quo, is developed in an attempt to alleviate the following problems that
have existed for years in the actual business environment:
(1) Subjective and Unsystematic pricing decision-making process – In the B2B marketplace, the sellers or the suppliers
provide a quotation taking into consideration a number of decisive factors, such as customer’s background, their
purchasing spending and purchasing record. Without any decision support and any identification of the relation-
ships among the factors influencing consumer behaviour, the decision-making process of price quotation will be
inaccurate.
(2) Time-consuming in the pricing decision-making process – Timely responses from the supplier side is one of the
key factors in winning an order. However, with a lack of decision support tools, it is difficult for decision-makers
to solely rely on their knowledge and experience in deciding an appropriate price within a short period of time.
Consequently, any delay in providing a prompt feedback from the supplier could potentially reduce the chance of
getting the B2B order.
The rest of this paper is organised as follows. Section 2 presents a literature review of the existing research and theories
related to the topic. The proposed fuzzy association rule mining-based system for providing pricing decision support is
presented in Section 3, followed by a case study to demonstrate the proposed data mining approach in Section 4. The results
and findings are discussed in Section 5. Finally, Section 6 gives the conclusions of the study and directions of future research.
2. Literature review
Conventionally, business transactions are made through face-to-face meetings, phone calls, and emails. The advancement of
information technology (IT) enables enterprises to place orders via the Internet. Such contactless transactions made via the
Internet have become a major sales channel for suppliers. In 2017, the gross merchandise volume of business-to-business
International Journal of Production Research 3
e-commerce transactions amounted to 7.66 trillion US dollars, up from 5.83 trillion US dollars in 2013 (Statista 2017).
While the rise of B2B online retail stores allows more potential business opportunities due to the increase of brand exposure
of suppliers to the potential customers worldwide, suppliers are forced to improve their efficiency in the RFQ process
(Hvam, Pape, and Nielsen 2006; García-Crespo et al. 2011). Given the ease for buyers in searching for a supplier and a
subsequent request for a quotation, the timeliness of suppliers in providing feedback for each buyer upon receiving their
quotation requests is crucial in winning the order. Otherwise, buyers might end up selecting a supplier who provides the
most competitive prices and the fastest response to a quotation request. In view of the need to shorten suppliers’ response
time in providing feedback for a quotation request, the pricing decision becomes a critical area the suppliers should focus
on, not only for the sake of accuracy, but also the speed of making such decisions. However, in the era of digitalisation, B2B
e-commerce business companies are currently grappling with the complex task of determining the right price to charge a
customer for a product or a service (Bichler et al. 2002). As such, the difficulty of making fast and accurate pricing decisions
has been increasing.
Data mining and artificial intelligence techniques have been extensively applied in the past decades for solving different
types of research and industry problems. With the advancement of the Internet, big data analytics is one of the hottest
trends that companies attempt to adopt for assisting their decision-making (Albright, Winston, and Zappe 2010; Minelli,
Chambers, and Dhiraj 2012; Waller and Fawcett 2013). The integration of data mining tools and big data analytics for the
extraction of implicit and potentially useful knowledge from large sets of raw data from the cloud has a vast potential in the
future. In the literature, there are some hybrid approaches, combining two or more data mining and artificial intelligence
techniques to build a more comprehensive model that complements the drawbacks of solely applying a single technique
(Liao, Chu, and Hsiao 2012). Fuzzy association rule mining is one of the hybrid approaches that have been adopted for
discovering knowledge at the parameter level through defining the quantitative values of the parameters in fuzzy terms
(Chen and Wei 2002). The associations or relationships among the parameters identified using this approach can help in
decision making for the solution of a given problem (Sánchez et al. 2009; Osman Hegazi and Abugroon 2016). The fact that
the classical association rule mining approach, in which Boolean Logic is adopted to convert numerical characteristics into
boolean attributes by the sharp partitioning of the datasets, is computationally unproductive in terms of accuracy, processing
time, prevention of redundant rule generation, resource requirement etc. (Aliev et al. 2007;Hoetal.2012), has motivated
researchers to introduce the fuzziness of parameters into the traditional association rule mining approach, which has been
proven to be able to extract more useful knowledge for generating decision support than many other approaches (Mabu
et al. 2011; Lee et al. 2015). In the mainstream literature, it has been applied to the transportation sector (Sowan et al. 2013),
fashion product development (Lee et al. 2015), stock market performance prediction (Ho et al. 2012), education industry
(Verma, Thakur, and Jaloree 2017) and the medical industry (Buczak et al. 2015). For instance, a prediction model proposed
by Sowan et al. (2013) integrates the fuzzy association rule mining approach for knowledge extraction. The extracted
knowledge is then stored in a fuzzy inference system that can be used for the prediction of a future value. The experimental
results using road traffic data sets demonstrated the merits and capability of integrating the fuzzy association rule mining
algorithm and the fuzzy logic technique for, respectively, mining hidden relationships within the data sets, and providing
decision support based on accurate prediction.
In order to drive company-wide strategic and competitive advantages, actionable conclusions should be made based on
quality data. In the literature, a number of studies attempted to improve the overall supply chain performance. Jain, Beny-
oucef, and Deshmukh (2008) proposed a fuzzy association rule mining approach to support decision makers by enhancing
the flexibility in making decisions for evaluating agility with both tangible and intangible attributes such as flexibility, prof-
itability, quality, innovativeness, pro-activity, speed of response, cost and robustness. Ho et al. (2008) developed a set of
nearly optimal fuzzy rules in quality enhancement based on the extracted fuzzy association rules in a supply chain network.
Li et al. (2012) developed a model for selecting third-party logistics service providers (3PLs) using fuzzy sets. Chen, Hao,
and Li (2014) developed supply chain models that incorporate both a loss-averse retailer and option contracts, providing
insights into the effect of loss-averse on decision making and performance of the supply chain. A social media competitive
analysis performed by He, Zha, and Li (2013) suggested that the adoption of social media and text mining has become
an effective technique to extract business value from the vast amount of available social media data. Nevertheless, while
numerous applications of big data analytics using data mining approaches, machine learning algorithms, and artificial intel-
ligence techniques have been developed across various vertical industries, such as the telecom and marketing industries,
there is a lack of development of data mining applications, especially for tackling the internal decision-making inefficien-
cies under the RFQ process. Most of the previous literature focuses on proposing optimisation models for formulating a
proper bidding strategy.
According to García-Crespo et al. (2011), the conventional and knowledge-based systems for machining price quotation
can be classified into either qualitative estimation methods or quantitative estimation methods. Qualitative estimation meth-
ods are those based on the products developed previously for determining the price of a new product (Niazi et al. 2006),
4K.H. Leung et al.
Figure 2. Price/cost estimation methods suggested by García-Crespo et al. (2011) and Niazi et al. (2006).
which can be further separated into two methods, i.e. intuitive methods and analogical methods (Niazi et al. 2006). Quanti-
tative estimation methods are those based on the design, features, and manufacturing process of the product for estimating
the total cost of the product. These methods can further be classified into parametric methods and analytical methods. Exam-
ples of qualitative and quantitative estimation methods suggested by García-Crespo et al. (2011) and Niazi et al. (2006)are
illustrated in Figure 2.
Though a number of cost estimation methods were proposed and developed in the literature, the primary idea of the
development of these methods was to estimate the cost or the price of a product based on the costs incurred in manufacturing
and/or assembling the parts to produce the finished product. Other relevant factors in adjusting the profit margin of an order,
such as the buyers’ purchasing behaviour and the product attractiveness, have not been fully taken into consideration. In fact,
the details of the mentioned factors, i.e. the buyers’ purchasing behaviour and the product attractiveness, can be obtained
through the cloud, given that B2B transactions are now made via online retail websites. For example, the popularity of a
product can be measured in terms of the average click rate of the web page of a product in a specified period of time. The
degree of intention of a buyer to purchase a product can be measured by the amount of time a buyer stays on the web page,
and the number of times the buyers revisits the same product web page. Given the massive amount of information or data
available from the B2B retail websites in today’s digital economy, the price of a product to be offered to a customer can be
more dynamic and flexible. However, it is understandable that, without any IT and decision support, frequent adjustment
of the price of a product is difficult or even infeasible. In the literature, there has been a lack of focus on examining the
potential of applying artificial intelligence and data mining techniques for assisting or re-engineering business activities that
have long been undertaken without the aid of any IT solutions. The decision-making activities under the RFQ process is
one of those being executed with heavy reliance on human intervention and domain knowledge, such that the decisions
made under the RFQ process are subjective and lacking in quality. To address the research gap in the literature, this paper
highlights the fuzzy association rule mining approach as a hybrid model combining two artificial intelligence techniques to
alleviate the suppliers’ operating obstacles under the RFQ process in today’s e-commerce business environment.
3. Architecture of Smart-Quo pricing decision support system
In the perspective of a supplier, there are some crucial decision areas involved in the RFQ process upon receiving a cus-
tomer’s request. As shown in Figure 1, these decision areas include, but not limited to, adjusting the quotation price for
International Journal of Production Research 5
a buyer, designing the customised terms and conditions, determining the ability to meet the requirements, such as deliv-
ery requirement and production specifications, etc. Amongst these decision areas, pricing is undeniably one of the most
influential factors affecting the selection of a supplier by the customers. It suggests that a proper pricing adjustment for
each individual customer is crucial in the perspective of a supplier. To facilitate suppliers in the pricing determination pro-
cess, especially in today’s e-commerce-based business environment, where B2B orders can be received in e-marketplace,
a pricing decision support system, namely Smart-Quo, is designed, so as to extract hidden relationships between customer
purchasing behaviour, product popularity, and historical quoted prices of the product. The architecture of the Smart-Quo is
shown in Figure 3. Through the identification of their possible inter-related relationships, Smart-Quo provides a suggested
quotation price for each product upon being requested by a customer. For example, if a buyer makes a quotation request
with five different products involved, the Smart-Quo enables the supplier to extract the past purchasing behaviour of this
customer from the cloud database (if any), information on the five products requested by this customer, and then provides the
suggested quotation price of the five products upon running the fuzzy logic module of Smart-Quo. The adoption of such a
framework allows a supplier to efficiently and effectively to not only comprehensively take all relevant factors into account
during the pricing decision-making process, but also speed up the RFQ process by providing proper decision support for
suppliers to make pricing timely and accurate decisions upon receiving each quotation request from the B2B e-commerce
online selling platforms. The architecture of the Smart-Quo consists of three modules, namely, (i) Data storage and retrieval
module, (ii) Data mining and relationship extraction module, and (iii) Pricing decision support module.
3.1. Data storage and retrieval module
Historical purchasing behaviour of repeat customers, i.e. buyers who, in the past, have made quotation requests and pur-
chases via the B2B e-commerce online platforms, and product popularity and pricing information are retrieved and placed
in the Data storage and retrieval module of the Smart-Quo. The information can be retrieved from the B2B e-commerce
online platforms, and the internal Customer Relationship Management (CRM) system of the supplier. The information for
each customer and product retrieved and stored in this module serves as the inputs for the Data mining and relationship
extraction module, so that relationships between customer purchasing behaviour, product popularity and historical pricing
of the products, can be extracted in the Data mining and relationship extraction module.
3.2. Data mining and relationship extraction module
To identify the hidden relationships in the purchasing behaviour of a particular customer, and the product popularity in
the past, processed data from the Data storage and retrieval module is extracted to this module. The data mining algorithm
adopted in this system integrates fuzzy sets theory and association rules theory, which had been proposed and used in several
previous studies, i.e. Hong, Lin, and Wang (2003), Lau et al. (2009), Ho et al. (2012), and Lee et al. (2015). In general, the
use of the fuzzy association rule mining approach can be classified into three stages, namely, (i) parameter setting stage, (ii)
relationship extraction stage, (iii) rule evaluation and selection stage. In the application of the fuzzy association rule mining
algorithm for extracting hidden relationships, parameters, as well as the fuzzy characteristics of each parameter, are defined
and stored in this module. As these parameter settings are crucial to the subsequent data mining process, it is recommended
to consult industry experts for defining the fuzzy membership sets and threshold values, such as the confidence values, in
accepting the possible association rules. The detailed procedures of applying the fuzzy association rule mining algorithm in
generating pricing decision support for suppliers during the RFQ process is depicted in the Case study section.
3.3. Pricing decision support module
The relationships identified using the fuzzy association rule mining approach in the previous module are coded into ‘IF-
Then’ rules. These rules are stored in the inference engine of the fuzzy logic sub-system of the Smart-Quo in this module.
The objective of including a fuzzy logic sub-system in the Smart-Quo is to provide pricing decision support to suppliers,
so that they can flexibly adjust the quotation prices for each customer, while taking all relevant factors, such as customer
purchasing behaviour and customer’s acceptance rate of quotations, into consideration. Therefore, for every buyer making
an RFQ, the supplier can retrieve relevant information from the cloud database in the Data storage and retrieval module,
generate and confirm the ‘If-Then’ rules in the Data mining and relationship extraction module, and then obtain the output of
the Smart-Quo, i.e. a suggested quotation price of all the products requested by the buyer. The supplier can thus determine
the final quotation price with reference to the suggested one offered by Smart-Quo, thereby reducing the reliance solely
on the decision maker while shortening the pricing decision making process and ensuring a quotation is submitted to the
designated buyer in a timely manner.
6K.H. Leung et al.
Figure 3. System architecture of the Smart-Quo.
4. Case study
The proposed fuzzy association rule mining approach for supporting a supplier’s pricing decision-making under the RFQ
process is validated through a case study. The Smart-Quo is implemented into a small and medium-sized (SME) B2B
e-commerce company founded in 2014. It is a Hong Kong-based wholesaler specialised in re-selling two lines of products,
International Journal of Production Research 7
i.e. sports equipment and electronic goods. The company offers a variety of products and provides worldwide shipping
with guaranteed delivery promises. The free trade policies, proximities to other worldwide cities, free flow of capital and
information, and business flexibility, contributed to the cross border e-commerce market growth in Hong Kong in the past
decade. To capture both local and international markets, like many retail giants in Hong Kong, in which traditional brick-
and-mortar stores established online purchasing platforms, the company launched its own B2B e-commerce online retail site
as the major sales channel. While the availability of big data increases the likelihood of upselling opportunities by enabling
e-retailers to track consumers’ online shopping preferences, the company has been facing serious challenges in its B2B
resell activities in its online retail site. One of which is the complex pricing management process, comprising of a series
of an internal operating procedures that require participation of management teams from sales, marketing, and customer
service department, for preparing a customised offer/quotation for each customer. Regular meetings are held among the
departmental heads, on a weekly basis, to:
(1) Process the quotation requests received in the previous week, and
(2) Customise the quotation for each customer based on previous sales performance with respect to the customer.
With the increasing number of competitors in the e-marketplace, the competitive business environment has forced the
company to provide more accurate and speedy response to each customer’s request. The company has been struggling to
maintain the service level for all potential customers by returning a quotation to each customer on time. The company
realised that weekly meetings are no longer efficient enough to provide speedy response to customer’s quotation requests.
The delay in providing timely feedback due to the inefficiency of making pricing decisions during the RFQ process has
become the bottleneck of the company in further expanding the business. Therefore, Smart-Quo is implemented in the case
company in an attempt to improve the efficiency in the RFQ operation, and reduce the heavy reliance of decision makers
to determine a proper pricing through time-wasting meetings. The core feature of Smart-Quo, that is, data mining and
relationship extraction using the fuzzy association rule mining approach, is discussed below. An overview of the Smart-Quo
working mechanism, is illustrated in Figure 4(a).
4.1. Parameter setting stage
In this stage, parameters are identified through a series of discussions and consultations with the industry experts, followed
by defining the fuzzy characteristics of each parameter. In this case study, a total of 13 parameters are identified, as shown
in Table 1. Data are retrieved from various sources, such as from the internal CRM system and the back-end cloud database
Figure 4. (a) Working mechanism of the Smart-Quo. (b) Membership functions of all pre-defined parameters.
8K.H. Leung et al.
Figure 4. Continued.
International Journal of Production Research 9
Figure 4. Continued.
of the B2B e-commerce online retail site. These parameters can be categorised into three types of data: (i) Customer profile
and purchasing behaviour, (ii) Product information, and (iii) Customer transaction records.
(1) Customer profile and purchasing behaviour – Customer profiles, with basic information, such as the customer ID,
company name, in charge staff name, contact number, and foundation year, are retrieved and stored in the centralised
cloud database in the Data storage and retrieval module of the Smart-Quo. Parameters under this category, i.e.
parameters A, C, D, E, F, G, J, and L, are also gathered and stored for further data mining purpose.
(2) Product information – The details of all listed products in the B2B e-commerce retail sites are retrieved and stored in
the centralised cloud database in the Data storage and retrieval module of the Smart-Quo. In particular, the historical
sales performance of each product (Parameter B), and the average monthly hit rate of each product (Parameter
K), that is, the number of sales of a product compared to the number of people who visit the specific product
webpage to look at that product, are identified as parameters that are taken into consideration for the extraction of
the relationship with other parameters.
(3) Transaction records – Historical transactions are stored in the cloud database of the Smart-Quo. Relevant data,
which includes the transaction ID, customer ID, order date, delivery date, payment date, product ID, unit of products
10 K.H. Leung et al.
sold, order’s total price, average number of SKUs in an order, and the total value of transported goods, are retrieved
along with parameters H, I, and M, which are, respectively, the number of SKUs ordered along with this product
for a particular customer, the expected ordering quantity of this product for a particular customer, and the discount
of this product for a particular customer.
Table 1. The details and fuzzy sets of each defined parameter.
Parameter (symbol) Type Fuzzy sets
Inputs:
Customer’s credibility (A) Customer A={(x,μA(x)|x∈[0, 100]}
μA(x)=
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
VL n(x)≤30
L25 ≤n(x)≤50
M40 ≤n(x)≤70
H60 ≤n(x)≤90
VH 85 ≤n(x)≤100
Last year’s sales record of the
same month of a particular
product (B)
Product B={(x,μB(x)|x∈[0, 40]}
μB(x)=⎧
⎪
⎨
⎪
⎩
Ln(x)≤20
M15 ≤n(x)≤30
H25 ≤n(x)≤40
Average hit rate per month of
all products of a particular
customer (C)
Customer C={(x,μC(x)|x∈[0, 45]}
μC(x)=⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
Ln(x)≤15
M10 ≤n(x)≤25
H20 ≤n(x)≤35
VH 30 ≤n(x)≤45
Ordering frequency per year
of a particular product by a
customer (D)
Customer D={(x,μD(x)|x∈[0, 2]}
μD(x)=⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
Ln(x)≤0.6
M0.5 ≤n(x)≤1.2
H1≤n(x)≤1.6
VH 1.5 ≤n(x)≤2
Total number of orders made by
a particular customer per year
(E)
Customer E={(x,μE(x)|x∈[0, 80]}
μE(x)=⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
Ln(x)≤20
M15 ≤n(x)≤40
H30 ≤n(x)≤60
VH 50 ≤n(x)≤80
Number of orders of this product
made by a particular customer
per year (F)
Customer F={(x,μF(x)|x∈[0, 40]}
μF(x)=⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
Ln(x)≤8
M5≤n(x)≤20
H15 ≤n(x)≤30
VH 25 ≤n(x)≤40
Average delivery lead time of
the orders requested by this
customer (G)
Customer G={(x,μG(x)|x∈[0, 30]}
μG(x)=⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
Ln(x)≤5
M3≤n(x)≤15
H12 ≤n(x)≤20
VH 18 ≤n(x)≤30
Number of SKUs ordered
along with this product for a
particular customer (H)
Transaction H={(x,μH(x)|x∈[0,20]}
μH(x)=⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
Ln(x)≤3
M2≤n(x)≤7
H5≤n(x)≤10
VH 8≤n(x)≤20
(Continued)
International Journal of Production Research 11
Table 1. Continued.
Parameter (symbol) Type Fuzzy sets
Outputs:
Current ordering quantity of
this product for a particular
customer (I)
Transaction I={(x,μI(x)|x∈[0,50]}
μI(x)=
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
VL n(x)≤10
L8≤n(x)≤15
M12 ≤n(x)≤25
H22 ≤n(x)≤36
VH 32 ≤n(x)≤50
Percentage of this customer
accepting the first quotation
(J)
Customer J={(x,μJ(x)|x∈[0, 100]}
μJ(x)=
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
VL n(x)≤30
L25 ≤n(x)≤50
M40 ≤n(x)≤70
H60 ≤n(x)≤90
VH 85 ≤n(x)≤100
Average hit rate of this product
per month for all customers
(K)
Product K={(x,μK(x)|x∈[0, 60]}
μK(x)=
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
VL n(x)≤15
L10 ≤n(x)≤25
M20 ≤n(x)≤35
H30 ≤n(x)≤45
VH 40 ≤n(x)≤60
Average hit rate of this product
per month for a particular
customer (L)
Customer L={(x,μL(x)|x∈[0, 20]}
μL(x)=⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
Ln(x)≤5
M3≤n(x)≤8
H7≤n(x)≤12
VH 10 ≤n(x)≤20
Discount of this product (M) Transaction M={(x,μM(x)|x∈[0, 100]}
μM(x)=
⎧
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎩
VL n(x)≤20
L15 ≤n(x)≤40
M35 ≤n(x)≤60
H55 ≤n(x)≤80
VH 75 ≤n(x)≤100
With the parameters identified, the fuzzy characteristics of each parameter, i.e. the fuzzy terms and fuzzy membership
functions, are defined. While there is no clear indication of the range and type of the membership function of each parameter,
users of the proposed system are recommended to evaluate and define the fuzzy characteristics based on the operating scale
of the company and the opinions of the decision-makers who are used to regularly prepare quotations and in determining
the quotation prices. With respect to the operating scale of the case company, the membership functions of each parameter
are defined and illustrated in Figure 4(b) and Table 1.
4.2. Relationship extraction stage, and the rule evaluation and selection stage
With the parameters identified and their fuzzy characteristics defined, the fuzzy association rule mining approach is used to
extracted any hidden relationships among the parameters identified. Through the use of this algorithm, ‘If-Then’ rules can
be generated and stored in the Pricing decision support module of the Smart-Quo for providing the operators with decision
support for determining the quotation price of each item requested by a customer. To illustrate the feasibility of the fuzzy
association rule mining approach for mining the relationships among the parameters under the three data categories, namely,
customer purchasing behaviour, product popularity, and transaction records, a case scenario, with five historical transaction
records of a particular product, is given in this section, as shown in Table 2. A list of the notations used is shown in Table 3
for effectively expressing the computational steps of the algorithm. As the algorithm integrates data mining techniques and
fuzzy set concepts to build useful association rules from quantitative data, the confidence value and threshold support count
need to be predefined by domain experts. The minimum support counts of each parameter are predefined and illustrated in
12 K.H. Leung et al.
Table 2. Five sample transaction records of the case company and the threshold support counts of each parameter.
Order ID A B C D E F G H I J K L M
001 88 18 15 1.2 50 25 14 7 35 80 42 9 30
002 80 30 8 0.8 30 35 7 1 21 95 20 12 10
003 75 20 25 0.5 20 5 7 5 15 70 25 15 12
004 95 10 12 0.2 15 10 20 0 20 75 20 10 20
005 90 12 15 0.7 40 30 14 2 20 85 15 7 25
Minimum support count 0.3 0.45 0.6 0.6 0.8 0.7 0.6 0.75 0.5 0.7 0.3 0.65 0.7
Table 3. List of notations used in the Smart-Quo FARM algorithm.
N Total number of historical order records
N={1, 2, ... ,n}The set of index of historical order records
RaThe ath order record, ∀a∈N
QtThe tth parameters
htThe number of fuzzy classes for Qt
Ht={1, 2, ... ,ht}The set of index of fuzzy classes for Qt
Cti The ith fuzzy classes of Qt,∀i∈Ht
αtThe predefined minimum support threshold of Qt
βxt The fuzzy set converted from the quantitative values of Qt
βxti The fuzzy membership values of Qtin Rain class Cti
countti The summation of βxti values
MAX-counttThe maximum value among countti
MAX-CtThe fuzzy class of Qtwith MAX-countt
IrThe set of itemsets with ritems
λThe predefined minimum confidence threshold
Table 2, and the threshold confidence value in the case scenario is set as 75%. Also, the fuzzy sets need to be converted and
kept for use in a sequence of the data mining process, such as generation of itemsets and calculation of support counts.
The algorithm description for generation of the itemsets and the calculation of the supports counts for the generation of
possible association rules are as follows.
Step 1. The quantitative values of parameters Qtin each order record Rainto fuzzy set βxt need to be converted
according to the predefined membership functions.Represent βxt as (βxt1/Ct1+βxt 2/Ct2+βxt3/Ct3+ ··· + βxti/
Cti).
Step 2. Calculate the support count countti of each Cti of Qt.
Step 3. Pick the maximum values of support counts of each parameter, MAX-countt, and find the corresponding
fuzzy class MAX-Ctfor the fuzzy characteristic of Qtfor the further mining process.
Step 4. Put the parameters into Iras items and set r=1. If the MAX-counttis larger than or equal to αt, keep it in
Ir. Otherwise, remove it from Ir.
Step 5. Generate combinations of items in Irby forming (r+1)-itemsets. For each itemset wwith items (w1,
w2,... ,wr+1), identify the maximum value of threshold support counts of items as αw. If the minimum value of
support counts of the items is larger than or equal to αw, temporarily put w in Ir+1.
Step 6. Check whether Ir+1is null or not. If not, go to the Step 7. If Ir+1is null and ris equal to 1, quit the algorithm.
If Ir+1is null but ris greater than 1, go to Step 11.
Step 7. Calculate the fuzzy membership value βxw of win Raas βxw =min (βxw1,βxw2,βxw3,... ,βxwr+1).
Step 8. Calculate the support count of w.
Step 9. If countwis equal to or larger than αw, keep win Ir+1. Otherwise, remove it from Ir+1.
Step 10. Check if Ir+1is null. If not, set r=r+1 and repeat Steps 5–10.IfIr+1is null and r=1, quit the algorithm.
If Ir+1is null but ris greater than 1, go to Step 11.
Step 11. Extract items from Irto construct possible rules for the following stage.
Step 12. Calculate the confidence value of each rule. If the confidence value of a rule is larger than or equal to the
predefined threshold confidence value, the rule is regarded as a useful fuzzy association rule.
Upon the completion of the above steps, the possible rules and all calculation data should be retained for the further develop-
ment of the algorithm of the Smart-Quo. Generated rules are evaluated and selected under the rule evaluation and selection
stage (step 12), the third stage of the algorithm under the Data mining and relationship extraction module of the Smart-Quo.
International Journal of Production Research 13
In the rule evaluation and selection stage, threshold confidence values are defined for calculating the confidence value of
rules. If the confidence value of rules does not satisfy the threshold confidence values, the rules will be discarded in order
to implement rule filtering for enhancing the Smart-Quo accuracy and preciseness. The detailed computational procedures
of this algorithm based on the five transaction records in Table 2are illustrated as follows.
Step 1. The quantitative values of all parameters need to be converted into fuzzy sets according to the predefined
membership functions. For example, the first parameter (parameter A) in the order ID 001 has a rating of 88, which
lies in ‘High’ and ‘Very High’ fuzzy classes, based on its membership function defined in Figure 4(b). It can be
converted into a fuzzy set which is characterised as (0.33/High +0.67/Very High). By applying such a fuzzy set
conversion for all parameters and all order records, the result of converted fuzzy sets for five historical transaction
records is shown in Table 4.
Step 2. After converting all fuzzy sets, the counts of each fuzzy class are calculated. Add up all the counts of each
parameter for further calculation. Take parameter A as an example, by adding up the number of counts of ‘High’
fuzzy class from five historical transaction records, the fuzzy counts of ‘High’ for parameter A is calculated as
(0.33 +0.67 +1+0+0) =2. The remaining counts of each fuzzy class in each parameter are calculated using
the same computation. The detailed result is shown in Appendix 1.
Step 3. Sort out the corresponding fuzzy class and the maximum number of counts for each parameter. Take param-
eter A as an example, the counts of the five fuzzy classes of parameter A, i.e. ‘Very Low’, ‘Low’, ‘Medium’, ‘High’
and ‘Very High’ are 0, 0, 0, 2, 2.6, respectively. Hence, the class of ‘Very High’ is nominated for parameter A. The
results of the remaining parameters are shown in Table 5.
Step 4. Compare the maximum counts and corresponding fuzzy classes of the 1-itemset with the predefined thresh-
old support counts. If the maximum count of a parameter is smaller than the threshold support count, the parameter
needs to be removed from the 1-itemset. In this case example, all maximum counts of all parameters are larger than
all the predefined threshold support counts listed in Table 2. Thus, no parameter should be removed.
Step 5. Generate 2-itemset combinations from the maximum counts of the 1-itemset. When the minimum number
of counts of two parameters is equal to or larger than the maximum of their predefined threshold support counts,
the combination of 2-itemset is selected. For example, for the itemset {A.Very High, B.Low}, the counts of ‘Very
High’ for parameter A and ‘Low’ for parameter B are respectively 2.6 and 2. Thus, the minimum value is 2. As
for the threshold support counts of parameter A and B, their predefined values are 0.3 and 0.45 respectively. Thus,
the maximum threshold support count for the itemset {A.Very High, B.Low}is 0.45. Since the minimum number
Table 4. Converted fuzzy sets.
Order ID Quantitative values of parameters using fuzzy sets
001 0.33
A.H+0.67
A.VH 0.2
B.L+0.3
B.M0.63
C.M 1
D.H0.67
E.H0.63
F.H 0.2
G.M+0.5
G.H 0.8
H.H0.17
I.H+0.38
I.VH 0.67
J.H0.3
K.H+0.4
K.VH 0.67
L.H 1
M.L
002 0.67
A.H0.5
B.H 1
C.L 1
D.M 1
E.M 1
F.VH 0.57
G.M 1
H.L0.8
I.M 0.5
J.VH 1
K.VH 0.4
L.VH 1
M.VL
003 1
A.H0.5
B.M0.63
C.H0.33
D.L0.33
E.M0.75
F.L0.57
G.M 0.8
H.M0.38
I.M0.67
J.H 1
K.M 1
L.VH 0.8
M.VL
004 1
A.VH 1
B.L0.6
C.L+0.25
C.M 1
D.L0.5
E.L 0.5
F.M 0.5
G.VH 1
H.L 1
I.M 1
J.H0.71
K.L 1
L.H0.33
M.L
005 1
A.VH 0.8
B.L0.63
C.M0.67
D.M0.67
E.H 0.71
F.VH 0.2
G.M+0.5
G.H0.67
H.L 1
I.M0.33
J.H0.63
K.L0.5
L.M0.67
M.L
Note: VH =Very High, H=High, M=Medium, L=Low, VL =Very Low.
Table 5. The maximum counts and corresponding
fuzzy classes of the 1-itemset.
Parameter item Counts Parameter item Counts
A. Very High 2.6 H. Low 2.67
B. Low 2 I. Medium 3.18
C. Low 1.6 J. High 2.67
D. Medium 1.67 K. Very High 1.4
E. High 1.33 L. High 1.67
F. Very High 1.71 M. Low 2
G. Medium 1.54
14 K.H. Leung et al.
of counts is larger than the maximum of the threshold support counts, i.e. 2 >0.45, the 2-itemset {A.Very High,
B.Low}is accepted for putting into the 2-itemset. The same logic is applied to other combinations for justifying
whether the combination should be included in the 2-itemset. As the 2-itemset is not null, the algorithm can be
continued.
Step 6. After identifying the parameter item to be included in the 2-itemset, the fuzzy counts of the itemsets in
2-itemset needs to be calculated by adding up the minimum number of counts of each item in the five order records.
Considering {A.Very High, B.Low}again as an example, in order ID 001, the counts of A.Very High and B.Low
are 0.6 and 0.2, respectively, thus the minimum value is 0.2. Besides, {A.Very High, B.Low}does not exist in order
ID 002. Hence, the count of order ID 002 is equal to 0. By adding up the minimum values of all the order records,
the counts of {A.Very High, B.Low}is (0.2 +0+0+1+0.8) =2. All the 2-itemsets are calculated in the same
way.
Step 7. As in Step 4, compare the results of Step 5 and Step 6. If the count of the 2-itemset is smaller than the
threshold support count, the parameter needs to be removed from the 2-itemset. As the count of {A.Very High,
B.Low}is 2, which is larger than the maximum value of threshold support count, i.e. 0.45, {A.Very High, B.Low}
of the 2-itemset is retained. On the other hand, the count of {A.Very High, G.Medium}is 0.4, which is smaller
than the threshold support count (0.6), thus {A.Very High, G.Medium}needs to be removed from the 2-itemset.
After comparing all the counts of the 2-itemset with the threshold support count, the reviewed 2-itemset is shown
in Table 6.
Step 8. As the reviewed Iris not null, repeat Steps 5–7 to generate higher level itemsets, such as 3-itemset, until
there are no further combinations that can be formed. The reviewed 3-itemset and 4-itemset are calculated and
shown in Table 6.
Step 9. Only the f-itemset with f ≥2 can be extracted to conduct possible fuzzy association rules. The confidence
value of each rule is necessary to be calculated in order to justify whether the rule should be considered as valid.
For example, the 2-itemset {A.Very High, I.Medium}, the possible rule is constructed as ‘IF {A.Very High}then {I.
Medium}’. The confidence value of this rule is calculated as:
(A.Very high ∩I.Medium)
(A.Very High)=2
2.6 =0.77
Table 6. The reviewed 2-itemset, 3-itemset and 4-itemset.
Parameter item Counts Parameter item Counts Parameter item Counts
2-itemset
{A.VH, B.L}2{B.L, J.H}1.53 {F.VH, I.M}1.51
{A.VH, C.L}0.6 {B.L, L,H}1.2 {G.M, H.L}0.77
{A.VH, D.M}1.67 {B.L, M.L}1.2 {G.M, I.M}1.15
{A.VH, E.H}1.34 {C.L, D.M}1{G.M, J.H}0.97
{A.VH, F.VH}0.71 {C.L, F.VH}1{H.L, I.M}2.47
{A.VH, H.L}1.62 {C.L, H.L}1.6 {H.L, J.H}1.33
{A.VH, I.M}2{C.L, I.M}1.4 {H.L, L.H}1
{A.VH, J.H}1.9 {C.L, L.H}0.6 {H.L, M.L}1.67
{A.VH, K.VH}0.4 {D.M, F.VH}1.67 {I.M, J.H}1.71
{A.VH, L.H}1.67 {D.M, G.M}0.77 {I.M, K.VH}0.8
{A.VH, M.L}1.6 {D.M, H.L}1.67 {I.M, M.L}1
{B.L, C.L}0.6 {D.M, I.M}1.47 {J.H, L.H}1.67
{B.L, D.M}0.67 {E.H, J.H}1{J.H, M.L}1.33
{B.L, E.H}0.87 {E.H, L.H}0.67 {L.H, M.L}1
{B.L, F.VH}0.71 {E.H, M.L}1.34 {G.M, I.M }1.15
{B.L, H.L}1.67 {F.VH, G.M}0.77
{B.L, I.M}1.8 {F.VH, H.L}1.67
3-itemset
{A.VH, B.L, E.H}0.87 {B.L, F.VH, I.M}0.71 {D.M, F.VH, G.M}0.77
{A.VH, B.L, F.VH}0.71 {C.L, D.M, F.VH}1{D.M, F.VH, H.L}1.67
{A.VH, B.L, H.L}1.67 {C.L, D.M, H.L}1{D.M, G.M, H.L}0.77
{A.VH, F.VH, I.M}0.71 {C.L, F.VH, H.L}1{F.VH, G.M, H.L}0.77
4-itemset
{D.M, F.VH, G.M, H.L}0.77
Note: VH =Very High, H=High, M=Medium, L=Low, VL =Very Low.
International Journal of Production Research 15
Take {A.Very High, E.High, J.High}of the 3-itemset as an example. The rule can be constructed as ‘IF {A.Very
High, E.High}then {J.High}’. The confidence value of this rule is calculated as:
(A.Very high ∩E.High ∩J.High)
(A.Very High ∩E.High)=1
1.267 =0.79
Step 10. After calculating the confidence values of all possible rules, compare them with the predefined threshold
confidence value, which is defined as 0.75 in this case example. If the possible rules with confidence values are
smaller than the predefined threshold confidence value, the possible rules are removed. On the other hand, if the
possible rules with confidence values larger than or equal to the predefined threshold confidence value, the possible
rules are retained as useful association rules. The result of useful association rules is obtained, with confidence
values ≥0.75 needing to be calculated. The possible association rules with output parameters I, J, K, L and M
included in the consequent part of the ‘If-Then’ rule, i.e. the then-part of the rule, are generated and shown in
Table 7.
Step 11. After filtering the possible rules with confidence values ≥0.75, the fuzzy association rule mining algorithm
ends after obtaining a set of useful fuzzy association rules. The generated set of useful fuzzy association rules are
put into a fuzzy logic subsystem of the Smart-Quo. In the fuzzy logic subsystem, there is an inference engine storing
all the ‘If-Then’ rules for governing the relationship between each input and output parameter. The generated fuzzy
association rules with the consequent part of the rules containing the output parameters are extracted and stored
as decision rules in the inference engine of the fuzzy logic subsystem. The decision rules are paraphrased into
statements in order to help the user to understand the rules, as they may lack data mining knowledge. Some samples
of decision rules that are paraphrased into statements are shown in Table 8.
Step 12. After extracting the decision rules, the MATLAB Fuzzy Logic Toolbox is used to store the decision rules
and membership functions of all the parameters. Also, a fuzzy inference engine is used to identify output fuzzy sets
after the inputs are converted into fuzzy sets. The output fuzzy sets constructed as an individual fuzzy region for
each fired decision rule based on the membership functions. A consequent fuzzy region is acquired by integrating
all individual fuzzy regions of all the fired decision rules. The centre of area (Y) needs to be applied to translate
output fuzzy sets into quantitative values. The calculation for identifying the centre of area (Y) of the consequent
Table 7. Fuzzy association rules with confidence values ≥0.75.
Rule Confidence value Rule Confidence value
If {A. VH}then {I. M}2/2.6 =0.77 If {G. M}then {I. M}1.15/1.54 =0.75
If {A.VH}then {J. H}2.0/2.6 =0.77 If {M. L}then {J. H}1.33/1.67 =0.80
If {B. L}then {I. M}1.8/2 =0.9 If {E. H, J. H}then {M. L}1/1 =1
If {D. M}then {I. M}1.47/1.667 =0.88 If {A. VH, E.H}then {M. L}1.27/1.334 =0.95
If {H. L}then {I. M}2.47/2.667 =0.93 If {B. L, E. H}then {M. L}0.87/0.87 =1
If {A. VH, E. H}then {J. H}1/1.267 =0.79 If {B. L, J. H}then {M. L}1.2/1.53 =0.78
If {B. L, F. VH}then {I. M}0.7143/0.7143 =1If{H. L, J. H}then {M. L}1/1.3 =0.77
Note: VH =Very High, H=High, M=Medium, L=Low, VL =Very Low.
Table 8. Some samples of decision rules that are paraphrased into statements.
Rule 1
IF Customer’s credibility is High AND
Average hit rate per month of all products of a particular customer is Medium AND
Average delivery lead time of the orders requested by this customer is High AND
Number of SKUs ordered along with this product for a particular customer is Low
THEN Percentage of this customer accepting the first quotation is Very High
Discount of this product is Low
Rule 2
IF Customer’s credibility is Very High AND
Total number of orders made by a particular customer per year is High AND
Number of SKUs ordered along with this product for a particular customer is Medium
THEN Average hit rate per month of this product for all customers is Low AND
Discount of this product is Very Low
16 K.H. Leung et al.
fuzzy region is given by Equation (1).
Y=N
j=1wjCjAj
N
j=1wjAj,(1)
where w,Cand Aare the weight, centre of gravity and area of individual fuzzy regions of decision rule j, respec-
tively. The quantitative values of the output parameters are computed and displayed after the quantitative values of
the input parameters are inputted in MATLAB Fuzzy Logic Toolbox, as shown in Appendix 2.
4.3. User interfaces of the Smart-Quo
The front-end user interfaces (UI) of the Smart-Quo are designed and integrated with the back-end database. The parameters
affecting the pricing decision in the RFQ process can be modified in the ‘Parameter Setting’ UI of the Smart-Quo, as depicted
in Appendix 3. In order to continuously improve the suitability of the system, the fuzzy characteristics, i.e. fuzzy terms and
membership functions, can be edited. In the case of receiving a new quotation request from a potential buyer via the B2B
e-commerce retail platform, the staff of the case company can select this buyer and the requested products by searching and
selecting the customer ID and product ID in the ‘Generate new rules’ UI of the Smart-Quo, as shown in Appendix 3. By
clicking the ‘Generate specific rules for this customer’ button, the back-end fuzzy association rule mining algorithm will be
activated to generate fuzzy association rules which will be further retained as ‘If–Then rules’ in the fuzzy inference engine of
the Smart-Quo for generating a proper pricing decision for this customer. Once the rules are generated, the decision-maker
can obtain the pricing decision support in the ‘Generate suggested price’ UI of the Smart-Quo, as shown in Appendix 3.
By clicking the ‘Generate suggested price for this customer’ button, the fuzzy logic subsystem will be activated to generate
a proper discount rate for each of the products requested by the current customer. In the result output UI, the decision
maker can view and edit the discount of each product. The confirmed prices of each product requested by this customer in
Smart-Quo can then be used for further preparation of the quotation for the customer. With the adoption of the Smart-Quo,
the decision makers no longer manually decide the price of each product based on their experience. The decision support
generated in the proposed system using the fuzzy association rule mining and fuzzy logic approach has brought benefits to
the case company, which are discussed in the next section.
5. Results and discussion
5.1. Improvement areas with the introduction of the proposed system
During a six-month pilot run of the system in the case company, the generated fuzzy association rules were recursively
challenged and revised until better ones were obtained. The aim of implementing the Smart-Quo is to enhance the efficiency
and accuracy in making quotation pricing decisions under the RFQ process. The improvement areas after the implementation
of the Smart-Quo are threefold: (i) Reduction in the average time in quotation planning, (ii) Reduction in the number of staff
participating in the quotation process, and (iii) Acceptance rate of the quotations. A summary of the improvements achieved
by the use of the Smart-Quo is shown in Table 9.
(i) Reduction in the average time in quotation planning
With the adoption of Smart-Quo, the average time in quotation planning is reduced by over 60%, as the time spent on
human-driven activities such as manually reviewing data from different documents and records, as well as judging in the
importance of different pricing factors based on experience and intuition, are eliminated. Without the adoption of Smart-
Quo, operators are required to use the traditional way to handle the quotation in e-commerce retail business. For example,
the staff need to check and assemble the quotation orders from the e-commerce retail website on Day 1, and then send
the orders to the manager, who reviews the orders and decides on the price of the quotations by personal judgement. This
process normally needs at least two days as it is a completely manual process. Finally, the manager will inform the staff
who is responsible for the integration of the price and quotation items through an excel message, and then and send it back
to the customer. The traditional way normally needs at least four working days to handle the quotation orders. Instead, with
the introduction of the Smart-Quo, the decision maker only needs to monitor the operation of the Smart-Quo processing of
the quotation flow by activating the system to generate the pricing discount automatically, with the total number of days to
prepare a quotation reduced to 1.5 working days. Such a reduction in the waiting time for customers to receive quotation
feedback potentially increases the chance of winning the order from the competitors.
(ii) Reduction in the number of staff participated in the quotation process
Before the launch of Smart-Quo, on average, a total of seven staff were involved in managing the entire quotation process
in order to maintain the quality of customer relationships due to the large number of quotations required to be prepared for
International Journal of Production Research 17
Table 9. Improvements achieved by the use of the Smart-Quo.
Performance indicators Without the adoption of
Smart-Quo With the adoption of
Smart-Quo Percentage of
improvement
Average time spent in quotation planning (working days) 4 1.5 62.5%
Number of staff involved in the quotation process 7 3 57.1%
Acceptance rate of the quotations 46% 78% 69.6%
the buyers who are awaiting feedback from the supplier. With the implementation of Smart-Quo, the staff no longer need
to manually extract the required information from the company database and the e-commerce retail website. Therefore,
the manpower required in the quotation department has been drastically reduced. Only three staff are now needed in the
quotation process, allowing the management to adjust the manpower in the quotation department.
(iii) Acceptance rate of quotation
Subjective decisions made before the introduction of the proposed system led to a relatively low acceptance rate of quo-
tations submitted by the case company. After the use of the Smart-Quo, the acceptance rate increased from 46% to 78%,
indicating that B2B customers are more satisfied with the offered pricing, after the company has adopted the Smart-Quo for
generating pricing decision support. With the increased acceptance rate, the proposed system, in other words, increased the
success rate of winning a bid, bringing more sales orders in long term.
5.2. Comparison with other existing work
The Smart-Quo is compared with other existing approaches to demonstrate its advantage for providing pricing decision
support in the B2B e-commerce environment. In the literature, there is a lack of development of data mining and artificial
intelligence-based approaches to deal with the internal decision-making inefficiencies under the RFQ process. Most of the
previous literature focuses on proposing optimisation models for formulating a proper bidding or pricing strategy. In view of
the lack of an AI-based decision support model for assisting industrial practitioners in the RFQ process, existing approaches,
which apply hypothetical, optimisation and game-theoretic approaches are selected for comparison. A summary is presented
in Table 10.
(i) Comparing Smart-Quo with existing hypothetical approach
Statistical analysis with the construction of a hypothetical model is a common technique for identifying if there is any cor-
relation among various subjects. However, confirming the correlation among various factors that affect the pricing strategy
in the RFQ process is not sufficient to provide industry practitioners with pricing decision support. Abbey, Blackburn, and
Guide (2015) developed hypothesis testing and regression analysis to investigate the correlation between pricing discounts
and attractiveness of remanufactured products. Though the study confirmed that there was a relationship between these
factors, and was subsequently developed a mathematical model under different consumer and product segments to illustrate
the relationship, the work was not adequate in facilitating industry practitioners to determine a proper pricing for their prod-
ucts. Without further development of data mining or artificial intelligence technique, the pricing decision under the RFQ
process still relies heavily on industry practitioners’ knowledge and experience, leading to a time-consuming, subjective,
and unsystematic pricing decision-making process. Therefore, the fuzzy association rule mining approach introduced in this
study can extract the hidden relationship among various parameters, such as the customer purchasing behaviours and sales
performance of the product. The relationships among these parameters are further used for the development of the Smart-
Quo to generate pricing decision support whenever a quotation request is received by the company, so that proper pricing
for the quotation can be determined without heavy reliance of one’s knowledge and experience.
(ii) Comparing Smart-Quo with existing non-fuzzy-based optimisation approaches
Though the integration of AI and DM techniques for streamlining the RFQ process is lacking in the literature, a number of
optimisation approaches are found to determine an optimal pricing and production policy. Xu et al. (2017) and Lu, Zhang,
and Tang (2016) developed an optimisation model for perishable products. Both of the studies sought to introduce a number
of variables that are influential in developing an appropriate pricing policy. Xu et al. (2017) suggested the production rate,
demand rate, on-hand inventory level, etc. to determine the optimal pricing for perishable items, whereas Lu, Zhang, and
Tang (2016) proposed a demand function, production cost function, replenishment rate, etc. to determine the optimal pricing.
These studies, however, have a limited level of applicability, as they specifically focused on perishable items. To a large
extent, the variables introduced, such as the demand function and production cost function, would not be generic for other
types of products. Furthermore, the applicability of most of the optimisation models are required to comply with some
18 K.H. Leung et al.
Table 10. Comparison of the Smart-Quo with other pricing-related modelling.
Hypothetical approach Non-fuzzy-based optimisation approach Game-theoretic approach in fuzzy environment
Smart-Quo Optimal pricing
for new and
remanufactured
products (Abbey,
Blackburn, and
Guide 2015)
Production and pricing
problems in make-to-
order supply chain
with cap-and-trade
regulation (Xu et al.
2017)
Optimal dynamic pricing
and replenishment
policy for perishable
items with inventory-
level-dependent
demand (Lu, Zhang,
and Tang 2016)
Pricing decisions for
substitutable products
with horizontal and
vertical competition
in fuzzy environments
(Wei and Zhao 2016)
Pricing and retail
service decisions in
fuzzy uncertainty
environments (Zhao
and Wang 2015)
Objective(s) To discover the hidden
relationships between the e-
commerce sales performance
of a product and consumer
purchasing behaviour,
thereby providing pricing
decision support
To investigate the
optimal pricing
of new and
remanufactured
products
To determine an optimal
joint pricing and
production policy for
perishable products
To determine the optimal
dynamic pricing
and replenishment
policy for perishable
items with inventory-
level-dependent
demand
To explore the effect
of the vertical and
horizontal competition
in a fuzzy supply
chain with two
manufacturers and
the common retailer
on pricing decisions
of two substitutable
products
To study the pricing
and retail service
decisions of a product
in a fuzzy supply
chain with one
manufacturer and two
retailers
Scope Pricing determination under the
RFQ process in e-commerce
B2B environment
The nature of pricing
for remanufactured
products, and the
impact of consumer
segments
Development of time
optimal control model
to maximise an
enterprise’s profit
Profit maximisation
problem for inventory
control, and pricing
decision support
Fuzzy uncertainties
existing in a common-
retailer chain
Fuzzy uncertainties
existing in a
common-retailer
chain
E-commerce-
oriented Yes NoNoNoNoNo
AI-based model Yes No No No No, only introduced the
fuzziness of variables No, only introduced the
fuzziness of variables
Tool(s) Fuzzy association rule mining,
fuzzy logic, big data analytics Hypothesis testing,
regression
analysis, mixed
mathematical
modelling
Mathematical modelling,
algorithm development Mathematical modelling,
algorithm development Game modelling,
fuzzy-based linguistic
expressions
Game modelling,
fuzzy-based linguistic
expressions
International Journal of Production Research 19
Variables
selected for
investigation
Price discounts, product web
page’s hit rate and sales
record, delivery lead time,
customer’s order frequency,
quantity and credibility
Price discounts,
attractiveness of
remanufactured
products, product
segments
Optimal pricing,
production rate,
demand, production
cost function, product
deterioration rate,
inventory level
Optimal pricing,
inventory level,
replenishment rate,
demand function,
production cost
function
Manufacturing cost,
market base, price
elasticity
Customer demand,
manufacturing
costs, service cost
coefficients
Level of
applicability Relatively high – quantitative
values of each variable are
retrievable from database,
serving as the input for
generating pricing decision
support using decision rules
derived via the proposed
mining algorithm
Relatively low –
only identified the
relationships among
the variables under
different product
segments
Relatively low – the
applicability of the
developed model
is restricted to the
specific assumptions
made and the business
scenario
Relatively low – require
human experience to
identify the fuzziness
of each variables and
knowledge in game
theory
20 K.H. Leung et al.
specific assumptions and business scenarios. Only a small minority of firms might be able to adopt optimisation models for
generating decision support in actual business. Compared with the optimisation approach, in this sense, the methodology
introduced in this study presents a generic tool for industry practitioners to mine the unique rules or relationships of the
parameters, based on their real data. These extracted rules can then be used for the development of a decision support
system to generate such as the Smart-Quo.
(iii) Comparing Smart-Quo with existing fuzzy-based game-theoretic approaches
Apart from optimisation methods, researchers have proposed game-theoretic approaches to deal with pricing decision under
a certain kind of scenario. Wei and Zhao (2016) explored the effect of the vertical and horizontal competition in a fuzzy sup-
ply chain with two manufacturers and a common retailer on pricing decisions of two substitutable products. Zhao and Wang
(2015) studied the pricing and retail service decisions of a product in a fuzzy supply chain with one manufacturer and two
retailers. Both of the studies introduced the fuzzy uncertainties in a common retailer chain, making the studies more capable
of reflecting the real business environment. However, similar to the optimisation methods, the fuzzy-based game-theoretic
approaches often preset a specific scenario for investigation. The study performed by Wei and Zhao (2016) considered a
fuzzy supply chain with two manufacturers and one common retailer; whereas Zhao and Wang (2015) investigated a fuzzy
supply chain with one manufacturer and two retailers. The applicability of fuzzy-based game-theoretic approaches are still
relatively low, given the complexity of the real business environment, and the requirement of firms’ expertise in game
theory. Comparing the proposed fuzzy association rule mining approach with fuzzy-based game-theoretic approaches, the
proposed approach offers a light-weight, systematic method to extract relationships among several factors influencing pric-
ing decisions. The generalisation capability of the Smart-Quo allows it to be implemented in e-commerce retailers, however,
complex their supply chain is and whatever types of products they are offering to B2B customers.
6. Conclusive remarks and future work
The e-commerce business environment not only reshapes B2C consumer purchasing behaviour, but also the B2B. While
the source and purchasing process of B2B consumers can all be performed via B2B e-commerce sites, the RFQ process is
also triggered by B2B customers who submit a request online. Conventionally, determining the price in a quotation by B2B
suppliers is generally manual, but is critical. It is a complex process that involves a large amount of variation and can only
be performed by an experienced person who has the ability to determine the total cost and profit margin. The complexity
in price determination increases with the increasing number of factors that can be taken into consideration in today’s B2B
e-commerce environment. Information, such as the availability of big data for each particular consumer and the popularity
of the products, can be retrieved from the cloud and should be effectively extracted for the mining of essential knowledge so
as to assist practitioners during the decision making process. As the selling channels have been diversified both online and
offline, utilisation of big data and effective management of data across the supply chain have become the rules for business
success in today’s digitalised supply chain and competitive market:
•Utilisation of big data: With the integration of suitable data mining, machine learning or deep learning techniques,
proper use of big data enables the extraction of hidden knowledge. In this paper, the use of fuzzy association rule
mining approach for extracting the relationships between B2B consumer purchasing behaviour and a quotation
pricing, is an example of how historical data can be utilised and how the hidden knowledge extracted from the
fuzzy association rule mining approach assists decision makers in making better pricing decisions in the future.
•Effective data management across the supply chain: Data can be used for further knowledge extraction only if
information is made available across the supply chain. This could be a tedious task if an individual business
competes as stand-alone entities in the supply chain. Nevertheless, technically, the blockchain technology has
the capability to transform the conventional e-commerce industry by removing the existence of third-parties and
decentralising unnecessary control along the supply chains in the aspect of e-commerce order fulfilment.
This paper integrates a fuzzy association rule mining and fuzzy logic approach into a single system, namely, Smart-Quo,
for providing merchants with decision support during the RFQ process. By discovering the hidden relationships between
the sales performance of a product and consumer purchasing behaviour, the proposed system improves the merchants’
decision-making ability in determining the price in the quotation. A case study presented validates the feasibility of the
proposed approach in formulating a pricing strategy tailor-made for each B2B customer. With the use of Smart-Quo, there is
a dramatic reduction in both the cycle time in making a pricing decision for each customer and the number of staff involved
in the quotation process, with an increase in the acceptance rate of the quotations. This clearly indicates the benefits of
embedding artificial intelligence techniques into light-weight IT solutions in the real business environment, particularly in
today’s B2B e-commerce business environment, thereby assisting practitioners in making decisions. It is suggested that,
depending on the operating scale and nature of business of a firm, factors affecting the pricing decision, aside from those
International Journal of Production Research 21
introduced in this paper, can also be taken into account by including them in the proposed fuzzy association rule mining and
fuzzy logic approach, so as to deliver a more sophisticated pricing strategy which can maximise profit for the suppliers. As
the efficiency of e-commerce supply chains hinges on data transparency and utilisation across the supply chain, researchers
are recommended to further utilise big data from the cloud for developing tools that assist practitioners in making better and
faster decisions in the field of B2B and B2C e-commerce business environments, through the applications and innovations of
artificial intelligence, data mining, machine learning tools, as well as the integration of blockchain technology, particularly
in the areas of e-commerce order fulfilment process, prediction of consumer purchasing habits in online sales channels, and
omni-channel selection for advertising and product selling.
Funding
The authors wish to thank the Research Office of The Hong Kong Polytechnic University for supporting the project (Project Code: RU5T
and G-YBMS).
Disclosure statement
No potential conflict of interest was reported by the authors.
References
Abbey, J. D., J. D. Blackburn, and V. D. R. Guide Jr. 2015. “Optimal Pricing for New and Remanufactured Products.” Journal of
Operations Management 36: 130–146.
Albright, S., W. Winston, and C. Zappe. 2010. Data Analysis and Decision Making with Microsoft Excel. Mason, OH: South-Western.
Aliev, R. A., B. Fazlollahi, B. G. Guirimov, and R. R. Aliev. 2007. “Fuzzy-genetic Approach to Aggregate Production–Distribution
Planning in Supply Chain Management.” Information Sciences 177 (20): 4241–4255.
Bichler, M., J. Kalagnanam, K. Katircioglu, A. J. King, R. D. Lawrence, H. S. Lee, G. Y. Lin, and Y. Lu. 2002. “Applications of Flexible
Pricing in Business-to-Business Electronic Commerce.” IBM Systems Journal 41 (2): 287–302.
Buczak, A. L., B. Baugher, E. Guven, L. C. Ramac-Thomas, Y. Elbert, S. M. Babin, and S. H. Lewis. 2015. “Fuzzy Association Rule
Mining and Classification for the Prediction of Malaria in South Korea.” BMC Medical Informatics and Decision Making 15 (1):
47.
Chen, X., G. Hao, and L. Li. 2014. “Channel Coordination with a Loss-Averse Retailer and Option Contracts.” International Journal of
Production Economics 150: 52–57.
Chen, G., and Q. Wei. 2002. “Fuzzy Association Rules and the Extended Mining Algorithms.” Information Sciences 147 (1–4): 201–228.
García-Crespo, Á., B. Ruiz-Mezcua, J. L. López-Cuadrado, and I. González-Carrasco. 2011. “A Review of Conventional and Knowledge
Based Systems for Machining Price Quotation.” Journal of Intelligent Manufacturing 22 (6): 823–841.
He, W., S. Zha, and L. Li. 2013. “Social Media Competitive Analysis and Text Mining: A Case Study in the Pizza Industry.” International
Journal of Information Management 33 (3): 464–472.
Ho, G. T. S., W. H. Ip, C. H. Wu, and Y. K. Tse. 2012. “Using a Fuzzy Association Rule Mining Approach to Identify the Financial Data
Association.” Expert Systems with Applications 39 (10): 9054–9063.
Ho, G. T. S., H. C. W. Lau, S. H. Chung, R. Y. K. Fung, T. M. Chan, and C. K. M. Lee. 2008. “Fuzzy Rule Sets for Enhancing Performance
in a Supply Chain Network.” Industrial Management & Data Systems 108 (7): 947–972.
Hong, T.-P., K.-Y. Lin, and S.-L. Wang. 2003. “Fuzzy Data Mining for Interesting Generalized Association Rules.” Fuzzy Sets and Systems
138 (2): 255–269.
Hvam, L., S. Pape, and M. K. Nielsen. 2006. “Improving the Quotation Process with Product Configuration.” Computers in Industry 57
(7): 607–621.
Jain, V., L. Benyoucef, and S. G. Deshmukh. 2008. “A New Approach for Evaluating Agility in Supply Chains Using Fuzzy Association
Rules Mining.” Engineering Applications of Artificial Intelligence 21 (3): 367–385.
Kannan, P. K., and P. K. Kopalle. 2001. “Dynamic Pricing on the Internet: Importance and Implications for Consumer Behavior.”
International Journal of Electronic Commerce 5 (3): 63–83.
Lau, H. C. W., G. T. S. Ho, K. F. Chu, W. Ho, and C. K. M. Lee. 2009. “Development of an Intelligent Quality Management System Using
Fuzzy Association Rules.” Expert Systems with Applications 36 (2): 1801–1815.
Lee, C. K. H., Y. K. Tse, G. T. S. Ho, and K. L. Choy. 2015. “Fuzzy Association Rule Mining for Fashion Product Development.”
Industrial Management & Data Systems 115 (2): 383–399.
Li, F., L. Li, C. Jin, R. Wang, H. Wang, and L. Yang. 2012. “A 3PL Supplier Selection Model Based on Fuzzy Sets.” Computers &
Operations Research 39 (8): 1879–1884.
Liao, S.-H., P.-H. Chu, and P.-Y. Hsiao. 2012. “Data Mining Techniques and Applications – A Decade Review From 2000 to 2011.”
Expert Systems with Applications 39 (12): 11303–11311.
Lu, L., J. Zhang, and W. Tang. 2016. “Optimal Dynamic Pricing and Replenishment Policy for Perishable Items with Inventory-Level-
Dependent Demand.” International Journal of Systems Science 47 (6): 1480–1494.
22 K.H. Leung et al.
Mabu, S., C. Chen, N. Lu, K. Shimada, and K. Hirasawa. 2011. “An Intrusion-Detection Model Based on Fuzzy Class-Association-Rule
Mining Using Genetic Network Programming.” IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and
Reviews) 41 (1): 130–139.
Minelli, M., M. Chambers, and A. Dhiraj. 2012. Big Data, Big Analytics: Emerging Business Intelligence and Analytic Trends for Today’s
Businesses. Hoboken: John Wiley & Sons Inc.
Niazi, A., J. S. Dai, S. Balabani, and L. Seneviratne. 2006. “Product Cost Estimation: Technique Classification and Methodology Review.”
Journal of Manufacturing Science and Engineering 128 (2): 563–575.
Osman Hegazi, M., and M. A. Abugroon. 2016. “The State of the Art on Educational Data Mining in Higher Education.” International
Journal of Computer Trends and Technology 31 (1): 46–56.
Sánchez, D., M. A. Vila, L. Cerda, and J. M. Serrano. 2009. “Association Rules Applied to Credit Card Fraud Detection.” Expert Systems
with Applications 36 (2): 3630–3640.
Sowan, B., K. Dahal, M. A. Hossain, L. Zhang, and L. Spencer. 2013. “Fuzzy Association Rule Mining Approaches for Enhancing
Prediction Performance.” Expert Systems with Applications 40 (17): 6928–6937.
Statista. 2017. “Statista Digital Market Outlook – Trend Report.” Statista Report 2017 – B2B e-Commerce. New York: United States.
Statista, p. 7.
Teich, J. E., H. Wallenius, J. Wallenius, and O. R. Koppius. 2004. “Emerging Multiple Issue e-Auctions.” European Journal of
Operational Research 159 (1): 1–16.
Verma, S. K., R. S. Thakur, and S. Jaloree. 2017. “Fuzzy Association Rule Mining Based Model to Predict Students’ Performance.”
International Journal of Electrical and Computer Engineering (IJECE) 7 (4): 2223.
Waller, M. A., and S. E. Fawcett. 2013. “Data Science, Predictive Analytics, and Big Data: A Revolution That Will Transform Supply
Chain Design and Management.” Journal of Business Logistics 34 (2): 77–84.
Wei, J., and J. Zhao. 2016. “Pricing Decisions for Substitutable Products with Horizontal and Vertical Competition in Fuzzy
Environments.” Annals of Operations Research 242 (2): 505–528.
Xu, X., W. Zhang, P. He, and X. Xu. 2017. “Production and Pricing Problems in Make-to-Order Supply Chain with Cap-and-Trade
Regulation.” Omega 66: 248–257.
Zhao, J., and L. Wang. 2015. “Pricing and Retail Service Decisions in Fuzzy Uncertainty Environments.” Applied Mathematics and
Computation 250: 580–592.
Appendices
Appendix 1. The fuzzy counts of 1-itemsets.
Parameter item Counts Parameter item Counts Parameter item Counts
A. Very Low 0 E. Very High 0 J. Low 0
A. Low 0 F. Low 0.75 J. Medium 0
A. Medium 0 F. Medium 0.5 J. High 2.67
A. High 2 F. High 0.63 J. Very High 0.5
A. Very High 2.6 F. Very High 1.71 K. Very Low 0
B. Low 2 G. Low 0 K. Low 1.34
B. Medium 0.8 G. Medium 1.54 K. Medium 1
B. High 0.5 G. High 1 K. High 0.3
C. Low 1.6 G. Very High 0.5 K. Very High 1.4
C. Medium 1.5 H. Low 2.67 L. Low 0
C. High 0.63 H. Medium 0.8 L. Medium 0.5
C. Very High 0 H. High 0.8 L. High 1.67
D. Low 1.33 H. Very High 0 L. Very High 1.4
D. Medium 1.67 I. Very Low 0 M. Very Low 1.8
D. High 1 I. Low 0.5 M. Low 2
D. Very High 0 I. Medium 3.18 M. Medium 0
E. Low 0.5 I. High 0.17 M. High 0
E. Medium 1.3 I. Very High 0.38 M. Very High 0
E. High 1.33 J. Very Low 0
International Journal of Production Research 23
Appendix 2. Fuzzy logic output result in MATLAB Fuzzy Logic Toolbox
24 K.H. Leung et al.
Appendix 3. User interfaces of Smart-Quo
