improving service processes based on visualiz ation of human...

8
Improving service processes based on visualization of human-behavior and POS data: A case study in a Japanese restaurant Tomohiro Fukuhara 1 , Ryuhei Tenmoku 1 , Takashi Okuma 1 , Ryoko Ueoka 2 , Masanori Takehara 3 , and Takeshi Kurata 1 1 Center for Service Research, National Institute of Advanced Industrial Science and Technology, 1-1-1 Umezono, Tsukuba, Ibaraki 305-8568, JAPAN Tel: +81-3-3599-8568, Fax: +81-29-862-6548, E-mail: [email protected] 2 Graduate School of Design, Kyushu University, Fukuoka 3 Graduate School of Engineering, Gifu University, Gifu, JAPAN Abstract A case study of service process improvement based on visualization of human-behavior and POS (point-of- sales) data in a Japanese restaurant is described. We developed a human-behavior sensing and visualization suite for supporting managers and employees in actual service fields to understand and improve their service processes by visualizing both of behavior and POS data. We had an experiment using the suite in the restaurant, and confirmed that managers and employees were able to understand their ordinary processes, and make plans for improving their processes by using the suite. An overview of the suite and experiment results are described. Keywords: Service process improvement, human-behavior sensing, data visualization, POS data analysis, quality- control circle 1 INTRODUCTION Today, service industries play an important role in economy [1]. The weight of service industries are gradually growing in developed countries [2]. As growth of service industries, new research fields called service science, management and engineering (SSME) and service engineering are emerging [2-4]. In these fields, researches are trying to improve productivity of service industries based on various approaches such as industrial engineering (IE) [5,6], operations research (OR), data mining (DM) [7], game theory [8], sensors, and so on. For understanding and improving efficiency of service processes, measuring methods of processes are needed. In the field of IE, various methods have been developed for measuring behavior and processes such as time-and- motion study [9] and work sampling [10]. Although these techniques enable us to understand both of macroscopic and microscopic states of service processes, they have limitations on observing behaviors with respect to the cost (time and money) and comprehensiveness of data. The aim of this study is to support managers and employees who are engaged in actual service fields to understand their current service processes comprehensively, and support them to improve their processes. For this aim, we are developing a human- behavior sensing and visualization suite. Our suite observes behavior of employees contiuously and comprehensively at low cost by using wearable sensors [11], and visualizes behavior data by combining POS (point-of-sales) data. By combining both of behavior and POS data, managers and employees can understand efficiency of their processes objectively, and can make plans for improving processes effectively. We had an experiment of the suite in a Japanese cuisine restaurant for measuring efficiency of service processes. In this restaurant, there is a quality-control circle (QC circle) which is a voluntary group of employees to improve their productivity. We collaborated members of the QC circle to understand and improve their processes. We had an experiment for about a month in the restaurant, and found that the suite assisted managers and employees to (1) observe their ordinary behaviors and sales data, and (2) support them to make plans for improving their processes, and (3) verify effects of the plan with actual data. We observed changes of behavior and POS data. This paper consists of following sections. Section 2 describes related work and requirements for the support system for improving service processes. Section 3 describes an overview of the human-behavior sensing and visualization suite. Section 4 describes an experiment of the suite in a Japanese cuisine restaurant. Section 5 describes discussion on results. In Section 6, we describe conclusion and future work. 2 RELATED WORK There are several related work on service process improvement. We describe related work from several disciplines, and we describe requirements for the support system for improving service processes. 2.1 Industrial engineering approach In the field of IE, various methodologies and techniques have been developed for measuring employees’ work; motion-and-time study [9] and work sampling [10] are major techniques to observe behavior and processes of workers. Although these techniques are good way to obtain data from work space such as factories, they have limitations on collecting data with respect to following factors: (1) cost of investigation, (2) comprehensiveness of observation, and (3) privacy of customers. For the first issue, traditional IE techniques require much time and money to have an investigation because they are based on observation by human observers. Because budget of each service field is limited, it is difficult to have a long term investigation. For the second issue, they have limitations on collecting data comprehensively. When managers want to evaluate their service processes, they have to have an investigation for several days or weeks. During these periods, it is hard to observe every behaviors of whole employees through their work time because the number of observers are limited. 1 The 1st International Conference on Serviceology

Upload: others

Post on 04-Feb-2021

1 views

Category:

Documents


0 download

TRANSCRIPT

  • Improving service processes based on visualization of human-behavior and POS data: A case study in a Japanese restaurant

    Tomohiro Fukuhara1, Ryuhei Tenmoku1, Takashi Okuma1, Ryoko Ueoka2, Masanori Takehara3, and Takeshi Kurata1

    1 Center for Service Research, National Institute of Advanced Industrial Science and Technology, 1-1-1 Umezono, Tsukuba, Ibaraki 305-8568, JAPAN

    Tel: +81-3-3599-8568, Fax: +81-29-862-6548, E-mail: [email protected] 2 Graduate School of Design, Kyushu University, Fukuoka

    3 Graduate School of Engineering, Gifu University, Gifu, JAPAN

    Abstract A case study of service process improvement based on visualization of human-behavior and POS (point-of-sales) data in a Japanese restaurant is described. We developed a human-behavior sensing and visualization suite for supporting managers and employees in actual service fields to understand and improve their service processes by visualizing both of behavior and POS data. We had an experiment using the suite in the restaurant, and confirmed that managers and employees were able to understand their ordinary processes, and make plans for improving their processes by using the suite. An overview of the suite and experiment results are described. Keywords: Service process improvement, human-behavior sensing, data visualization, POS data analysis, quality-control circle

    1 INTRODUCTION Today, service industries play an important role in economy [1]. The weight of service industries are gradually growing in developed countries [2]. As growth of service industries, new research fields called service science, management and engineering (SSME) and service engineering are emerging [2-4]. In these fields, researches are trying to improve productivity of service industries based on various approaches such as industrial engineering (IE) [5,6], operations research (OR), data mining (DM) [7], game theory [8], sensors, and so on. For understanding and improving efficiency of service processes, measuring methods of processes are needed. In the field of IE, various methods have been developed for measuring behavior and processes such as time-and-motion study [9] and work sampling [10]. Although these techniques enable us to understand both of macroscopic and microscopic states of service processes, they have limitations on observing behaviors with respect to the cost (time and money) and comprehensiveness of data. The aim of this study is to support managers and employees who are engaged in actual service fields to understand their current service processes comprehensively, and support them to improve their processes. For this aim, we are developing a human-behavior sensing and visualization suite. Our suite observes behavior of employees contiuously and comprehensively at low cost by using wearable sensors [11], and visualizes behavior data by combining POS (point-of-sales) data. By combining both of behavior and POS data, managers and employees can understand efficiency of their processes objectively, and can make plans for improving processes effectively. We had an experiment of the suite in a Japanese cuisine restaurant for measuring efficiency of service processes. In this restaurant, there is a quality-control circle (QC circle) which is a voluntary group of employees to improve their productivity. We collaborated members of the QC circle to understand and improve their processes. We had an experiment for about a month in the restaurant, and found that the suite assisted managers and employees to

    (1) observe their ordinary behaviors and sales data, and (2) support them to make plans for improving their processes, and (3) verify effects of the plan with actual data. We observed changes of behavior and POS data. This paper consists of following sections. Section 2 describes related work and requirements for the support system for improving service processes. Section 3 describes an overview of the human-behavior sensing and visualization suite. Section 4 describes an experiment of the suite in a Japanese cuisine restaurant. Section 5 describes discussion on results. In Section 6, we describe conclusion and future work. 2 RELATED WORK There are several related work on service process improvement. We describe related work from several disciplines, and we describe requirements for the support system for improving service processes. 2.1 Industrial engineering approach In the field of IE, various methodologies and techniques have been developed for measuring employees’ work; motion-and-time study [9] and work sampling [10] are major techniques to observe behavior and processes of workers. Although these techniques are good way to obtain data from work space such as factories, they have limitations on collecting data with respect to following factors: (1) cost of investigation, (2) comprehensiveness of observation, and (3) privacy of customers. For the first issue, traditional IE techniques require much time and money to have an investigation because they are based on observation by human observers. Because budget of each service field is limited, it is difficult to have a long term investigation. For the second issue, they have limitations on collecting data comprehensively. When managers want to evaluate their service processes, they have to have an investigation for several days or weeks. During these periods, it is hard to observe every behaviors of whole employees through their work time because the number of observers are limited.

    1The 1st International Conference on Serviceology

    gotoタイプライターテキストProceedings of the 1st international conference on Serviceology, pp.1-8 (2013)

    gotoタイプライターテキスト

    gotoタイプライターテキスト

  • For the third issue, traditional IE techniques are not suited to actual service fields where ordinary customers are buying goods or receiving services. When human observers observe behavior of employees in actual service fields, they might hinder natural interactions between customers and employees. Furthermore, human observers might harm privacy of customers who are not willing to notify what they purchased or received. Although there are studies of observing behavior of employees and processes in service fields such as restaurants [12] and hospitals [13], it takes much time and money to observe processes. In this paper, we propose a human-behavior sensing and visualization suite that can collect data continuously and comprehensively at low cost by using wearable sensors. Our suite can work in actual service fields without harming privacy of customers because the suite only observes behavior of employees. Customers and employees can behave naturally in service fields. 2.2 Operations research and data mining

    approaches There are many studies on improvement of productivity of services from viewpoints of operations research (OR) and data mining (DM). In the context of OR, studies aiming at improving revenues are called revenue management [14, 15] or yield management [16,17]. Yield management has been developed in airline industries since 1970’s because airline companies had risks on having flights with vacant seats; some seats can be cancelled and others might not be used because passengers missed the flight (called no-show). For recovering this loss, airline companies developed the revenue management system that maximizes revenues by selling seats more than the capacity of a plane [16]. For fulfilling seats, the system sells seats at multiple prices; for leisure travelers, they want to purchase tickets at discount rate, and for business persons, they do not care about prices. The point of the system is to decide the number of seats for selling at regular price. Belobaba proposed a method called Expected Marginal Seat Revenue to calculate optimal number of seats for selling at regular price [18]. Hotels, theaters, and restaurants have similar problems, and therefore they also have similar solutions. In the context of DM, various studies have been reported which are based on computing probability of purchase of goods or services from data. Takenaka et al. proposed a method to predict demands of customers in retail stores and restaurants by combining sales data and environment data such as weather, the day of week, social events, and so on [19]. Ishigaki et al. proposed a method to categorize both of customers and products by using POS data which are linked to IDs of customers [7]. These studies mainly based on customers’ history of purchase data, and they did not focus on behavior of employees. For improving processes, behavior data of employees is needed because processes can be improved by identifying inefficient processes. Furthermore, calculation of optimal parameters is sometimes difficult because demands of customers would vary according to seasons and social events. In this paper, we aim to improve service processes by using not only sales data but also behavior data of employees. By using both of behavior and sales data, we can find issues that hinder efficiency, and improve their processes effectively. 2.3 Virtual reality approach Burger et al. proposed a test bed for evaluating service processes using virtual reality (VR) [20]. Their prototype

    system called ServLab enables managers and employees to design and evaluate new service processes in a virtually constructed service field. Simo et al. also proposed a test bed called SINCO for designing service operations using VR [21]. Hyun et al. also proposed service field simulator which uses the omni-direction immersive display to visualize a virtualized service field [22]. These studies are suitable for designing and evaluating service processes, however, these studies are not based on analysis of actual behavior and sales data in service fields. For improving actual service processes, analysis of service fields from viewpoints of behavior of employees and various business data of the fields is needed. We aim to create a human-behavior sensing and visualization suite to improve service processes by using both of behavior and sales data of service fields. 2.4 Sensor based behavior analysis approach For understanding behavior of employees, various studies that use sensors have been reported. Inoue et al. proposed an indoor positioning system that uses beacon devices embedded in a building [23]. Sumi et al. proposed a sensor system that can analyze social interactions among people [24]. Choudhury and Pentland proposed a sensor system called sociometer that can collect human interaction data [25]. With sociometer, human interaction data such as who was accompanied, how long s/he talked, and how often s/he moved can be recorded. Kim et al. uses sociometer to analyze shoppers’ behaviors, and found several correlations between actions of customers and interests for items [26]. Olguín and Pentland proposed an approach to measure behavior of employees by using the sociometer for analyzing and improving productivity of employees in organizations such as banks and hospitals [27,28]. Ara et al. proposed an approach to analyze behavior data of humans by linking with other data called performance indicators such as financial profit, amount of communication, employee satisfaction (ES), and customer satisfaction (CS) [29]. They created a feedback system that support employees to find their current communication states, and improve them by visualizing interactions among managers and employees [30]. Huang et al. proposed a self-training system that supports nurses to learn the way to transfer patients from bed to a wheel chair based on behavior sensing [31]. In the business domain, Toward logistics Co. Ltd., which is a Japanese logistic company, succeeded to improve their services by using car driving sensors. They developed a system called TRU-SAM (Truck Support, Administration and Management) for recording and visualizing driving data of a truck [32]. TRU-SAM records speed, engine revolutions, fuel efficiency, sudden breaks, and shakes of the truck body, and the system gives feedback to drivers by visualizing the data. Drivers can find his/her drive operations. With TRU-SAM, this company reduced fuel costs, the number of car accidents, and as a result, they saved the insurance cost. This company succeeded to improve their processes based on measuring and visualizing data. Ueoka et al. proposed a scheme for measuring and assisting employees to improve service processes by using wearable sensors and POS data [33]. This scheme called CSQCC (Computer-supported Quality Control Circle) aims at improving processes by assisting a group of employees that has a motivation for improving productivity. Ueoka reported a case study of human-behavior sensing and service process improvement in a Japanese cuisine restaurant. In this paper, we also aim at improving service processes in actual fields. For

    2

  • achieving this aim, we focus on assisting a group of employees that has a motivation to improve their productivity. We provide the suite to this kind of group. 2.5 Requirements for human behavior sensing and

    visualization suite Previous studies have limitations on improving service processes with respect to the use of behavior data and POS data. We consider that a system that can support managers and employees to understand and improve their processes should have following functions: (1) Behavior sensing function that can collect behavior

    data of employees through their work time (about eight hours) continuously. This sensing should be executed in a building. The sensing should not harm privacy of customers.

    (2) Data visualization function for supporting managers and employees to understand their current behaviors and processes. This function should provide visualization of data by combining behavior data and sales data.

    For the first function, behavior sensing should be able to collect behavior data of employees continuously and comprehensively in indoor environment. Traditional IE techniques have limitations on observing behavior of employees comprehensively. Comprehensive observation helps managers and employees to evaluate efficiency of their current service processes. Actually, we confirmed that comprehensive observation enabled managers and employees to find out inefficient processes during the lunch time operation (see section 4.3). Furthermore, the function should not harm privacy of customers; customers and employees can behave naturally even if observation is executed. For the second function, the system should have a function for visualizing behavior data by combining business data. For improving processes, combining behavior data and sales data is needed because if we look at behavior data only, we cannot evaluate the behavior as good or bad for their business. For evaluating each behavior, we need to analyze behavior data by combining sales data. For visualizing data, we need a map of service fields because managers and employees want to analyze behavior and POS data by linking with specific rooms or areas of the fields. So, the system should have a function for creating a map of the fields instantly. Target users of the system are (1) managers who want to understand their current service processes objectively, and (2) employees who also want to understand daily processes and want to change their processes. 3 HUMAN-BEHAVIOR SENSING AND

    VISUALIZATION SUITE In this section, we describe an overview of our human-behavior sensing and visualization suite. Fig. 1 shows an overview of the suite. The suite consists of following components. 1. Interactive 3D indoor modeller 2. Pedestrian dead-reckoning (PDR) 3. Sensor data fusion (SDF) 4. Data visualization tool Interactive 3D indoor modeler allows users to create a 3D model of a service field. Users can create a 3D model of the field from still images which are taken by ordinary digital cameras [34]. Because the modeler estimates parameters of a camera automatically, users can easily create a 3D model of the field from images.

    Pedestrian dead-reckoning (PDR) allows users to track the location and orientation of a person [35]. We have developed a sensor which contains multiple sensors such as barometer, magnetometer, accelerometer, gyroscope, thermometer, and an RFID reader. By using these sensors, PDR estimates position, orientation, and velocity of a person. The error rate of location estimation by PDR only is 4.3%, and this can be corrected by using action recognition method to 2.0% [35]. Actually, locations of employees are estimated by using the sensor data fusion (SDF) which corrects locations estimated by the PDR. The battery life of PDR sensor is 12 hours; it is possible to observe employees’ whole behaviors through their work time. Batteries are charged every night by using the sensor station which can be treated by employees easily. SDF estimates the location and orientation of a person. Compared to PDR, SDF estimates the location and orientation more precisely by combining several sensor data such as the map data of the field which is created by the interactive 3D indoor modeller, camera data which is obtained from video cameras in the field, and RFID sensors [36]. The average error rate of location estimation using SDF is 2.2 meteres [37], which is enough precision for analyzing behavior of employees. Data visualization tool allows users to browse both of behavior data and various field data such as POS (point-of-sales) data and nurse-call data in nursing care homes. Fig. 2 shows the screen image of the visualization tool. The tool provides several visualizations of service fields such as the number of customers in dining areas, the number of orders, and staying time of workers in each area. 4 SERVICE PROCESS IMPROVEMENT IN A

    JAPANESE CUISINE RESTAURANT In this section, we describe a case study of service process improvement using the human-behavior sensing and visualization suite. We had an experiment of the suite

    Fig.1 Overview of the human-behavior sensing suite.

    Video cameras

    Visualization tool

    RFID tags

    Interactive 3D indoor modeler

    POS data Pedestrian Dead Reckoning (PDR)

    An employee equipped with PDR

    Sensor data fusion (SDF)

    Field data

    PDR data

    Environment sensor data

    Visualization of staying time and number of orders

    Visualization of waiting staff and customers

    Fig.2 Screen image of the visualization tool. Customers and waiting staff are displayed on the 3D model of the restaurant. POS data is displayed in the right panel.

    Waiting staff

    POS data

    Customers

    3

  • Tab. 1 Terms of experiment. Term Period Total Jan 12 to Feb 9, 2001 First term (Before) Jan 12 to 18, 2011 QC circle Feb 1, 2011 Second term (After) Feb 3 to 9, 2011

    in a Japanese cuisine restaurant, which is located at Ginza, Tokyo, to observe changes of behaviors of waiting staff, and its effect on POS data of the restaurant. We first describe overviews of the experiment and the restaurant, and describe activities in the quality-control (QC) circle that used our suite to confirm and improve their service processes. 4.1 Overview of the experiment The aim of this experiment was to evaluate whether the suite can support managers and employees to understand their service processes by combining behavior data and POS data. We also aim at the support of managers and employees to improve their processes by using our suite. We collaborated with the QC circle of the restaurant to support their observation and improvement of their processes. The experiment was held for 29 days from January 12 to February 9, 2011. Among 29 days, QC circle members selected the first week (first term) for confirming their ordinary processes; they discussed issues that hinder their productivity, and made plans for improving their processes (see section 4.3). After making plans, they implemented the plans during the last seven days (second term). Tab. 1 shows the terms of experiment. QC circle meeting was held on February 1st, 2011. This experiment was approved by the committee for ergonomic experiments of national institute of advanced industrial science and technology. 4.2 Overview of the restaurant The restaurant consists of two floors, and has 229 seats. Total area of the restaurant is about 800 square meters. Business hours are from 11:30 a.m. to 11:00 p.m. through Monday to Saturday. On Sunday, it closes at 10 p.m. B1 floor holds 129 seats, and has sushi bar, tables, and two zashiki rooms (private rooms). In the pantry, waiting staff prepare drinks. Foods are cooked at the kitchen, and conveyed to the dining areas by carts. B2 floor consists of zashiki rooms only, and there are 100 seats. Fig. 3 shows an overview of the B1 floor. We mainly investigated behaviors of waiting staff in the B1 floor. With respect to the types of staff, there are three types of staff working in the restaurant: (1) waiting staff, (2) bussers, and (3) cooks. The main tasks of waiting staff are preparing tables, taking orders, serving drinks and foods, and accounting. Bussers convey foods from kitchen to dining areas, and they clean tables after customers leave. Cooks stays at kitchen and cook according to orders. In this experiment, we mainly observed behaviors of waiting staff. Orders of customers are registered in POS database, and forwarded to the kitchen to start cooking. In the POS data, there are two types of orders, i.e., initial orders (IOs) and additional orders (AOs). IOs are the first time order for a group of customers; when the group arrived and orders several dishes, these dishes are registered as IOs. AOs are orders that are accepted after the IOs; when the group call a waiting staff to have more drinks or foods, these orders are registered as AOs.

    Tab. 2 Example of POS data of a day of the restaurant

    No.

    Grou p ID

    Order

    T ype

    Table No.

    Staff

    Time

    Checkout

    Menu

    Unit price

    Quantity

    Reserved

    1 5 IO 7 A 17:21 22:18 Beer 650 2 No2 5 AO 7 B 17:30 22:18 Sashi-

    mi 980 1 No

    3 5 AO 7 B 17:30 22:18 Udon 580 1 No4 6 IO 9 C 18:00 21:12 Banq-

    uet 0 20 Yes

    5 6 IO 9 C 18:00 21:12 Sushi 1980 20 Yes

    Tab. 2 shows an example of the POS data of the restaurant. The data has several columns; No. indicates the sequence of records, Group ID is the ID of each group, Order Type indicates initial or additional orders, Table No. is the table that a group seated, Staff is the ID of staff who took orders, Time indicates the order time, Checkout is the checkout time. Menu, Unit Price, Quantity are orders by the group. The last column named Reserved indicates whether orders are reserved or not. For example, there were two groups ID5 and ID6 appeared in Tab. 2. Group ID5 seated at table no. 7, and ordered two beers as IOs, and ordered a sashimi plate and a bowl of udon noodles as AOs. This group checked out at 22:18. Group ID6 is customers with reservation. The group is allocated to seat at table no. 9. The group reserved a banquet course which contains 20 sushi set (note that the price of banquet is 0. This is because actual orders of the banquet are listed in the following records). 4.3 Process improvement by the QC circle On February 1st, 2001, members of the QC circle had a meeting for discussing issues that hinder efficiency of their service processes based on the data obtained in the first term. Fig. 4 shows the scene of meeting; members used the visualization tool to confirm behavior and POS data of the first term. Through discussion, they found following issues: (1) waiting staff often leaves dining areas during the lunch time, (2) staying time of waiting staff in dining areas was shorter than expected, and (3) bussers’ work and waiting staff’s work are not separated completely.

    Fig.3 An overview of the B1 floor of the restaurant. The floor consists of sushi bar, tables, and zashiki rooms (private rooms). There are 129 seats in the B1 floor.

    Sushi bar

    Tables

    Zashiki room

    Pantry

    Kitchen

    Cash desk

    Entrance

    Zashiki room

    B1 floor (dining areas)

    4

  • For the first issue, members noticed that waiting staff often left dining areas during the lunch time by looking at trajectories of staffs. Fig. 5 shows the trajectory; a waiting staff left dining area during the lunch time and went to the office. They guessed the reasons for this as that this staff received a reservation request from customer via phone, she had to check reservation list which is stored at the office, so she had to return back to the office during the lunch time. From this, members found the necessity of staying at dining area during busy hours such as the lunch time and dinner time. For the second issue, staying time of waiting staff was shorter than expected. The ratio for staying in dining areas was 41.8%, which was not a proper ratio as waiting staff. So, they decided to increase the staying time in dining areas. As a key performance indicator (KPI) for evaluating improvement of their plans, they decided to use the stay ratio in dining areas. As improvement plan, they decided to prepare for banquets which are planned in the evening after the lunch time. For the third issue, work of waiting staff and bussers are not separated; waiting staff convey foods from kitchen and cleans tables that are main tasks of bussers. Therefore, members decided to separate works of waiting staff and bussers clearly. 4.4 Results For verifying results of improvement plans, we confirmed following variables: (1) stay ratio of waiting staff in dining areas which was defined as a KPI by QC members, (2) number of additional orders per customer which is not defined as a KPI by QC members, but it is an valuable index to understand effects of the first plan, and (3) walk distance of waiting staff for evaluating whether workloads

    Tab. 3 Categories of operation hours of the restaurant Category Hours Lunch time 11:00 - 15:00 Tea time 15:00 - 18:00 Dinner time (Core dinner time)

    18:00 - 23:00 (19:00 - 22:00)

    increased by implementing the improvement plans in the second term. We divided business hours of the restaurant into four categories: (1) lunch time, (2) tea time, (3) dinner time, and (4) core dinner time. Tab. 3 shows the category of operation hours. This classification is based on our analysis of POS data. We use these categories to analyze behavior and POS data. 4.4.1 Stay ratio of waiting staff in dining areas Fig. 6 shows the stay ratio of waiting staff in dining areas for each hour. The X-axis shows hours, and Y-axis shows the average stay ratio of waiting staffs for each hour. The data showed the average ratio through the first and the second terms respectively. By looking at the ratio by each time category, the graph shows that (1) there are few differences between two terms in the lunch time, (2) the ratio was decreased in the tea time in the second term, and (3) the ratio was increased in the dinner time in the second term. For the first point, there were few differences between two terms in the lunch time. By summing up the ratio for each term, there was only 1.0 point difference between two terms. We consider that it was difficult to change processes in the lunch time because this period is so busy. For the second point, the ratio was decreased during the tea time in the second term. This was because waiting staff went back to backyards and prepared for banquets which were planned in the dinner time. At 4 p.m., the ratio decreased from 45.4% to 36.5%. Finally, the ratio was increased in the dinner time. For example, the ratio increased from 43.9% to 51.3% (+7.4 points) in the second term. We observed effects of improvement plans made by QC members. 4.4.2 Number of additional orders As one of effects of the improvement plan, we confirmed a difference on the number of additional orders. Fig. 7 shows the number of additional orders per customer. The data shows average number for each hour in the first and the second terms. The figure also shows that the number of additional orders in the dinner time was increased in the second term.

    Fig.4 Scene of the QC circle meeting. Members discussed issues that hinder productivity by using the visualization tool. They checked their movement and POS data.

    Fig.5 Trajectory of a waiting staff during the lunch time in a day. White arrow indicates the direction of movement of the staff. She left the dining area and went to the office during the lunch time.

    Kitchen

    Dining area

    Office

    Trajectory of a waiting staff during the lunch time in a day(Arrow indicates the direction of move)

    Fig.6 Comparison of the stay ratio of waiting staff in dining areas between two terms.

    30%

    35%

    40%

    45%

    50%

    55%

    11 12 13 14 15 16 17 18 19 20 21 22

    Stay ra

    tio in

     dining areas (%)

    Hour

    First term Second term

    Lunch time Tea time Dinner time

    5

  • For example, the number at 7 p.m. was increased to 1.2 from 0.7. Throughout the dinner time, the number was increased from 4.2 to 5.9 (+1.7) in the second term. We also found that there was a significant difference between two terms in the core dinner time. Fig. 8 shows the comparison of the numbers of additional orders between two terms. We confirmed that there is a statistical significance between two terms by Mann-Whitney’s U test (U=121, n1=n2=21, p=.012 < .05, two-tailed). The medians were 0.7 for the first term, and 1.1 for the second term. 4.4.3 Walk distance Fig. 9 shows the walk distance of waiting staff per customer. This parameter represents an index of workloads of waiting staff. The figure shows the average distance per customer for each hour. We found that there were no significant differences in each time categories. 5 DISCUSSION In this experiment, we supported members of the QC circle to improve processes by visualizing behavior and POS data of their ordinary processess. We confirmed changes on both of behavior and POS data in the second term of experiment. We observed increases of the stay ratio and the number of additional orders per customer in dinner time. Our suite assisted QC circle members to confirm their ordinary processes, supported them to find issues, and make plans for improvement.

    Fig. 10 shows comparison of the number of additional orders in the core dinner time between January and February. The figure shows the data through 2011 to 2013. There was a significant difference between January and February only in 2011; we could not find significant differences in 2012 and 2013. This means that waiting staff took orders much more than other years by staying at dining areas. We should consider other factors that affect the number of additional orders. One of such factors is the reservation of seats. Because customers who reserve seats often order course dishes beforehand, there are few chances to order another foods and drinks. Fig. 11 shows the number of reservation through a year. The figure shows that the number is especially high in December. This is because there is a custom to have a party at the end of a year in Japan. There is also a custom to have a new year’s party in January. The figure shows that the number in January is not high as much as in December. By comparing January and February, the number is higher in January (M=515, SD=69.7, Mdn=491) than in February (M=452, SD=59.3, Mdn=460). There is no significant difference between January and February (U=24, n1=n2=6, p=.240). From this figure, we consider that effects of reservation on the number of additional orders is low, and there was a positive effect by the process improvement.

    Fig.7 Number of additional orders per customer. Statistical significance was found in the core dinner time (see also Fig. 8).

    0.00.20.40.60.81.01.21.4

    11 12 13 14 15 16 17 18 19 20 21 22Num

    ber o

    f add

    ition

    al orders 

    per customer (d

    ishe

    s/custom

    er)

    Hour

    First term Second term * p < .05

    *

    Lunch time Dinner timeTea timeHour

    Fig.8 Comparison of the number of additional orders per customer in the core dinner time. There was a statistical significance between two terms (by Mann-Whitney’s U=121, n1=n2=21, p=.012 < .05, two-tailed).

    First term Second term

    0.5

    1.0

    1.5

    2.0

    2.5

    Num

    ber o

    f add

    ition

    al o

    rder

    s in

    the

    core

    din

    ner t

    ime

    *

    * p < .05

    Fig.9 Walk distance per customer (meters/customer).

    0

    50

    100

    150

    200

    11 12 13 14 15 16 17 18 19 20 21 22

    Walk distan

    ce per customer 

    (meters/custom

    er)

    Hour

    First term Second term

    Lunch time Dinner timeTea timeHour

    Fig.10 Comparison of the number of additional orders in the core dinner time between January and February. The figure shows data of January and February through 2011 to 2013. There is a significant difference between January and February only in 2011.

    0.0

    0.5

    1.0

    1.5

    2.0

    2.5

    3.0

    Num

    ber o

    f add

    ition

    al o

    rder

    s pe

    r cus

    tom

    er

    Jan 2011(n=21)

    Feb 2011(n=21)

    Jan 2012(n=21)

    Feb 2012(n=21)

    Jan 2013(n=19)

    Feb 2013(n=21)

    *

    * p < .05

    6

  • The results obtained from this experiment imply that efficiency of service can be improved by introducing an appropriate plan which is planned based on observed behavior and POS data. In many service fields, managers make plans for improving processes based on their intuitions and experiences. Although these human-senses are important for achieving efficiency in some case, but it is difficult to apply this approach to other fields. Meanwhile, we aim to support process improvement by measuring and visualizing actual behavior data of the waiting staff and POS data of the restaurant. Based on actual data, QC circle members were able to create a plan for improving their processes. Our future work is to improve the suite so that anyone who is in service industries can use the suite for improve their processes. For achieving this goal, we have two issues: (1) usability of the suite, and (2) feedback function of the system. The first is the usability of the suite. In service industries, various people are working whose age and education are quite different. There exist people who are not familiar with sensors and computers. In this experiment, waiting staff were not familiar with sensors at the first time, so we had to assist them to use sensors. For allowing anyone in service fields to use our suite, we have to improve the usability of the suite. The second is a feedback function of the data. In this experiment, behavior and POS data were not provided during the experiment terms. If people can easily check their behavior and POS data quickly, the data would motivate them to improve their work. As Ara et al. [30] and Huang et al.[31] reported the impact of feedback of sensor data to employees, we consider that short-term or real time feedback would improve the quality of services drastically. For enabling employees to confirm their behavior and POS data quickly is our future work. 6 CONCLUSION In this paper, we reported a case study of human-behavior sensing and visualization suite in a real service field. Our suite can observe behavior of employees with wearable sensors, and visualize behavior data by combining POS data of the restaurant. We collaborated with QC circle of the restaurant, and supported its members to find issues in their ordinary processes by visualizing behavior and POS data. They were able to make plans for improving their processes. We confirmed effects of the plans as the stay ratio of employees in dining areas, which was defined as a KPI by circle members. We also confirmed an side effect of the plans as the number of additional orders increased in the

    second term. For evaluating the workload, there were no significant differences on walking distance between two terms. Our future work is to improve the usability of the suite so that anyone who is in service industries can use our suite, and to create a feedback function for employees so that they can confirm their performance instantly. ACKNOWLEDGMENT This work was supported by the Ministry of Economy, Trade and Industry (METI) of Japan. The authors thank Ganko Food Service Co., Ltd., and all staffs in the Ganko Ginza 4-chome restaurant for their great cooperation. REFERENCES [1] The Council on Competitiveness, 2005, Innovate America:

    National Innovation Initiative Summit and Report (http://www.compete.org/publications/detail/202/innovate-america/, accessed on 2013-05-30).

    [2] Spohrer, J. and Maglio, P. P., 2008, The emergence of service science: Toward systematic service innovations to accelerate co-creation of value, Production and Operations Management, 17(3), 238-246.

    [3] Maglio, P. P., Srinivasan, S., Kreulen, J. T., and Spohrer, J., 2006, Service systems, service scientists, SSME, and innovation, Communications of the ACM, 49(7), 81-85.

    [4] Bullinger, H.J., Fähnrich, K.P., and Meiren, T., 2003, Service engineering: Methodical development of new service products, International Journal of Production Economics, 85(3), 275-287.

    [5] Miwa, H., Nakajima, M., Fukuhara, T., and Nishimura, T., 2013, Proposal of handing-over support system for nursing-care service with service engineering approach, In Y. Shimomura, and K. Kimita (Eds.), The Philosopher's Stone for Sustainability, Springer, 131-136.

    [6] Shimmura, T., and Takenaka, T., 2011, Improving restaurant operations through service engineering, Proceedings of International Conference on Advances in Production Management Systems (APMS).

    [7] Ishigaki, T., Takenaka, T., and Motomura, Y., 2010, Category mining by heterogeneous data fusion using PdLSI model in a retail service, Proceedings of the 2010 IEEE International Conference on Data Mining, 857-862.

    [8] Nishino, N., Fukuya, K., and Ueda, K., 2012, An auction mechanism considering seat reservations in movie theater services, In H. Sasaki (Ed.), International Journal of Organizational and Collective Intelligence (IJOCI), 2(1), 63-76.

    [9] Pigage, L. C., and Tucker, J. L., 1954, Motion and time study, In R.W. Fleming, and B.D. Dennis (Eds.), The University of Illinois Bulletin, 51(73) (available at http://hdl.handle.net/2142/9385, accessed on 2013-05-30).

    [10] Ray, P. S., 2005, Work sampling, In A. B. Badiru (Ed.), Handbook of Industrial and Systems Engineering, chapter 7, CRC Press, Boca Raton, FL.

    [11] Kurata, T., Kourogi, M., Okuma, T., Ishikawa, T., Ueoka, R., Tenmoku, R., and Makita, K., 2012, Human-behavior sensing and visualization for service quality control, Proceedings of CSCW 2012 Workshop: Exploring Collaboration in Challenging Environments: From the Car to the Factory and Beyond (available at http://workshops.icts.sbg.ac.at/cscw2012/, accessed on

    Fig.11 Number of reservations through a year (from analysis of POS data of the restaurant through 2008 to 2013).

    200

    300

    400

    500

    600

    700

    Month

    Num

    ber o

    f res

    erva

    tions

    Jan(n=6)

    Feb(n=6)

    Mar(n=5)

    Apr(n=4)

    May(n=4)

    Jun(n=4)

    Jul(n=4)

    Aug(n=4)

    Sep(n=4)

    Oct(n=4)

    Nov(n=4)

    Dec(n=4)

    7

  • 2013-08-01). [12] Kimes, S. E., and Mutkoski, S. A., 1991, Assessing

    customer contact: Work sampling in restaurants, Cornell Hotel and Restaurant Administration Quarterly, 32(1), 82-88.

    [13] Shimizu, S., Tomizawa, R., Iwasa, M., Kasahara, S., Suzuki, T., Wako, F., Kanaya, I., Kawasaki, K., Ishii, A., Yamada, K., and Ohno, Y., 2011, Nursing business modeling with UML: From time and motion study to business modeling, In A. B. Eldin (Ed.), Modern Approaches To Quality Control, chapter 22, InTech.

    [14] Kimes, S. E., 2004, Restaurant revenue management, Cornell Hospitality Reports, 4(2).

    [15] Bertsimas, D., and Shioda, R., 2003, Restaurant revenue management, Operations Research, 51(3), 472-486.

    [16] Smith, B. C., Leimkuhler, J. F., and Darrow, R. M., 1992, Yield management at American Airlines, Interfaces, 22(1), 8-31.

    [17] Kimes, S. E., 1989, Yield management: A tool for capacity-considered service firms, Journal of Operations Management, 8(4), 348-363.

    [18] Belobaba, P. P., 1989, Application of a probabilistic decision model to airline seat inventory control, Operations Research, 37(2), 183-197.

    [19] Takenaka, T., Ishigaki, T., and Motomura, T., 2011, Practical and interactive demand forecasting method for retail and restaurant services, Proceedings of International Conference on Advances in Production Management Systems (APMS).

    [20] Burger, T., Kim, K. J., and Meiren, T., 2009, Visualizing and testing service concepts, Proceedings of the First International Symposium on Services Science, 149-159.

    [21] Simo, R., Satu, M., Essi, K., and Antti, L., 2012, A laboratory concept for service prototyping: Service innovation corner (SINCO), Proceedings of ServDes 2012: Service Design and Innovation Conference.

    [22] Hyun, J., Habuchi, Y., Park, A., Ishikawa, T., Kourogi, M., and Kurata, T., 2010, Service-field simulator using MR techniques: Behavior comparison in real and virtual environments, Proceedings on International Conference on Artificial Reality and Telexistence (ICAT2010), 14-21.

    [23] Inoue, Y., Sashima, A., and Kurumatani, K., 2009, Indoor positioning system using beacon devices for practical pedestrian navigation on mobile phone, In D. Zhang, M. Portmann, A. H. Tan, and J. Indulska (Eds.), Proceedings of the 6th International Conference on Ubiquitous Intelligence and Computing, 251-265.

    [24] Sumi, Y., Yano, M., and Nishida, T., 2010, Analysis environment of conversational structure with nonverbal multimodal data, Proceedings of International Conference on Multimodal Interfaces and the Workshop on Machine Learning for Multimodal Interaction, 44:1-44:4.

    [25] Choudhury, T. and Pentland, A., 2003, Sensing and modeling human networks using the Sociometer, Proceedings of the 7th IEEE International Symposium on Wearable Computers, 216-222.

    [26] Kim, T. J., Chu, M., Brdiczka, O., and Begole, J., 2009,

    Predicting shoppers' interest from social interactions using sociometric sensors, Extended Abstracts of 27th Annual CHI Conference on Human Factors in Computing Systems (CHI 2009), 4513-4518.

    [27] Olguín, D. O., and Pentland, A. S., 2009, Sensible organizations: A sensor-based system for organizational design and engineering, Proceedings of the International Workshop on Organizational Design and Engineering (IWODE09) (available at http://iwode09.ist.utl.pt/~iwode09.daemon/doku.php?id=program, accessed on 2013-05-30).

    [28] Olguín, D. O., and Pentland, A., 2010, Sensor-based organisational design and engineering, International Journal of Organisational Design and Engineering, 1(1/2), 69-97.

    [29] Ara, K., Kanehira, N., Olguín, D., Waber, B. N., Kim, T., Mohan, A., Gloor, P., Laubacher, R., Oster, D., Pentland, A. S., and Yano, K., 2008, Sensible organizations: Changing our businesses and work styles through sensor data, Journal of Information Processing, 16, 1882-6652.

    [30] Ara, K., Akitomi, T., Sato, N., Takahashi, K., Maeda, H., Yano, K., and Yanagisawa, M., 2012, Integrating wearable sensor technology into project-management process, Journal of Information Processing, 20(2), 406-418.

    [31] Huang, Z., Nagata, A., Kanai-Pak, M., Maeda, J., Kitajima, Y., Nakamura, M., Aida, K., Kuwahara, N., Ogata, T., and Ota, J., 2012, Posture study for self-training system of patient transfer, IEEE International Conference on Robotics and Biomimetics (ROBIO), 842-847.

    [32] TRU-SAM (Truck Support, Administration and Management) [Computer system]. Toward Logistics Co., Ltd. (available at http://www.tru-sam.jp/, accessed on 2013-05-30) (in Japanese).

    [33] Ueoka, R., Shinmura, T., Tenmoku, R., Okuma, T., and Kurata, T., 2012, Introduction of computer supported quality control circle in a Japanese cuisine restaurant, Proceedings of the 4th International Conference on Applied Human Factors and Ergonomics (AHFE), 6632-6641.

    [34] Ishikawa, T., Okuma, T., and Kurata, T., 2009, Interactive indoor 3D modeling from a single photo with CV support, Proceedings of the 3rd International Workshop on Ubiquitous Virtual Reality (IWUVR2009).

    [35] Kourogi, M., Kurata, T., and Ishikawa, T., 2010, A method of pedestrian dead reckoning using action recognition, Proceedings of the Position Location and Navigation Symposium (PLANS), 2010 IEEE/ION, 85-89.

    [36] Ishikawa, T., Kourogi, M., Okuma, T., and Kurata, T., 2009, Economic and synergistic pedestrian tracking system for indoor environments, International Conference of Soft Computing and Pattern Recognition (SOCPAR '09), 522-527.

    [37] Ishikawa, T., Kourogi, M., and Kurata, T., 2011, Evaluation and application of service-worker tracking system in real shop floors, Transactions of the VR Society of Japan, 16(1), 23-34. (in Japanese)

    8