DISCUSSION ASSIGNMENT INSTRUCTIONS

AI as Complement to Cloud CRM 1

AI as Complement to Cloud CRM:

Intelligent Virtual Assistant, Chatbot, Speech Analytics

Phase 2

Sam Towne

AI as Complement to Cloud CRM 2

Contents

Introduction......................................................................................................................................3

Problem Statement...........................................................................................................................4

Feasibility Study..............................................................................................................................4

Project Plan....................................................................................................................................12

Conclusion.....................................................................................................................................14

AI as Complement to Cloud CRM 3

Introduction

Phase two of the Synectic IVA project will provide a literature review of current artificial

intelligence design practices and a framework for comparing the old and new system. The

development of Artificial Intelligence (AI) is complex. The challenge is that some of the larger

and best-known AI projects, like Siri, Amazon Alexa, and Google are proprietary and are not

publishing their design strategies. This is for good reason, because there is no true AI leader and

there is much competition. Many AI design strategies lean on existing cloud computing and

computer system frameworks but also add their own twist to the that is specific to artificial

intelligence. For instance, Soumya’s 2018 article outlines an architecture that is designed to build

empathy and consciousness into the intelligent machine but also relies heavily on traditional

frameworks (Banerjee 208). Balke De Vos and Padget use traditional frameworks for the

development of the core AI design, but then recommend the use of intelligent agents for the

testing portion of the AI to speed up the product development (Balke, De Vos, Padget 2013). It

seems that many designs involve both institutional frameworks such as ETSI, ISO, and AI

focused models for system comparison. Phase 2 of the IVA project will compare the new and old

systems using the following ETSI standards from Cloud Standards Coordination Phase 2 (SR

003) and Cybersecurity (TR 103):

1. Cloud Computing Standards and Open Source; Optimizing the Relationship between

standards and Open Source in Cloud Computing (SR 003 382) 2. Interoperability and Security in Cloud Computing (SR 003 391) 3. Cloud Computing Standards Maturity Assessment; A New Snapshot of Cloud

Computing Standards (SR 003 392) 4. Identification of User Needs (SR 003 381)

AI as Complement to Cloud CRM 4

5. Cyber Security; Design Requirements Ecosystem (TR 103 369)

The rest of this paper is ordered as follows. A literature review relating to artificial intelligence

and IVA design strategy will be performed. Then a system analysis and design of the old and new

systems will be performed.

Literature Review

One of the challenges of designing an artificial intelligence system is that “human

behavior is complex and difficult to formalize mathematically” (Helbing and Balietti 2011). This

complexity in human behavior makes it hard to develop an intelligent virtual assistant that is

appropriately accommodating for the large varieties of requests that can be presented. “Human

behavior is characterized by stochasticity. Logic (programming) approaches tend to not account

for this, but rather focus on proving “ideal” conditions and agent behavior” (Balke, De Vos,

Padget, p. 381, 2013). If ideal conditions are not met they recommend having a recovery method

that the AI can use to get back on track. There must be a system for planning for random events

to gracefully recover from unexpected user responses.

The amount of data available is massive, and because it is growing rapidly, there are also

multiple approaches for providing the virtual human (VH) with the searching tools it needs to

provide relevant data to the customer. The problem is not getting the data. The problem is getting

the right data without requiring the human to dig through pages and pages of results. It must

seem like an easy process to the user. Three of these approaches are point-wise, pair-wise, and

list-wise. Their strengths are combined and termed “boosting”. “Boosting successfully combines

the idea of learning sets with machine learning. Known and resolved questions serve as inputs in

the computer. After learning, the computer formulates rules based on the inputs” (Kao, Fahn,

AI as Complement to Cloud CRM 5

2013). Boosting provides a method that allows large and different sets of data to be used for

building and testing the IVA’s baseline knowledge against the required function it must perform.

Values, culture, and emotional intelligence are important considerations for the intelligent virtual

assistant. Another framework consideration is value sensitive design (VSD). This framework

incorporates human values into the structure of the system. VSD involves values, stakeholders,

and value tensions. Stakeholders are the people who interact with the virtual assistant. Values are

what the stakeholders care about. This can be time, money, service or other things. Value

tensions occur when one value conflicts with another or is increased at the expense of another

value (Harbers, Neerincx, 2017). This method recommends situated cognitive engineering (SCE)

for the design of the virtual assistant because “it is specifically geared toward the development of

human-machine interaction.” Harbers and Neerincx recommend adding VSD into the SCE

design process to make the human machine interactions more natural. Having a good

understanding of the user’s values makes it so that the virtual assistant can respect those values.

This makes for a better user experience. Although less research has been formed on emotional

intelligence within software agents, Kazemifar, Aghaee, Koenig, and Oren are hoping to provide

a framework that takes the user experience to the next level are. They are incorporating

memories and emotional experiences into the AI memory. Their tests involved creating a

network of intelligent agents and having them interact with each other which was based upon

Soar. Soar is “a general cognitive architecture for both modeling cognitive systems and

implementing systems that exhibit intelligent behavior. Episodic memory in Soar encodes and

stores the entire contents of the top state of Working Memory with a time stamp, which can be

used in retrieval” (Kazemifar, Aghaee, Koenig, Oren, 2013). Basically, this is creating intelligent

AI as Complement to Cloud CRM 6

virtual assistants that have a memory and can attach positive and negative memories to certain

interactions. The goal is not to have intelligent robots that portray emotions, but rather to have

them be able to understand emotion. This way they can respond in appropriate ways.

Culture can also be modeled in software. “As the development of virtual agents focuses

increasingly more on the social aspects of human interaction, it becomes crucial to address the

notion of culture and how it affects human behavior” (Degens, Paiva, Prada, Hofstede, Beulens,

Aylett, 2016). Degens, Paiva, Prada, Hofstede, Beulens, and Aylett propose the social importance

dynamics framework. This type of understanding of culture is important to design a system with

agents that accommodate to the local culture in a natural way.

The Synectic IVA project has an aggressive development and testing timeline.

Crowdsourcing bootstrapping (CB) is a way to rapidly build AI. The Crowdsourcing bootstrap

framework is a good method for this type of project. It enables the IVA to accumulate large

amounts of foundational knowledge quickly. “The CB framework uses crowdsourcing for large

scale data collection” (Borish, Lok, 2016). Crowdsourcing usually involves providing a web

interface that allows users to provide direct input towards a problem, or to provide sample data

through the participation in surveys or other data collection tools. A challenge with

crowdsourcing is getting enough people motivated to spend the time to record the data. Crowd

design frameworks tend involve either linear competition or iterative improvements as reward

systems for the “crowd”. “Linear competitions might be single or multistage and reward workers

with staged payments or a winner-take-all prize” (Wu, Corney, Grant, 2015). Iterative

improvements are often the situation where workers are the “crowd” and receive monetary

AI as Complement to Cloud CRM 7

reward for improving upon the existing system. For the Synectic IVA project, iterative

improvements could be a valuable testing method during the crowdsourcing phase.

Video game developers also focus on tools that allow for rapid AI development. This is

because “video game AI is often specifically designed for each game” (Safadi, Fonteneau, Ernst,

2015). This challenge caused the developers to make a system that enabled them to use the same

AI across multiple game types by what they term “detachment”. The focus is “creating AI that

operates solely on concepts” (Safadi, Fonteneau, Ernst, 2015). This conceptual design of the

framework allows an AI to be broken into small chunks that can be implemented quickly to meet

changing requirements. “Agile or adaptive enterprise architecture (EA) capability is a key

strategic capability that plays an important role in describing the structure, behavior, social,

technology, and facility elements of an adaptive enterprise” (Gill 2013; ISO/IEC 42010 2007). If

new business arrive that require AI, the conceptual AI foundation can be reused in altered

environments.

Distributed artificial intelligence is the concept of having agents with different purposes

and skills, that work together to solve problems. The collaboration between AI agents is called

collective intelligence (CI). “The objective of this discipline is to build a theoretical framework

and tools for modelling agents with planning and communication skills” (Mazilescu, 2017). This

enables the virtual agents to have specialties, just like humans specialize in certain areas. It also

enables large problems to be worked on collectively from different angles by multiple agents in

collaboration.

System Design Methods

AI as Complement to Cloud CRM 8

The below diagram is the support request use case of the existing system. The trigger is

when a customer calls the support group. A support agent receives the phone call and creates a

support ticket. The support agent provides the customer with the ticket number and sends an

email at the end of the conversation. Then the support ticket is made available for the next

available technician to work. Support agents work the service requests and provide status updates

to account management, customer, and management. This requires the support agent actor to

receive incoming calls, create the new service request, work the service request, and update

customer, sales and management throughout the process.

AI as Complement to Cloud CRM 9

Use Case Title Support Request Existing System Primary Actor Customer Level Kite Stakeholders Customers, management, support agents, account management and

sales Precondition The customer is already in the cloud CRM system. The calling party is

an authorized contact. Minimal Guarantee No ticket is made. The scope of work is miscommunicated. The

customer is unable to reach an available support agent and hangs up. A ticket is made, but no follow up occurs.

Success Guarantee A support ticket made. The customer receives an email with their ticket number and expected resolution timeline. The problem is resolved over the phone.

Trigger The customer calls with a support request. Main Success Scenario

Support calls are received without long wait times. Customers are satisfied with the level of service.

Extensions A customer request is sent to an account manager if it is sales related or to management as needed.

AI as Complement to Cloud CRM 10

Use Case Title Support Request New System Primary Actor Customer Level Kite Stakeholders Customers, management, support agents, account management and

sales Precondition The customer is in the CRM. Minimal Guarantee No ticket is made. The customer is frustrated with the IVA and hangs

up before the process is complete. Success Guarantee A support ticket made. The customer receives an email with their

ticket number and expected resolution timeline. Trigger A customer call is transferred to the IVA. Main Success Scenario

Support calls are received without long wait times. Customers are satisfied with the level of service.

Extensions A custo mer request is sent to an account manager for sales related inquires or to management as needed.

AI as Complement to Cloud CRM 11