Method for analyzing sequential services using EEG: Micro-meso analysis of emotional changes in real flight service

Capturing customers ’ emotional changes in sequential service should be realized using physiological measurements to assess customer delight. Questionnaire-based customer surveys may miss significant and dissipating emotional responses. This study developed a micro ‑ meso analysis method of capturing emotional changes for sequential service using electroencephalograph (EEG) measurement, dealing with both service encounters (micro-level) and servicescape (meso ‑ level) over a couple of hours. Customers ’ emotion states were defined based on emotional arousal and valence. Emotional responses caused by human interactions were evaluated


Introduction
To be successful and stay ahead of competitors, delighting customers by providing outstanding customer experiences that elicit positive emotions is essential [1].For more than two decades, marketing literature has debated the definition of customer delight.The dominant view is that customer delight represents a distinct construct resulting from high satisfaction and is a deeply positive emotional state consisting of joy and surprise [2][3][4].Therefore, appealing to customers' emotions is vital to customer delight.Based on this understanding, the recent literature on customer delight has investigated the necessary components of the construct.For example, whether surprise is a mandatory element of customer delight [4].The present study adopted a definition of customer delight consisting of joy and surprise, with surprise being regarded as an amplifier.The antecedents of customer delight can be divided into three overarching areas-customer-related, employee-related, and contextual factors [4].
Customer delight positively influences repurchase intentions, wordof-mouth, and loyalty to the provider [1,4].Additionally, it benefits employees interacting with delighted customers, enhancing their customer orientation, job skills, and work engagement.Prior survey studies on delight show the robustness of these effects and their consistency across service delivery modes, including face-to-face, self-service, and online [4].Surveys based on questionnaires in those studies remain effective, but they lack the capability to explore the dynamic nature of emotions during the customer experiences [5].
The present study targeted sequential services over a couple of hours that involved multiple interactions between service providers and customers (i.e., service encounters as employee-related factors) in a changing service setting and physical environment (i.e., servicescape [6] as contextual factors).Flight service is a typical example of such sequential services where customer emotions fluctuate.Besides the average emotional evaluation, the peaks are also important [7].However, customer surveys examining emotional responses are generally conducted after the entire service via questionnaires.Such surveys often miss key emotional responses and feedback [8] since they only analyze predefined service contents and lack capturing immediate emotional responses (see the upper part of Fig. 1).Additionally, cognitive responses can dissipate over time, and interaction details are likely to be forgotten.The longer the customer experience, the more pronounced this problem becomes.The same issue can also be addressed on servicescape.Questionnaires may not accurately distinguish the effect of the servicescape on emotions.
Therefore, capturing customers' emotional responses in sequential service through time-series physiological measurements is a big challenge in service marketing and design (e.g., [5,9,10]).Customer journey mapping is a commonly used tool in these areas and often includes visualizing an emotional curve similar to that in the center of Fig. 1.However, in practice, most of these curves represent assumed and inferred emotions from planning or observation, not direct measurement.Both emotional changes caused by each service encounter and servicescape need to be considered.According to the viewpoint of aggregated levels of interactions and institutions in a service ecosystem [11], in the present study, service encounter is regarded as micro-level and servicescape as meso-level.Therefore, a hybrid of micro-level analysis and meso-level analysis for servicescape is required.
This study explored capturing the emotions of passengers during an actual flight service, which fluctuates over flight phases, such as during takeoff and during the cruise, and interactions with the flight attendant.In-flight environments as servicescape may also change according to the flight phase.Previous studies on service quality, satisfaction, and loyalty in passengers' flights [12][13][14][15][16][17][18] primarily analyzed post-flight questionnaires and did not measure changes in emotions during the flight.A previous study on the transition of passenger comfort and emotions in each flight phase [19] also used questionnaires and interviews, and different passenger activities during the cruise phase were considered.
Theories of emotion are mainly based on cognitive psychology, with contributions from physiological psychology and other disciplines [20].Russell's dimensional model [21] is a well-known model of emotion (Fig. A1).This model, named the circumflex model of affect, provides a two-dimensional map of coordinates by the degree of valence and arousal.Emotions are distributed on the map in terms of pleasure and arousal.This dimensional model is often used in consumer behavior studies as it has high explanatory power [22].Especially, "high positive affect" (i.e., positive valance with high arousal) in the map is relevant to joy and crucial for customer delight.
Measurement of emotional states that are not self-reported can be classified into four categories: autonomic nervous system measures, startle response measures, behavioral measures, and brain activity measures [23].Among the brain activity measures, the number of research studies in neuromarketing using electroencephalograph (EEG) devices has increased due to their cost efficiency, high temporal resolution, and mobility advantages [24,25].Non-invasive brain signal recording techniques are used in neuromarketing to directly measure the response of a customer's brain to the marketing stimuli, replacing traditional survey methods [26].
Studies using EEG in neuromarketing to analyze consumers largely focus on their reactions to marketing approaches, decision-making [27], and device quality.TV commercial engagement or success analysis are representative applications [24].The demand for acquiring and analyzing EEG signals in real-time is increasing as techniques on emotion analysis are being applied to various domains, such as automobiles, robotics, healthcare, and customer-support services [28].Additionally, several studies have examined the applications of EEG, such as mental workload [29,30], task concentration [31], psychiatric health [32], and prediction of functional outcomes after stroke [33,34].EEG is also used in quantifying physiological biomarkers of microwave brain stimulation devices [35].EEG is suitable for naturalistic studies as many EEG systems are portable and can be combined with other measurement tools [5,36,37].Verhulst et al. [5] discussed the potential of neuroscientific methods in the service field, focusing on a single touchpoint.Mobile EEG measurements are practically expected to analyze long-duration customer experiences in actual services by being less burdensome for test subjects (i.e., customers).However, there is a Fig. 1.Micro-meso analytic method of emotional changes during a customer journey.

T. Hara et al.
paucity of studies using EEG in real-world experiences.Fewer studies cover experiences of long duration.
In the present study, we aimed to develop a micro-meso analysis method to measure emotional change in sequential services over a couple of hours.We conducted an experiment on actual flight service using mobile EEG measurement.This method analyzes emotional changes at service encounters (i.e., micro-level analysis) and through servicescape (i.e., meso-level analysis) using the obtained EEG data.Fig. 1 illustrates the difference between these two analytic approaches on a customer journey.For comparison, the upper part of the figure shows self-reported and limited emotional responses by a customer through a questionnaire survey.The micro-level analysis verifies immediate emotional responses caused by interactions, resulting in (a) assuring self-reported positive emotions and (b) detecting unreported positive emotions.The meso-level analysis visualizes the duration of a significant emotional state called "high positive affect" during a customer's journey regardless of direct interactions and contributes to (c) detecting unreported positive emotions, including the effect of the servicescape.We assume that single-channel EEG measurements enable the integrated analysis of micro-level and meso-level approaches for real-world services.
The present study aimed to supplement existing literature on the interactions between physiology (EEG), behavior (emotional changes), and customer service research.The contributions of the present study include: • Succeeded in emotion estimation using single-channel EEG measurement over a couple of hours during the actual service where questionnaires are the mainstream methods.
• Developed an integrated method that verifies immediate emotional responses by human interactions (i.e., service encounters) and visualizes emotional peak periods throughout the customer journey using the same EEG measurement data.The visualization includes effects by servicescape and unreported positive emotions by customers.
• The analytic method contributes to empirical studies on sequential services in marketing and the design field by enabling the extraction of "high positive affect," which needs to be identified for customer delight.

Estimation of emotion by single-channel EEG
Customers' emotional states in this study were defined based on Russell's dimensional model.Customers wore a hairband-type EEG device during a target customer journey, and the valence and arousal of emotional reactions were estimated based on the measured EEG.Among mobile EEG devices such as Emotiv Epoc and NeuroSky Mindwave, the present study employed a NeuroSky device with one front biosensor, not requiring any conductive medium to be applied on the test subject's scalp.Given that our experiments were conducted for several hours in a real-world service environment with other customers present, the device's discreet appearance and impact on the wearer's (i.e., test subject's) experience [38] were crucial considerations.
The data-processing of signals and emotion estimation software using the single-channel EEG device was provided by Dentsu Science-Jam Inc.The authors of this paper did not develop the software.Using the EEG device, Dentsu ScienceJam Inc. and scholars at Keio University have been working extensively on measuring and data processing for particular elements of human emotions or states of mind.They have established a set of procedures to denoise and remove artifacts, including blinks, body movements, and electrical noise, in real-time with low computation costs [39,40].Another research group has also independently verified that the device could reliably remove environmental noise and unintended frequencies [41].
The provided estimation software is based on the model developed by the machine learning algorithm [42], which tested the EEG device for emotion estimation in a mobile application.The methodology encompassed filtering, feature extraction, and pattern recognition to infer emotional valence from EEG readings.Finite impulse response (FIR) and Butterworth infinite impulse response (IIR) filters were used for artifact reduction, while the data were transformed into power spectrums through the Fast Fourier Transform (FFT).Following the application of standardization and robust scaling techniques, a moving average of the scaled data was calculated to consider emotional changes over time, spanning from one to 30 s. Optimized Hz data were extracted from power spectrums by classifying frequencies and applying a genetic algorithm (GA).Regression (linear and support vector regression [SVR]) and classification (support vector machine [SVM]) and k-nearest neighbor [k-NN]) were incorporated for valence estimation.Different combinations of methods and parameters were rigorously tested using a grid search, and a 10-fold cross-validation ensured model robustness, preventing overfitting.The prediction accuracy of the valence score was over 70 % when classified as high or low despite having only one sensor.Similarly, another model for arousal was developed [42].Their prediction accuracy has been improved through parameter tuning in practice as a commercial product [43].
Scores of valence and arousal were output and recorded at intervals of 1 s with a value from 0 to 100.These are normalized values based on the result of calibrations before the start of EEG measurement to account for individual differences in EEG.
According to the two-dimensional space [44,45], the customer's state at a particular time was identified using the degrees of valence (v t ) and arousal (a t ).

Micro-meso analysis method to capture emotional change in sequential service
The proposed micro-meso analysis method consists of (A) detecting significant emotional responses caused by interactions and (B) visualizing the duration of emotional peak periods as high positive affect throughout the customer journey.
(A) Micro-level analysis: detecting significant emotional responses caused by interactions.
Data of v t and a t before and after interactions were used to determine whether their interaction type affects customer emotion at a service encounter level, as shown in the middle layer of Fig. 1.The data used were the mean of valence and arousal for 10 s before and after the occurrence of each interaction event (e k ).We have notated them as v be e k , v af e k , a be e k and a af e k .If an interaction event occurred at time t lasts and its duration is longer than or equal to 5 s (i.e., [t, t + (d − 1)], d ≥ 5), the 10 s-interval for the after data is set as A paired t-test (p < 0.05) of two groups: a before and after data set (e. g., ({v be e k }, {v af e k }) for valence) was performed.A significant difference between the two groups indicated that valence or arousal fluctuated according to the type of service encounter.
(B) Meso-level analysis: visualizing the duration of high positive affect throughout a customer journey In this study, we defined the following emotional state: • High positive affect (high valence and arousal levels, v t ≥ 75 and a t ≥ 75) These thresholds were set based on an emotion rating map using the Self-Assessment Manikin (SAM; e.g., [45,46]).Psychological research employs SAM [47], a non-verbal visual rating technique, to separately measure pleasure and arousal.For example, in a 9-point SAM test, considering low emotional intensity, a score of 7 on the valence and arousal scale indicated the minimum intensity of pleasure and surprise, respectively.After converting the scale to (0, 100), the "high positive affect" range became (75, 100), which was narrower than the four quadrants of emotion in Russell's dimensional model and corresponded to the emotion labels of "delighted" or "excited."High positive affect is a significant emotional state with peaks of better customer experience.
Intervals wherein high positive affect lasted for more than 5 s were extracted and visualized, as shown in the bottom layer of Fig. 1.The threshold of 5 s was set in consideration with noise detection.By summing the extracted intervals, the occurrence and density of high positive affect in customer journey could be calculated, including the effects by servicescape.

Experiment targeting actual flight 2.3.1. Experiment participants
With the cooperation of the airline company ANA (All Nippon Airways), measurement experiments were conducted in October 2020 on actual domestic flights.The research sample set consisted of 10 participants (7 men, 3 women), with an age range from 20 to 40 years (8 were in their 20 s).Based on the questionnaire administered at the time of application, 10 participants were selected from 30 applicants such that there would be no bias in passengers' familiarity with flights.This selection process was done using the questions exhibited in Table B1.
This study was conducted with the approval of the Ethics Review Committee of the Graduate School of Engineering, the University of Tokyo (KE20-28).Participants were paid an honorarium for their participation in the experiment.The amount was based on the number of hours of experiment.There was no incentive based on behavior during the flight.

Experimental procedure
Fig. 2 shows the experimental procedure for each participant.EEG measurements were conducted 20 times.Each participant flew the outbound and inbound flights on different days.We explained participants that the reason for wearing the mobile EEG device was to obtain EEG data on emotional fluctuations during the flight, in order to analyze the effects of the in-flight environment and customer service on these fluctuations.Participants were asked to board their return flights after their lunch break without going outside the airport.The team of flight attendants did not change between the outbound and inbound flights, but the flight attendants who served the participants were different in some cases.
First, prior to the flight, the EEG measurement device was attached to the participant's head and calibrated just before boarding.The calibration time was more than 75 s.During this calibration, individual differences among participants were considered.We asked the participants to spend their time as usual during the flight.However, we asked the participants to refrain from eating and using the bathroom during the flight as much as possible and not to be too absorbed in any one task, such as working on the computer, to avoid any interference with the EEG measurement.
The flight attendants involved in customer services were also asked in advance to "serve the participants as usual" during the flight.However, since contact with passengers on domestic flights is limited to basic duties, we asked them to adopt more than one approach to the participant during the flight.Here, approach includes not only actually serving the participant but also refraining from serving because of consideration.Videos around the participants were only recorded during the flight, including interactions with flight attendants.
In a questionnaire after each flight, the participant was asked to draw the emotional curve (Fig. 1) of the flight experience from boarding to disembarkation and note reasons or events at peaks of the curve.At the end of the experiment, we interviewed the participants based on the emotional curves of inbound and outbound flights.

Experimental flights and seats
We selected the same round-trip flight between Haneda and Kumamoto airports for this experiment based on three conditions: Although there were some changes in the flight schedule, the flight time was about 2 h.Only one participant could not take the same flight and took the flight between Haneda and Naha airports.
Participants were required to wear masks on board to prevent COVID-19 infection.Therefore, flight attendants had difficulties reading the needs of the passengers from their facial expressions; however, there was no obstacle to the measurement because the participants were being served.
The participants were seated in the last row of the cabin, and one of the authors was seated directly next to the participants to avoid interference with the general passengers and to cope with any measurement problems.There was no interaction between the participant and the author.The Hawthorne effect due to this [48] was not considered, as the participants were not asked for proactive actions and communication with flight attendants.
The seat pairs in the last row of the cabin were not available on the 3 out of 20 flights, as they had been reserved by the general passengers before the authors' booked their flight.Thus, the seat pairs for the participant and the author were replicated in other rows in the cabin on these three flights while maintaining the designed functions.

How the proposed method was applied to flight service
This section describes how the generic method described in Section 2.2 was applied to flight service.The elements of service encounter and servicescape in the flight service included: • Service encounters: direct services and in-flight announcements by flight attendants • Servicescape: in-flight environment and changing flight phases

Micro-level analysis of direct services and in-flight announcements
Direct services Direct services by flight attendants during the cruise were classified by coding [49], a qualitative data analysis method.We extracted direct service scenes in a service process from the videos and classified them by assigning a heading (code) to each service.The classification by coding comprised three steps: 1 Extracting the scenes in which flight attendants served the participants, 2 Coding each scene based on the verb associated with the action, and 3 Grouping of the same type of service as a category.
Using the coded service data, we obtained the mean of valence and arousal for 10 s before and after the start of each coded service.We assumed a single effect for each category, such as asking about a refill, evoking positive affect.
In-flight announcements after the initial climb Flight attendants made in-flight announcements after the initial climb.The micro-analysis method was applied for the data before and after the start of these in-flight announcement events.The announcements, including greetings, estimated arrival, turbulence forecasts, and seatbelt sign instructions, helped passengers anticipate seatbelt sign changes and drink service.While frequent flyers appreciated the arrival time, turbulence information was aimed at relieving less experienced passengers.Thus, we assumed that in-flight announcements after the initial climb would evoke positive affect.

Meso-level analysis across flight phases
The flight phases were divided into the following: during parking, from pushback to engine start, during taxiing, from the start of takeoff roll to takeoff, during the initial climb, during the climb, during the cruise, during the descent, during the initial approach, during the final approach, during landing, and from taxiing to stopping.According to these flight phases, positive emotional transitions were visualized and summarized in each phase.
Most previous studies on the comfort of the aircraft cabin environment considered seat related-factors, such as seat space [50,51].However, when considering pleasure and discomfort related to emotions in servicescape, airsickness due to aircraft-specific turbulence cannot be ignored.The weightlessness caused by vertical aircraft motion may cause discomfort in the human body.As a prerequisite, before analyzing the effect of servicescape in detail, we confirmed whether vertical aircraft motion due to turbulence called quick drops in the air happens and affects passenger emotions or not (Appendix C).

Measurement
The average measurement period of 20 flights was 7715 s.The average value and standard deviation of noise detection periods by the measurement processing software were 648.4 s and 626.6 s, respectively.These periods had missing measurement values, and we performed linear interpolation for them.
Fig. E1 summarizes the valence and arousal estimations that allow for the proposed meso-micro analysis considering flight phases and customer services.The measurements and estimations over a long period of time and organized according to the two-dimensional model of emotion were confirmed.

Duration of high positive affect in each flight phase (meso-level analysis)
The data of four flights, which had long unmeasured periods and could not be sufficiently correlated with the flight data, were excluded in the meso-level analysis.The bottom table in Fig. 3 shows the standard duration of each flight phase and the number of flights out of 16 in which high positive affect lasted at least once.For those applicable flights, the graph in Fig. 3 compares the average density of total duration of high positive affect for each flight phase.For example, during the initial climb, 8 out of 16 flights had high positive affect, which lasted more than 5 s at least once.Among the 8 flights, the average density of total durations was 0.25.If the duration of the phase is 60 s, as in standard, the average duration of high positive affect is 15 s in total.
These results were obtained from the analysis: • High positive affect could be seen across all flight phases, not only in service encounters during the climb, cruise, and descent.• In the period from the start of takeoff roll to takeoff, high positive affect tended not to be present in large populations, but when it did, it could last for a long time.• During the climb, cruise, and descent, the duration of which was longer than other phases, high positive affect was present in most participants at least once.The average density of 10 % in those long flight phases implied that most participants had multiple and lasting delight or excited experiences on board.
Fig. 4(A) shows the result for Participant 1 that visualizes service encounters and the duration of estimated high positive affect.The horizontal axis represents the elapsed time from starting EEG measurement.Since the flight recorder was turned on independently of the EEG measurement and boarding, the beginning of the elapsed time synchronized with recorded flight phases was 600 s in this case.The variety of direct services performed during the service encounters is presented in Table 1 and explained in Section 3.3.States of high positive affect were observed with and without the service encounters.Fig. 4(B) shows the selfreported emotion curve for Participant 1 after the experiment.The authors adjusted the time scale to contrast with Fig. 4(A) based on what was elicited from the interview with Participant 1 about the written emotion curve, including the events during the peak period.
Participant 1 reported a lasting pleasant state during the flight since it was not a daily experience.Thus, the written emotion curve was on the upper side of the baseline, and a few peaks of emotion change were identified.Contrarily, the measurement data showed a greater number of high positive affect.In this case, much of the high positive affect occurred in the servicescape and self-activities, not in service encounters.For example, many periods of high positive affect during the climb and cruise were found while Participant 1 was reading a book.These periods were not included in the self-reported emotion curve.Before disembarking, periods of high positive affect were observed several times through direct services of special goods provision and conversations.This is consistent with the self-report of being left with a lasting impression.
Other results in Appendix D show the following cases.
• Participant 2, whose self-reported positive emotional peaks were matched well with a subset of the estimated high positive affect.

Table 1
Direct service by flight attendant classified as a basic service.

Symbol Type Description Frequency (a) Alcohol sheet provision
Hand out an alcohol sheet to prevent COVID-19.

(b) Asking about a drink
Ask passengers what they would like to drink.

(c) Serving a drink
Pour the asked drink into a cup and serve the drink.

(d) Drink cup collection
Collect the cup when passengers finish the drink provided.
• Participant 3, whose self-reported positive emotional peaks in selfactivities were not confirmed by the objective measurement result, although several emotional peaks in service encounters were detected by the proposed method.• Participant 4 showed the highest proportion of positive affect.The prolonged dense sections revealed specific flight events and customer services noted in the self-report.• Participant 5, who was unfamiliar with flying, showed both large positive and negative emotional peaks.Positive emotions in the service encounters, which are fewer than the others, could be confirmed in the self-reports and measurements.

Differences among direct services
Flight attendants performed 113 direct service sessions in the experiment, which were coded into 11 types: 4 basic services and 7 individual services (Tables 1 and 2).Basic services were provided to all passengers on board, while individual services were offered only to passengers who needed them.
After dividing the data by each direct service code, the micro-level analysis was applied.A graph summarizing the mean and variance of valence and arousal before and after the direct service is shown in Fig. 5.
We examined whether there was a difference in valence and arousal before and after each type of direct service.As a result, there was a significant difference in arousal for (b) asking about a drink (p = 0.009) and (d) collecting the drink cup (p = 0.016).There was a significant difference in both valence and arousal for (f) asking about a refill (p = 0.030 and p = 0.005) and (i) conversations started by flight attendants (p = 0.037 and p = 0.002), indicating that the emotional effect tended to be higher.
The following were the results in terms of emotional change: • When (b) asking about a drink, arousal tends to increase, but valence is not evoked.• (f) Asking about a refill and (i) conversation started by flight attendants tend to cause high positive affect.• There was no significant difference in the change in valence and arousal in other direct services, and no crucial effect on emotional change was found.
In the case of (b) asking about a drink and (c) serving a drink, the mean of valence was lower, as shown in Fig. 5.When we checked both cases in the recorded videos, a huge decreasing valence (negative) was found when the participant was asked about a drink while reading a book, representing an interruption.This suggests that these two services may cause positive or negative affect.

In-flight announcements after the initial climb
A total of 20 in-flight announcements by flight attendants were made after the initial climb, one for each flight.As a result of applying the micro-level analysis, we found a significant difference in the change in valence by the in-flight announcements (p = 0.031).However, there was no significant difference in the change in arousal.Therefore, positive emotional change tends to be evoked.

Theoretical implications 4.1.1. Emotion estimation in real services and methodological contributions
In the present study, we achieved emotion estimation using singlechannel EEG measurement over a couple of hours in the actual service, where questionnaires are the mainstream method.The developed micro-meso analysis method processed the time series of estimated emotions.Experiments in actual flight service showed how the method captures emotional changes for sequential services.The meso-level analysis automatically quantified and visualized the "high positive affect" as significant emotional peak periods throughout the customer journey.This quantification and visualization included effects by servicescape and self-activities.Additionally, the micro-level analysis verified emotional responses during service encounters according to classifications of interactions by coding.The verification results revealed which classifications of interactions bring significant change in emotion (e.g., (f) asking about a refill and (i) conversations started by flight attendants).
Therefore, the analytic method contributes to existing empirical research and tools on sequential services in marketing by comprehensively evaluating emotional changes, reviewing existing touchpoints, and focusing on the critical touchpoints that leave a lasting impression on customers (i.e., "moment of the truth").The present study supplements the literature on the interactions among physiology (EEG), behavior (emotional changes), and customer service research.

Dynamics of customer delight: the interplay of joy and surprise
Two individual services of (f) and (i) were found to cause high positive affect, while basic services did not, indicating the effectiveness of individual service provisions suggested in the service excellence pyramid [1] even in dynamic and micro situations.For (k) special goods provision, another individual service, both self-reported and measured reactions during certain customer interactions were found during some individual customer journeys, although no significant differences were found in a paired t-test (Section 3.3.1).
In the present study, customer delight was framed around moments of high positive affect-instances of positive valence paired with high arousal.Surprise was regarded as an amplifier of delight.However, considering the results regarding the occurrence of high positive affect, emotional shifts, and self-reported feedback that were linked to individual services, we cannot deny certain situations where customer delight could be predominantly driven by surprise.This is because we did not deal with the sole effect of high arousal without positive valence.Thus, within the current research scope, we preliminarily analyzed the temporal dynamics between valence and arousal tied to high positive affect by individual services.
• For (f), responsive to timely drink refill needs, an immediate upswing in both valence and arousal tended to be observed post-service, as depicted in Fig. 6(A).

Table 2
Direct service by flight attendant classified as an individual service.

Symbol Type Description Frequency (e) Hand towel provision
Hand out a hand towel to passengers as they eat their meals.

(f)
Asking about a refill Ask passengers who seem to want a refill for another drink.

(g) Individuallywrapped straws provision
Provide individually-wrapped straws to passengers who are concerned about COVID-19 infection.

(h) Guide to baggage storage
Guide passengers to stow their baggage under the seat in front of them during takeoff and landing.

(i) Conversations started by flight attendant
Conversation with passengers started by the flight attendant, going back and forth one or more times.

(j) Conversations started by passenger
Conversation started by the passenger, going back and forth one or more times.

(k) Special goods provision
Give special goods other than the primary offerings such as cards and in-flight magazines.
• Longer interactions, notably (i) and (k), presented the patterns shown in Fig. 6(B) and (C)-an instant arousal increase followed by either partially delayed or subsequent valence amplification.These were observed, for example, when special goods were given, and then the customer appreciated it with an explanation received.• Fig. 6(D) showcases situations with prevalent high valence at service onset, with a subsequent arousal boost.This pattern, influenced by the passenger's own comfort or prior interactions such as those initiated by flight attendants, was also discerned in (i) and (k).
In short, surprise by unexpected moments was a driver of heightened arousal, particularly for (i) and (k) plays a role as antecedent, sustainer, or amplifier of positive valence.The present study contributes to research investigating the interplay of joy and surprise forms customer delight in sequential service dynamically.Note that in the subsequent valence pattern of Fig. 6(C), if the overlap between the two metrics was less than 5 s, the proposed method would not classify it as high positive affect.

Advancements in mobile EEG devices
With mobile EEG devices, the freedom of customer behaviors, interaction with service providers, and better wearing experience are maintained.The present study demonstrated the feasibility of measuring the emotional changes using mobile EEG devices in the context of actual services, including servicescape and human interaction.Another new approach of EEG measurement covering long-duration experiences can be stimulus presentation in immersive experiences in virtual reality(VR) [52,53], but research is lacking regarding how it differs from traditional stimulus presentation and for emotion classification using VR [52].

Physiological measurement and employee-related factors of flight service
Meyer et al. [54] analyzed physiological effects of flight conditions on vulnerable passengers by measuring their heart rate in simulated conditions, resulting in suggesting sympathetic arousal or reductions in parasympathetic drive.But only examined the level of arousal of the passengers.Few studies have analyzed passengers' valence changes during flights.EEG studies in-flight cases largely examine pilots' fatigue, emotions, and hypoxia rather than those of passengers.For passengers, questionnaires were still used to depict the transition of comfort and emotions (e.g., [19]).The present study succeeded in estimating the occurrence and density of high positive affect for passengers during flight services using a single EEG measurement, which have not been investigated in previous studies.
One category of antecedents for customer delight is employeerelated factors described in Section 1. Sezgen et al. [55] used text mining to clarify that the friendliness and helpfulness of flight attendants affect passenger satisfaction.Fukushima et al. [56,57] described the cognitive competencies of flight attendants behind such individual customer services, distinguishing between skilled and new flight attendants.The proposed method may enhance these early studies of the front-line staff by measuring the impact of these competencies from a customer viewpoint in a sequential service.The concept of "service not to serve" presented by Fukushima et al. [57] can be further analyzed using a dynamic understanding of emotion estimation and consideration, similar to that discussed in this section and Fig. 6.

Enhancement of customer behavior studies
In Participant 1, we identified periods of high positive affect during the climb and cruise, which were not captured in the self-reported emotion curves.Our method contributes to customer behavior studies by detecting unreported positive emotions, encompassing servicescape effects.In contrast, self-reported emotions may only partially show up in measurement results due to the limited scope of state estimation analysis.
The results explained in Section 3.3.2demonstrated a significant valence increase during in-flight announcements among participants.Interestingly, they were not standout events in the self-reported emotion curve.Our study verified this unconscious shift toward positive emotion.
The single axis-based method for emotional curves in customer journey maps (Fig. 1) struggles to distinguish between "positive affect" and "high positive affect" in Russell's dimensional model.It may result in overlooking outstanding customer experience and customer delight.Our study advances customer journey mapping and service design [58] by enabling long-term, real-time SAM-equivalent measurements, promoting customer delight.This contribution is supported by the results for Participant 2 in Fig. D1, whose continuous physiological measurement of emotion includes discrete self-reported responses.

Extensions to other sequential services beyond flight service
The present study primarily investigated flight services within domestic economy class, which served an undetailed majority of passengers simultaneously.The implications of our findings could be extended to a broad array of sequential services beyond flight service.Acquiring additional customer information beforehand allows service providers to tailor their offerings, especially during service encounters.The proposed method is more applicable to a variety of services than the previous study [5] since it allows multi touchpoints and long-term duration analysis.
Chitturi et al. [59] highlighted that the primary nature of the service (hedonic or utilitarian) influences the role of surprise in evoking customer delight, with utilitarian services potentially benefiting more from surprise elements.This suggestion needs to be considered when we interpret the results and validate the mechanism of customer delight on different sequential services.

Customer characteristics in emotion estimation
While we should be careful about the characteristics of each person in emotion estimation through biometrics [60], it is essential as a first step to find common trends independent of customer characteristics quickly and within a reasonable cost.The present study succeeded in testing the empirically assumed emotional effects through interactions and confirmed that some of them hold regardless of customer characteristics.Based on this fundamental understanding, detailed analysis to identify meaningful customer characteristics associated with a good customer experience [61] is expected.

Practical implications
We provided practical suggestions to enhance in-flight service, including appropriate timing, by objectively verifying the emotional effects of individual direct service.For instance, offering customer services such as "a conversation started by flight attendants before the seatbelt sign is turned on during descent" can relieve negative feelings and positively affect passengers.In addition, identifying moments when passengers are experiencing positive emotions and offering them a slightly unexpected, yet personalized touch can be effective.Such an approach, including "special goods provision," can foster customer delight by interweaving joy and surprise, as illustrated in Fig. 6(D)."Asking about a drink refill" works effectively with current service skills.Other individual direct services need to be further investigated by considering other customer characteristics, such as preferences related to direct service on-board.Basic direct service of asking about a drink is inevitable but should be carefully considered because it may deteriorate positive valence if passengers are absorbed in other things.In-flight announcements during the initial climb do not evoke customer delight directly but tend to bring unconscious positive change based on emotional valence regardless of flight familiarity.
Service firms can enhance the competitiveness of their businesses by finding the moment of positive emotions for customer delight (i.e., high positive affect) using physiological measurement.The present study took the measured data home after experiments and analyzed it in the laboratory.In the future, a quick visualization of the results of the analysis on-site will help marketers (or service designers) perform complemental interviews with participating customers immediately focusing on periods and events about significant measured emotional changes.
Flight events or customer service did not significantly differ between outbound and inbound flights.However, due to Haneda Airport congestion, outbound departures, and inbound arrivals have marginally longer waits and more self-reported negative feelings than their counterparts (Figs. 4, D1, D3).Participants inexperienced with flying often had negative emotional responses to turbulence and airsickness during their first outbound trip (Fig. D4).While the current method and visualization did not address "high negative affect" and thus did not show distinct differences, extending the method to consider negative affect might reveal how they overshadow positive experiences and delight.

Limitations
The study has several limitations to acknowledge.First, this study was limited by its relatively smaller sample size.Similar concerns are often raised about the use of neuroscience technologies in consumer behavior research.However, the basic biological composition of the brain is more consistent than other behavioral variables [25].Given this study's pioneering nature and the cost of analyzing long-duration EEG measurements in a real-world service environment compared to other studies [5,45,53], this sample size is reasonable for laying the groundwork for more extensive experiments in this area.
During the experiment for one participant, the EEG measurements were occasionally interrupted due to the strong influence of noise caused by physical characteristics and movement habits of the participant.Prior to the main experiment and participant selection, a preliminary assessment should be conducted to check the suitability of EEG measurement for each participant.Additionally, the results of measurements during the preliminary assessment in the daily conditions can be used as a baseline for the analysis in the experimental conditions for each participant.
Regarding the 10 s interval of interaction events in the microanalysis method explained in Section 2.2, its duration may need adjustments in practice for different services.It may depend on how service events elicit emotional responses at a touchpoint level.Considering the interplay of valence and arousal described in Section 4.1.2is also relevant to this issue.
During the experiments, both participants and cabin attendants wore masks, which may have influenced and hindered the participants' positive emotions caused by emotional contagion.However, we suspect that this effect was limited in the present study.The ability of customers (i.e., participants) to interpret employee emotions plays a role in this process.Eggenschwiler et al. found that customers can decode employee smiles even behind masks [62].Notably, smiles behind mask amplify perceived warmth, and mask wearing can covers negative emotions.The cabin department of the airline organization recognized that masks made communication harder but found that emphasizing emotions through voice tone and eye expression helped.This approach was shared throughout the organization, and its effectiveness is confirmed by the result presented in Fig. D4.

Conclusion
In this study, we successfully conducted emotion estimation using single-channel EEG measurement over a couple of hours in the actual service, where questionnaires are the mainstream method.The micro-meso analysis method of emotional changes verifies immediate emotional responses by human interactions and visualizes emotional peak periods throughout a customer journey.The visualization includes effects by servicescape and unreported positive emotions.Compared with questionnaire-based evaluation, the developed method contributes to empirical studies on sequential services in marketing and design by enabling extracting "high positive affect," which needs to be identified for customer delight.

Appendix B
We recruited individuals interested in participating by posting on social networking sites.Therefore, students from the author's laboratory and people familiar with the research did not participate.
Table B1 represents questions used for selecting 10 participants from 30 applicants.Of these, five participants were selected for being familiar with flights, who answered "yes" to 3 or more questions in the category "knowledge of air travel" and 4 or more in the category "knowledge of behaviors during flight."Five participants whose scores were lower than those described were selected as being unfamiliar with flights.During this selection process, the reported frequency of flying and gender balance were also considered to try to avoid bias within each group.
Three participants in the unfamiliar with flying group scored the lowest, with an overall score of 4. One participant had never flown before, and another (Participant 5 in Appendix D) had flown only once before in her early childhood.

Table B1
Questions to determine the familiarity with flights (developed with the airline company).

Category
Question Answer Knowledge of air travel I know that certain baggage is not allowed as carry-on baggage.Yes/No I know that the door used for passenger boarding and disembarking is on the left side of the aircraft in the direction of travel.
Yes/No I can locate my assigned seat without asking the flight attendant for directions.
Yes/No I know where the lavatories are located in the cabin.
Yes/No Knowledge of behaviors during flight I know that I can use any of the storage compartments if they are available.Yes/No I know that baggage can be placed under the seat in front of me, and not just in the compartment above.
Yes/No I know which side Mt.Fuji is visible from (the left or right window) on the flights I frequently ride.
Yes/No I know that seat belts need to be fastened while seated.
Yes/No I know that there is a signal that all doors are closed. Yes/No

T
. Hara et al.

Fig. 3 .
Fig. 3. Number of flights in which high positive affect lasted and their duration density in each flight phase.Error bars represent standard deviation.The shade of the graph reflects the grade of the number of applicable flights.

Fig. 4 .
Fig. 4. Result of visualizing the duration of high positive affect in each flight phase (participant 1, inbound).(A) measurement and (B) self-reported emotional curve.(B) represents the adjusted timescale of each flight phase for alignment with (A).

Fig. 5 .
Fig. 5. Means of valence and arousal before and after direct services by flight attendant.(A) valence and (B) arousal.Direct services (g) and (h) were not included because n = 1.

Fig. 6 .
Fig. 6.Observed patterns of the interplay between valence and arousal due to a service event.(A) simultaneous (B) delayed valence, (C) subsequent valence and (D) additional arousal.

Fig. D1 .
Fig. D1.Results of visualizing the duration of high positive affect in each flight phase (Participant 2, outbound).(A) measurement and (B) self-reported emotional curve.(B) represents the adjusted timescale of each flight phase for alignment with (A).

Fig. D2 .
Fig. D2.Results of visualizing the duration of high positive affect in each flight phase (Participant 3, inbound).(A) measurement and (B) self-reported emotional curve.(B) represents the adjusted timescale of each flight phase for alignment with (A).

Fig. D3 .
Fig. D3.Results of visualizing the duration of high positive affect in each flight phase (Participant 4, inbound).(A) measurement and (B) self-reported emotional curve.(B) represents the adjusted timescale of each flight phase for alignment with (A).

Fig. D4 .
Fig. D4.Results of visualizing the duration of high positive affect in each flight phase (Participant 5, outbound).(A) measurement and (B) self-reported emotional curve.(B) represents the adjusted timescale of each flight phase for alignment with (A).Appendix E Fig. E1 shows 16-quadrant emotion heatmaps of measurement data for all experiments.Fig. E1(A)-(J) represents the results of the outbound and inbound flights for each participant in a 16-quadrant emotional map.The state of high positive affect is mapped in the upper left quadrant.The scales of heatmaps are measured seconds of each emotional state, which vary by participants.Participants 1 -5 match the participants shown in Fig. 4 and Figs.D1-D4.

Fig. E1 .
Fig. E1.Emotion heatmap of all measurement data of valence and arousal.(A)-(J) represent the results of the outbound flight (left) and inbound flight (right) for each participant.The scales of heatmaps are measured seconds of emotional states.