Cloud-based Conversational Agents for User Acquisition and

Engagement

Manas

es Jes

us Galindo Bello

Cumulocity IoT Core R&D, Software AG, Germany

Keywords:

Dialog Systems, Chatbots, Cloud Computing, Artiﬁcial Intelligence, Natural Language Processing.

Abstract:

The beneﬁts of cloud computing have driven different companies from diverse sectors to migrate their products

and services to the cloud. In the last decade many businesses have adopted web and mobile applications to

offer better customer service as well as used social networks for advertisements and marketing campaigns

aiming to acquire, engage and retain their customers. This paper presents a case study combining the areas

of chatbots, cloud computing and customer service, acquisition and engagement targeting the gastronomy

industry; it evaluates and compares the implementation of a chatbot as a cloud-native application (Platform

as a Service) versus one built utilizing an authoring tool (Software as a Service); and it demonstrates how a

gastronomic business could attract with ease new customers by interacting with them using chatbots embedded

into instant messaging apps.

1 INTRODUCTION

Cloud computing has experienced a signiﬁcant

growth over a short period of time and companies

have implemented and/or migrated their products and

services to the cloud. With the addition of cognitive

services and natural language processing capabilities,

cloud providers are currently offering a broad catalog

of services that can be used to implement a plethora of

intelligent applications such as conversational agents

also known as chatbots. In the beginning there was the

computer and web interfaces, then smartphones came

along and mobile interfaces were introduced. Chat-

bots are going to revolutionize the software industry

likewise web and mobile technologies did. Chatbots

are a user interface that is changing the way people in-

teract with machines. Every day people perform sim-

ple tasks such as searching for images on a text-based

chatbot as well as using voice commands requesting

to play one’s favorite song or asking for navigation

instructions. Whether aware or not, people are al-

ready immersed in the era of chatbots. Companies

know that to achieve success they must offer excellent

customer service in a variety of channels, and chat-

bots are another way to reach the customers and pro-

vide them a 24/7 service. Communicating with cus-

tomers, showing empathy, being reliable at all times

and responsive to their needs are essential to create

customer loyalty (Parasuraman et al., 1991).

1.1 Problem Description

Something that has been seen over the years is that

technology can improve customer service and gastro-

nomic companies have implemented web systems and

mobile apps to achieve this goal. However, most of

the times such software brings high costs to the com-

panies as the user interface (front-end) as well as the

business logic (back-end) have to be developed, de-

ployed and maintained. In the majority of the cases,

gastronomic businesses opt to acquire such software

from third parties and thus save some implementation

costs. Nowadays people use mobile apps to locate

restaurants and check their details to make reserva-

tions. Also, users are bound to install distinct apps

on their smartphones for ordering pizza, coffee, tacos,

the groceries, and the list continues. It can be annoy-

ing to create a user proﬁle for each and every app

and add the payment method afterwards. It should

not be overlooked that smartphones do not have in-

ﬁnity storage and many users have faced the problem

of not being able to install apps due to a lack of stor-

age space. In the other hand, chatbots do not need

a particular graphic user interface when they get ac-

cessed inside messaging apps, thus keeping the same

user experience of the platform and messaging app

their run on. So, could a gastronomic business im-

prove its customer service by adopting a chatbot in

order to acquire customers and keep them engaged?

528

Bello, M.

Cloud-based Conversational Agents for User Acquisition and Engagement.

DOI: 10.5220/0007766105280534

In Proceedings of the 9th International Conference on Cloud Computing and Services Science (CLOSER 2019), pages 528-534

ISBN: 978-989-758-365-0

1.2 Objectives

Cloud computing and software engineering practices

shall be followed to develop two versions of a chatbot:

one from scratch deployed as a cloud-native applica-

tion on a Platform as a Service, and another utilizing

a chatbot-building tool (Software as a Service) with

the purpose of evaluating the beneﬁts and drawbacks

of each approach, and determine which option would

be more suitable for a gastronomic business. It shall

be analyzed whether it is feasible or not for a gas-

tronomic business to acquire and engage customers

by the adoption of a chatbot. This shall be done em-

ploying current customer acquisition and engagement

techniques in combination of a chatbot implemented

for one of the most popular messaging apps. The main

functionality of the chatbot shall be to provide rele-

vant information about a gastronomic business, e.g. a

restaurant or coffee shop. This information shall be

its location, opening hours and the menu.

1.3 Research Statements

In accordance with the foregoing objectives, it is pos-

sible to hypothesize that:

1. A restaurant can opt to use chatbots to offer a bet-

ter customer service providing relevant informa-

tion such as the menu and opening hours.

2. Customers may get a satisfying experience inter-

acting with a chatbot and getting the needed infor-

mation promptly. Hence, they are willing to come

back and recommend the place to others.

3. A restaurant may acquire new customers and keep

them engaged by means of a chatbot.

4. A chatbot deployed on a Platform as a Service

may offer better performance and scalability than

a chatbot-building tool (Software as a Service).

2 RELATED WORK

Researching about cloud computing, customer ser-

vice and conversational agents is not new. These

ﬁelds have been widely discussed over the past decade

– in both academic and practitioner literature – due

to new methodologies and technologies that have

emerged to revolutionize customer service through

social networks and conversational agents available

day and night thanks to the power and beneﬁts of the

cloud. The effectiveness of traditional advertising and

the insights generated through business-to-consumer

messages on Facebook were analyzed (Vries et al.,

2017) in a research for brand-building and customer

acquisition attempts. “Can machines think?” was the

question that drove Turing to propose a new sort of

problem that could be described in terms of a test

called the imitation game (Turing, 1950). Turing in-

troduced the idea to develop human-machine interac-

tions in 1950. From 1964 to 1966 Weizenbaum devel-

oped a text-based computer program named ELIZA

(R. Rising et al., 1978) with which one could establish

a conversation in the English language. Since then,

conversational agents have continuously been the sub-

ject of research and controversial debates concerning

moral aspects such as the fear of reducing the value

of human interaction and communication, the limita-

tions of dialog systems compared to real natural lan-

guage interactions between humans, and others. Con-

versational agents, dialog systems, virtual assistants

and chatbots are terms that have been used indistinctly

in different papers but referring to the same concept

where the basic idea is to establish a conversation

with a machine in a dialogical way using natural lan-

guage. Several researchers (Fadhil, 2018; Brandtzaeg

and Flstad, 2017; Shawar and Atwell, 2007) have ana-

lyzed how useful chatbots can be and why people use

them, ﬁnding that the major reasons are productiv-

ity, entertainment and curiosity. During 2000s several

companies implemented chat services on their web-

sites for customer service using chatbots in order to

reduce the costs of having a call center and the bene-

ﬁts of adopting chatbots for live support were evalu-

ated (Braun, 2003). Predictions have been made (Or-

acle, 2018) stating that 75% of cloud technology sup-

pliers will enrich their current products with artiﬁcial

intelligence and machine learning capabilities; chat-

bots will turn out to be highly specialized and with the

capacity to learn and collaborate with other chatbots,

and by 2020 the dominant part of customer service

will be led by chatbots.

3 FUNDAMENTALS

Though there is not a universal deﬁnition for cus-

tomer acquisition, one clear and simple is “the num-

ber of newly acquired customers per week” (Vries

et al., 2017). Customer service, acquisition and en-

gagement are interrelated areas always in a constant

cycle. Customers have to be acquired and companies

shall provide them constant quality service; moreover,

responding promptly to their needs is a way to main-

tain them engaged and loyal. Acquisition and engage-

ment techniques vary depending on the target market

and segment of customers; thus one technique that

works perfectly for one business may not necessarily

work for another. Before modern technology and so-

Cloud-based Conversational Agents for User Acquisition and Engagement

529

cial networks existed, one technique was used and it

is still being used in the present day: words-of-mouth.

In the gastronomy industry, it is impossible to adver-

tise the taste of a particular meal or drink, therefore

people tend to review online comments about a spe-

ciﬁc restaurant or to ask directly to friends that have

already visited such place. Analyzing online cus-

tomers’ behavior and shopping tendencies by means

of artiﬁcial intelligence has become a common prac-

tice of big companies aiming to build a customer en-

gagement ecosystem (Maslowska et al., 2016).

Chatbots are nothing more than computer pro-

grams in charge of performing speciﬁc tasks based

on user input in the form of a conversation

(Galindo Bello, 2018a). Chatbots are a new way to

expose software services through a conversational in-

terface (Shevat, 2017). Nevertheless, the chatbot is

not to be mistaken as the service itself as it is only the

interface to consume the service. In contrast to desk-

top applications and mobile apps, chatbots are not de-

ployed as native applications of speciﬁc platforms. In

that sense, chatbots are not installed in mobile mes-

saging apps rather connected to these apps through

an application programming interface (API). Figure 1

presents the most common architecture of a chatbot

exposing its services to a messaging platform that it

is used by messaging clients (apps or browser-based).

This type of architecture keeps secure the clients and

provides to chatbot developers a clean interface for

connecting to the platform. Users may install a mes-

saging app in different operating systems (OS) like

iOS or Android, and chatbot developers do not need

to worry about the client-side implementation as chat-

bots are OS-agnostic.

Artiﬁcial intelligence (AI) and machine learning

(ML) capabilities have been enhanced during the past

decades. Often people think that a chatbot is an arti-

ﬁcial intelligent program, but that does not hold true

for all chatbots. For instance, a chatbot can be imple-

mented to take as an input the name of a city and re-

turn the current weather of that city; AI is not required

in such case. Most elaborated chatbots can present the

menu of a restaurant with rich interactions for each

day of the week and grouped by the type of the meal.

Though the menu would create chat ﬂows, AI is not

required either. Chatbot developers would choose to

incorporate AI and ML capabilities depending on the

goal of the chatbot.

The process to recognize human speech uses nat-

ural language processing (NLP) to derive the user’s

input (also known as user intent) from the natural

language of the user (Shevat, 2017). NLP can be

integrated into voice-based and text-based chatbots.

Currently, some cloud providers offer cognitive ser-

vices and they include NLP frameworks and APIs,

e.g. IBM Watson Assistant (formerly Watson Conver-

sation). Industry-leading AI powers the underlying

natural language models of Watson Assistant

which

understand users and provides training recommenda-

tions for the chatbot on the ﬂy. There are also open

source options to add NLP capabilities when develop-

ing chatbots, e.g. NLP.js is a general natural language

library for Node.js. Visual authoring tools (VAT) can

be used to create a chatbot without the need of know-

ing programming languages and connect it to differ-

ent messaging platforms. Such tools offered as SaaS

may have AI and NLP capabilities and are able to host

the chatbots, so the user only needs to worry about the

design and speciﬁc tasks that the chatbot will perform.

Building a chatbot using a VAT, e.g. Chatfuel

, can be

as simple as assembling LEGO blocks.

Chatbots are different from one another in vari-

ous aspects such as the implementation, e.g. a chat-

bot for Facebook is not implemented in the same way

as one for Telegram. Also, a chatbot may have a

single-user focus to serve as a personal assistant or

to have a team-focus to be used in chat rooms for

a particular task, e.g. a poll bot. Another differ-

ence is the way a user may interact with them and

currently there are voice-based and text-based chat-

bots. Well-known voice-based chatbots that are cur-

rently present in smartphones are Apple’s Siri and

Google Assistant. Voice-based chatbots are activated

by the user employing a voice command, e.g. “hey

Siri”, or by pressing a button. Text-based chatbots are

usually present in messaging apps such as Facebook

Messenger (or Messenger for short), Slack, Telegram

and others, and users may start a conversation with

a chatbot by writing a text message. At the time of

this paper, Messenger is the leading consumer plat-

form for chatbots (Shevat, 2017) and it provides bot-

building APIs

. A chatbot implemented for Facebook

can interact with a user through Messenger (mobile

or browser-based), and this implementation connects

the chatbot to a page or app inside Facebook using a

webhook, which is the Messenger’s core bot experi-

ence. This webhook resides inside the chatbot’s code

and it is used to receive and send messages.

Figure 1: Architecture of a messaging platform connected

to a chatbot service.

Refer to https://www.ibm.com/cloud/watson-assistant

Refer to https://chatfuel.com

Refer to https://developers.facebook.com/docs/messen

ger-platform/introduction

CLOSER 2019 - 9th International Conference on Cloud Computing and Services Science

530

4 CHATBOT PROTOTYPES

Experimenting by means of a prototype is the

sine qua non of the scientiﬁc prototyping method

(Galindo Bello, 2018b). Nevertheless, building and

deploying a chatbot are not straightforward tasks as

it is required to manage the cognitive services of the

interface and bind it to an external API and services

of the target platform. Moreover, extensibility, scal-

ability and maintenance have to be taken into ac-

count while engineering a full-ﬂedged chatbot as it

is a comprehensive matter worthy of a book. In or-

der to achieve the objectives of this research, it was

chosen Messenger Platform Core set of bot-building

APIs. Node.js was chosen to the implement the chat-

bot’s back-end server for handling the allocation of

requests (REST API), MongoDB to store the users’

data and IBM Cloud (Platform as a Service, PaaS)

to deploy the chatbot as a cloud-native application

and considering its cognitive services for future work.

The chatbot was continuously enhanced until all re-

quirements were incorporated. Rich interactions were

added (i.e. buttons, image gallery) and a basic NLP

engine was integrated to derive the user intents. Fig-

ure 2 shows the architecture of the system where a

user sends messages using the Messenger platform

(mobile or browser-based). The Facebook’s page is

linked to the webhook which receives the user’s pub-

lic data and processes all intents. If the user does not

exist in the database, his/her information is stored.

Figure 2: High-level system architecture of the imple-

mented chatbot prototype.

The webhook receives two types of events: mes-

sages and postback messages. If the user intent con-

tains only text, the NLP engine derives it and de-

termines the action to be executed by the message

module. This module executes the action and pre-

pares the response payload. The response handler

takes the payload and sends the response message

to the user employing the Messenger’s Send API. If

the NLP engine cannot determine the action, the mes-

sage is stored as text in the database for later analy-

sis and the response handler sends a default message.

When the user taps a button inside the conversation,

the webhook receives a postback event and the post-

back module prepares the payload of the requested

option, passes it to the response handler and the re-

sponse message is sent to the user.

Once the prototype reached a mature grade and

experimentation was conducted, a second prototype

was required for the ﬁnal experiments but utilizing a

chatbot authoring tool (Software as a Service, SaaS).

Chatfuel was used to jump-start the implementation

of a second prototype keeping the same mature grade,

quickly integrate AI/NLP capabilities and ease the

data collection and analysis. Following the design

of the ﬁrst prototype, creating the second one was

straightforward and required zero lines of code. The

visual authoring tool was advantageous to quickly

create rich interactions and the main menu showing

the options in form of a gallery of images with but-

tons. An option to share the chatbot with other Mes-

senger users was also incorporated.

5 EXPERIMENT AND RESULTS

Attributable to the nature of the system (Messenger

chatbot) and its inputs (user intents), black box testing

was done following a logic-based technique (cause-

effect) where each user intent derives a chatbot action.

System and acceptance testing were conducted with

the end users and all the functional requirements were

tested in both cases.

Aiming to validate the foregoing research state-

ments, experimentation was conducted by testing the

chatbot prototypes (PaaS and SaaS implementations)

directly with the users in their natural environment,

which in this case were customers of the university’s

canteen (Mensa) chosen randomly. Direct observa-

tion plus user feedback helped to detect the ﬂaws of

the prototypes. In order to improve usability and user

experience, the PaaS prototype was enhanced after

each test (iteration) and new users were acquired on

each one. The SaaS prototype was utilized during the

last iteration.

Cloud-based Conversational Agents for User Acquisition and Engagement

531

The chatbot was able to send messages on arbi-

trary days asking to all the reachable users to kindly

respond some questions. Segmentation was done

based on the users’ data. Table 1 presents the ﬁnal

number of users and main segments. More segments

were identiﬁed by employing the chatbot to ask ques-

tions only to the users that were students and only 71

answered the questions. Only 4 students indicated to

have previous experience with chatbots, 59 would use

a chatbot to place an order, 3 would use a chatbot to

pay for an order and 65 would like to get a free beer.

Users were asked about their messaging app prefer-

ence and how often they are in Messenger ﬁnding

that WhatsApp is the most used app with a total of 69

users, followed by Telegram with 20 users. None of

the users marked Messenger as their most used mes-

saging app, nevertheless they use it sometimes (53

users) or seldom (36 users). Qualitative data did not

undergo into content analysis following a particular

method; it was primarily gathered by user feedback

and utilized to improve user experience and usability

of the chatbot prototype. Some data were collected

during the experiment at Mensa and other data were

provided by the users in the form of text messages us-

ing the feedback option of the Messenger chatbot. In

total, only 47 users wrote their feedback messages.

Table 1: Segmentation of Users.

Segment Number of Users

Acquired directly at Mensa 30

Acquired by words of mouth 59

Total reachable users 89

Female / Male 51 / 38

Vegan or vegetarian 14

Enrolled in the university 80

6 EVALUATION & DISCUSSION

This section presents the validation of the hypothe-

ses and evaluation of different parts of the research

work. A comparison of the PaaS and SaaS prototypes

is done to discuss which approach would be more

suitable for a gastronomic business. Facebook and

Chatfuel provide analytics tools, graphs and insights

activity. Descriptive statistics was used to summarize

all the collected data, construct the different graphs

and interpret the results.

6.1 Interpretation of the Findings

During the experimentation phase a total of 30 cus-

tomers were directly acquired and asked to talk with

their friends about the idea of using chatbots in restau-

rants. Advertising was done by creating a Face-

book page for the Mensa and following the tech-

nique business-to-consumer messages. Words-of-

mouth was used to acquire more customers as it has

proven to be one of the most effective acquisition

techniques (Wangenheim and Bayn, 2007). The ﬁ-

nal number of customers increased almost 200% in a

lapse of four weeks, which can be represented as an

average of 22 acquired customers per week. There-

fore, it can be stated that both techniques function

effectively for customer acquisition as previous stud-

ies (Vries et al., 2017; Wangenheim and Bayn, 2007)

demonstrated it. There were students without Face-

book proﬁles and others who refused to participate

in the experiment due to privacy concerns or just not

being interested in chatbots. As it is in any other sec-

tor, users are free to choose whether or not they try

a product and buy it afterwards. Messenger was not

the users’ ﬁrst option for text-based communication

notwithstanding that they have Facebook proﬁles. On

average only 65% of all the acquired customers were

re-engaged during the experimentation phase.

Gastronomic businesses have used incentives such

as giving discount codes or free food if the customers

provide their feedback with an online survey. This

technique has proven to be effective to retain and en-

gage customers, thus making them to come back to

their establishments and consume more. Although the

customers (students segment) of this experiment did

not receive a free beer as an incentive, 81.2% of them

indicated their desire to get it. From the same seg-

ment, only 5% indicated having used chatbots before

and 73% would be willing to use them as a means to

place an order. However, only 3.7% would pay an

order using a chatbot due to privacy concerns.

One user asked if the Mensa chatbot could pro-

vide also the price and calories of the meals – this

would be an added value for certain users. Providing

the information that users want and being available

at all times by means of a chatbot may help restau-

rants to improve their customer service and have a

more effective channel of communication. This is to-

tally congruent with previous studies (Brandtzaeg and

Flstad, 2017; Fadhil, 2018; Shawar and Atwell, 2007)

reporting that users employ chatbots when there is an

incentive or added value.

6.2 Objectives and Hypotheses

The objective of implementing a chatbot to determine

if it would be suitable for a gastronomic business was

reached. It was demonstrated that it is feasible for a

restaurant to offer services by means of a chatbot and

CLOSER 2019 - 9th International Conference on Cloud Computing and Services Science

532

thus acquire and keep customers engaged. In the last

decade restaurants have improved their service and

acquired more customers by technological updates

such as the adoption of mobile apps and online sys-

tems. In a similar way, it is possible to adopt chatbots

as another channel of communication and customer

service. Furthermore, chatbots can be used for adver-

tising in order to acquire new customers and to cre-

ate marketing campaigns in social networks and thus

keeping engaged the different segments of customers.

These facts from the foregoing literature were proved

with the implemented chatbot, hence hypothesis 1 and

3 are valid. Previous studies state that technology may

improve customer service, quality customer service

attracts more customers and happy customers recom-

mend a product or service. Therefore, by implement-

ing a chatbot to offer 24x7 service and providing to

the customers the desired information, they may con-

tinue visiting a restaurant and recommend it to others;

hence hypothesis 2 is valid. The PaaS implementation

of the chatbot was designed to be a platform-agnostic

application and not depend on a proprietary cloud ser-

vice; its modules are stateless and run inside contain-

ers which can be scaled automatically on demand; a

load balancer improves the distribution of workloads

and can also be scaled on demand. These capabil-

ities for a better performance and scalability can not

be conﬁgured in Chatfuel, hence hypothesis 4 is valid.

6.3 Chatbot Implementations

The PaaS prototype evolved until all functional re-

quirements were incorporated and tested with the

users. The NLP engine was limited as it does not fol-

low powerful AI algorithms nor has ML capabilities,

and in distinct occasions it did not perform properly.

On repeated occasions the users wrote messages in

German language and the NLP engine was not able to

understand the user intents as it was only set for En-

glish. For the chatbot menu (options), it was observed

that a simpler menu with options as buttons had a bet-

ter usability than a menu as an image gallery. The

quality of the implemented PaaS prototype was eval-

uated according to the standard ISO/IEC 9126-1, so

the user was able to utilize the chatbot (functional-

ity) and received correct information when using the

chatbot options (reliability), the chatbot responds im-

mediately to user inputs (efﬁciency), it is compliant

with the system architecture (maintainability) and it

can be deployed on any PaaS that supports Node.js

applications and NoSQL databases (portability).

The SaaS prototype was implemented in a very

short period of time thanks to the drag-and-drop fea-

tures and all functional requirements were covered.

The non-functional requirements were partially ad-

dressed as some parts are managed by the chatbot-

building tool, i.e. hosting the chatbot, creation of a

webhook, HTTPS ports and database. NLP capabili-

ties were integrated using the tool’s AI engine, how-

ever it was not able to process some user inputs and it

also failed to process messages in German as it only

understands English. A bug was observed on Mes-

senger running under iOS, Android, some browsers

(Safari, Chrome) and it happened randomly; some

users were able to write text and others were not. The

tool’s documentation suggests purchasing a PRO ver-

sion to solve the bug. Notwithstanding that the au-

thoring tool truly eases the creation of chatbots, its

documentation is not extensive and it is not a ﬂexible

tool. For instance, it is not possible to broadcast mes-

sages only to users celebrating their birthdays because

this would require the execution of custom code, but

the tool does not have such feature. Furthermore, it is

not possible to export the gathered user data neither

import data from other chatbot.

Which approach would be more suitable for a gas-

tronomic business? It depends. Several aspects need

to be considered such as the approximate number of

initial users, who will be in charge of the implementa-

tion and maintenance and how fast it is required to go-

live. An important aspect to take into account when

choosing an approach is the main goal of the chat-

bot and the type of information that it will provide to

the customers. For instance, the Mensa is the type of

restaurant that changes its menu every week; this im-

plies changing the pictures of the meals, description,

ingredients, prices, calorie information and other de-

tails. For this type of restaurants, a PaaS implemen-

tation would be more suitable. Also, if there will be

active advertising and engagement through notiﬁca-

tions and target of segments, the PaaS implementation

would be ideal. In the case of a small restaurant or

cafeteria with menus maintaining the same products

and prices over the year, then the information is static

and a chatbot authoring tool would be more suitable,

even at no cost when starting with a free version and

considering an upgrade depending on the number of

users and beneﬁts to receive within the upgrade.

6.4 Business Perspective

Online systems have reduced the stress of making

table reservations, waiting in queues at restaurants,

placing orders and also making payments. It is then

crucial for gastronomic businesses to keep up with

technology and continue offering quality service em-

ploying different channels – chatbots adoption is one

of them. Social networks have tremendously in-

Cloud-based Conversational Agents for User Acquisition and Engagement

533

creased their popularity and number of monthly ac-

tive users over the past years, and it should be used

by gastronomic businesses as a strategy to acquire

and engage customers. Social networks and chatbots

could signiﬁcantly reduce the costs of traditional ad-

vertising, customer service, acquisition and engage-

ment compared to online systems and mobile apps.

For instance, a restaurant could employ a chatbot to

send messages to all female customers on the Interna-

tional Women’s Day inviting them to come and get a

free beverage or special discount. This may be a win-

win situation because the customers would consume

more due to the special celebration. Besides saving

implementation costs, gastronomic businesses should

not overlook all the beneﬁts of adopting chatbots.

7 CONCLUSION AND FUTURE

WORK

In the same way that hardware has evolved from

computers to mobile and wearable devices, also soft-

ware has evolved and continues to do so. Compa-

nies continue to migrate their products and services

to the cloud; mobile interfaces have become plain,

simpler and more usable; and chatbots are becoming

a common user interface as well a trend. For better

or worse, AI and ML capabilities continue to grow

and chatbots will become more intelligent in the next

years. Companies have already started to incorporate

chatbots within their Facebook business pages, web-

sites and other social networks such as Twitter. Soon

chatbots will be an essential part of customer service.

It was demonstrated in this research that users

would be willing to try a new trend and use the ser-

vices of a chatbot when there is added value. In the

same way that restaurants use web and mobile apps

to offer services, chatbots will also become another

channel for communication with the potential beneﬁt

of having lower costs of implementation and mainte-

nance compared to web and mobile applications.

Running different modules of the system in con-

tainers makes it easy to manage and secure applica-

tions independently of the infrastructure that supports

them. With this approach, if one module is down

(fails), others can continue working while the failed

module gets restarted. Cloud-native capabilities such

as Serverless allow running dynamic workloads and

pay-per-use compute time in milliseconds. As part of

the future work, the implemented PaaS chatbot is be-

ing redesigned following cloud software engineering

principles to migrate it fully to a microservice archi-

tecture and thus target more instant messaging apps.

REFERENCES

Brandtzaeg, P. and Flstad, A. (2017). Why people use chat-

bots. In Lecture Notes in Computer Science.

Braun, A. (2003). Chatbots in the Customer Communica-

tion (in German: Chatbots in der Kundenkommunika-

tion). Springer.

Fadhil, A. (2018). Can a Chatbot Determine My Diet?: Ad-

dressing Challenges of Chatbot Application for Meal

Recommendation. University of Trento.

Galindo Bello, M. J. (2018a). Discretionary Adoption of

Chatbots for Customer Acquisition and Engagement

in the Gastronomy Industry. Department of Applied

Computer Science, HS-Fulda.

Galindo Bello, M. J. (2018b). Scientiﬁc prototyping:

a novel approach to conduct research and engineer

products. In 2018 IEEE 9th International Confer-

ence on Software Engineering and Service Science

(ICSESS), pages 32–35. IEEE PRESS.

Maslowska, E., Malthouse, E. C., and Collinger, T. (2016).

The customer engagement ecosystem. In Journal of

Marketing Management, volume 32, pages 469–501.

Routledge.

Oracle (2018). Cloud predictions. https://www.oracle.com/

assets/2018-cloud-predictions-4242085.pdf, [Ac-

cessed: 26-10-2018].

Parasuraman, A. P., Berry, L., and Zeithaml, V. (1991).

Understanding customer expectations of service. In

Sloan Management Review, volume 32, pages 39–48.

R. Rising, G., Weizenbaum, J., and H. Freeman, W. (1978).

Computer power and human reason: From judg-

ment to calculation. In The American Mathematical

Monthly, volume 85.

Shawar, B. and Atwell, E. (2007). Chatbots: Are they really

useful? In Journal LDV Forum, volume 22, pages 29–

49.

Shevat, A. (2017). Designing Bots: Creating Conversa-

tional Experiences. O’Reilly UK Ltd.

Turing, A. (1950). Computing machinery and intelligence.

In Mind, volume 59, pages 433–460.

Vries, L., Gensler, S., and Leeﬂang, P. (2017). Effects of

traditional advertising and social messages on brand-

building metrics and customer acquisition. In Journal

of Marketing, volume 81.

Wangenheim, F. and Bayn, T. (2007). The chain from cus-

tomer satisfaction via word-of-mouth referrals to new

customer acquisition. In Journal of the Academy of

Marketing Science, volume 35, pages 233–249.

CLOSER 2019 - 9th International Conference on Cloud Computing and Services Science

534