Conversational AI integration with Contact Center platform

ABSTRACT

Conversational AI is called Virtual Assistance, which allows free flow communication between user and BOT with help of NLP technology. It evolved over time from simple FAQ to intelligent Virtual Assistance (BOT) with leverage of Gen AI. It focuses on bringing automation towards simple and medium business use cases in Finance Domain and it helps to enhance contact center operation. It uses easy deployment tools to improve overall lead time. Today’s Virtual assistance uses LLM features to bring more human like conversation which helps better customer experience. integration between Contact Center & Conversational AI platform more seamless, which offers customers with channels of choice and at the right time. Most BOT platforms offer native ASR/TTS (for Voice BOT) or integration with best-in-class 3^rd party ASR/TTS, which enable better understanding of customer ascent, dialect. With Enterprise data integration, Analytic & build in intelligence help Business to predict customer needs and seek appropriate measures. As per the industry wide research, it is learnt that the conversational AI market is estimated to be valued at USD 10.65 Bn in 2024 and is expected to reach USD 44.38 Bn by 2031, growing at a compound annual growth rate (CAGR) of 22.6% from 2024 to 2031. Financial Organizations across the globe are keen and in the process of adopting BOT technology to improve automation, enhance customer experience & reduce overall operation costs.

What is conversational AI in contact center?

Conversational artificial intelligence (AI) is a technology that makes software capable of understanding and responding to voice-based or text-based human conversations. It uses machine learning; natural language processing understands user intent and form responses. Bots can continuously learn from conversations with customers, so they’re able to deliver more helpful responses as time goes on. 

Interface provide user to interact with back-end system using natural language say English, and this interface called as conversational interface. Conversation is the exchange of messages between end user and BOT to complete specific tasks.

Customer/User will say in different ways, words, dialect & accent, are the major Challenge to design conversational BOT. This can be addressed with the help of Natural Language processing Engine (NLP Engine), its NLP engine of Virtual assistance make sense of human language that is meaning full. NLP is used to determine a user’s intention (Intent Recognition) and to extract information (Entity Extraction) from the user’s utterance and to carry on a conversation with the user to execute and complete the task. The responsibility of NLP engine is to understand or qualify user intent and extract Entity/variable and store in data table to use for processing. NLP Engine supports multiple languages. Currently Contact Center platforms are offers with native BOT capabilities or easy integration with known players in the market. The focus is on improving self-service, allowing omnichannel customer interaction, real time guidance to customer service agents.

Evolution of Bots

Basic FAQ BOT (1st Gen)- Action BOT (2nd Gen)- Intelligent BOT (3rd Gen)- Generative AI BOT(4th Gen)

Components of Conversational AI:

1. Machine Learning

Machine learning consists of algorithms, features, and data sets that systematically improve over time. The AI recognizes patterns as the input increases and can respond to queries with greater accuracy. Success of conversational AI depends upon % accuracy in understanding customer intents & can be able to predict accurately

2. Natural Language Processing

Conversational AI uses NLP to analyze language with the aid of machine learning. Language processing methodologies have evolved from linguistics to computational linguistics to statistical natural language processing. Combining this with machine learning is meant to significantly improve the NLP capabilities of conversational AI in the future. Natural Language Processing (NLP) Engine and its key tasks to understand user utterance or intent & attributes/variables.

3. Data

The success of conversational AI depends on training data from similar conversations and contextual information about each user. Using demographics, user preferences, or transaction history, the AI can decipher when and how to communicate

4. Conversation Design

Conversational Design is a crucial step toward deployment. The first step is to understand all possible questions customers may ask & define appropriate responses from the BOT, to build a conversation prototype. This prototype should then be reviewed and agreed upon with multiple stake holders before moving on to the building phase.

 How Does Conversational AI Work? 

Conversational AI allows humans to interact with machines through four steps:

Step 1: It starts with receiving information from humans, which may be written text or spoken words. As the input is spoken, voice recognition converts it into text by Speech to Text (ASR) that is in a machine-readable format.

Step 2: The second step is for the application to grasp what the text means. To understand the intent behind the text, conversational AI uses natural language understanding (NLU) – a part of NLP algorithms.

Step 3: Utilizing dialog management, the application determines the response based on its understanding of the text’s intent. By orchestrating responses and converting them into a human-readable format through natural language generation (NLG), dialog management forms the other aspect of NLP. For Voice, it converts text back to speech by Text to Speech (TTS) that is understandable to the user with proper accent & dialect. Depending upon user or customer conversation it kicks start the automation to bring relevant dynamic data to present back to user and store for future use. With advancement of technology and shift in customer adoption, it is feasible to integrate with LLM to bring more human like conversation.

Step 4: Depending on the platform, either the application delivers the response in text or uses speech synthesis (artificial generation of human speech) so that the user receives it over a voice channel or irrespective of channel of customer choice.

Performance of Conversational AI

It depends on multiple factors stated below:

1. Conversational Design

Considering all possible conversations, the user may need to satisfy specific requirements. Conversational designers will play a key role in designing conversation for specific Business use cases.

2. Right ASR & TTS engine for voice communication

Customer preferences to have voice conversation and demand to select right ASR, TTS engine are challenges. % Fail to understand customer intents and proper response where user can understand it better.

3. NLP configuration & Training

Understanding the intent behind the text & generating the right response is very crucial for NLP engine. Intent Identification is the key parameter for success of any conversational design. Regular and the right amount of training is very crucial. NLP engineers must monitor the failure intent and ensure regular training to improve accuracy in intent detection and better response.

4. Platform offering

What are features supported, how flexible is the integration to look for system of records and other data to drive the conversation, flexibility of integration with best of class ASR/TTS, integration with LLM, quick to deploy, Integration with contact center, omnichannel integration, support of multiple languages etc.

5. Choice of right Business use cases

Business teams decide how the right business case can qualify automation which brings immense business benefits and does not comprise compliance, privacy.

6. Availability of API’s

What is the maturity of APIs available and what Business like to achieve is the key parameters for the success

7. Deployment & availability

Technology & Enterprise architect to decide on optimum architecture design and ensure low latency between NLP, ASR/TTS & Business Applications. Also, the decision to leverage native contact center capabilities or 3^rd party players is crucial Business decision.  

 Benefits of Conversational AI

Using conversational AI, organizations can:

• Automate Customer Interactions – Conversational AI shares answers to simple, transactional queries. It also provides personalized advice – with CRM integration
• Replace the IVR – The use of voicebots enables organizations to follow this trend, personify a brand, and meet the needs of each customer using their preferred channel.  Which helps to avoid navigation through complicated IVR menus.
• Increase Sales – Conversational AI can facilitate a consistent and convincing selling strategy. For example, a chatbot that tracks how a customer uses the website can offer support when they take a long time to check out. Also, it can proactively reach out to a customer with a discount on a product that they revisit but never purchase to drive sales. Subscription offers may also work well.
• Conduct Sentiment Analysis – With advanced conversational AI, businesses can analyze customer sentiment and fine-tune processes. For example, many conversational AI systems categorize interactions as positive, negative, or neutral based on the customer’s use of language. Through this process, a chatbot can respond accordingly and provide more personal experience.
• Reduce Costs – Conversational AI lowers staffing requirements, handling tasks such as answering customer queries at no extra charge. It also requires no time off and is not prone to human error & also 24x7 availability.

Industry wide Comparison of Leading BOT Frameworks (Indicative)

Industry Trend with BOT Implementation across globe (Indicative)

 1) Erica, by Bank of America

2) Amex bot, by American Express

3) EVA, by HDFC Bank

4) Amy, by HSBC Bank (Hong Kong)

Impact of conversational AI & Gen AI on Contact Center Operation 

Omnichannel Customer Engagement

• Offers Intelligent self-service and assisted service options on any channel of customer choice

Personalized CX automation & Intelligent routing

• Detects intent and automate customer requests via conversation.
• Enable cross-channel customer experience by allowing context switching.
• Ensure CSAT by monitoring live customer sentiment and transferring requests to the right agent at the right time and for the right reason

Empower/Engage Agents

• Empower agents by providing coaching and automating routine agent activities.
• Improves CX by improving EX and making agents more productive

Knowledge and insight

• Automate FAQ’s.
• Provides agents and customers with the next best actions based on the customer’s requests.
• Reports and dashboards to monitor and improve automation/NLU & take business decisions.

Call Summarization

• Consistent format & accurate summary information
• Reduction of Average Call Handling Time by 30-40 seconds
• The agent can edit summary to verify the discussion with the client
• Eliminate manual noting during client call and information lost

IVR Handoff

• Reduction of Average Handling time of agent
• Minimize client frustration by providing relevant information to the Agent
• Improve Agent productivity in handling client queries

Knowledge Management

Extract unstructured PDF Documents, and customize prompt in line with user intent and data available part of repository Information (prompt) then send to LLM to get requested information. Response is presented to the user.

 ASR/TTS Integration with NLP Engine

NLP Platform supports native ASR/TTS for Voice BOT, and it is flexible to integrate with 3rd party ASR/TTS tools such as Deep gram, Eleven Lab, AWS Transcribe, Microsoft, Google etc.

Automatic Speech Recognition (ASR) performance measured with the parameters below:

WER (Word Error rate) - is a common metric of the performance of a speech recognition or machine translation system. The WER metric ranges from 0 to 1, where 0 indicates that the compared pieces of text are identical, and 1 indicates that they are completely different with no similarity. WER is the ratio of errors in a transcript to the total words spoken. For example, a 20% WER means the transcript is 80% accurate. ASR systems developed by Google, Microsoft and Amazon have a WER of 15.82%, 16.51% and 18.42% respectively.

WIL (Word Information Loss). This value indicates the % of characters that were incorrectly predicted. The lower the value, the better the performance. Word information loss rate of 0 being a perfect score.

TTS: - the task of generating natural sounding speech for the given text input.

Consideration of Conversational AI integration with contact center platform

Contact Centers providers now offer native BOT capabilities; however, it also allows to integrate external BOT platforms for the specific Business use cases & better performance. Below are the possible integration approaches with external BOT platforms.

Approach 1: - Contact Center IVR+ASR/TTS integration with conversation AI.

Approach 2: - Contact Center IVR integration with conversation AI ASR/TTS.

How to boost conversational experience with Gen ai?

Generative AI (GenAI) significantly enhances conversational AI by enabling more natural, creative, and contextually relevant responses. GenAI can generate text, images, and other media, allowing conversational AI systems to go beyond simply answering questions to actively contribute to a conversation in a more dynamic and engaging way. 

Here's how GenAI contributes to Conversational AI:

• Enhanced Creativity and Originality:

GenAI can generate diverse and original content, allowing conversational AI systems to offer dynamic responses rather than pre-defined responses to user queries. This makes interactions more engaging and less robotic. 

• Natural Language Generation:

GenAI models, like Large Language Models (LLMs), are trained on massive amounts of text data, enabling them to produce human-like text that flows naturally and contextually. 

• Contextual Understanding and Response:

GenAI can analyze the context of a conversation to generate responses that are relevant and appropriate, rather than just focusing on keywords. 

• Content Generation for Knowledge Bases:

GenAI can be used to create and maintain knowledge bases, providing chatbots with a wealth of information to draw from when answering user queries. 

• Personalized Experiences:

GenAI can be used to personalize interactions with empathy, tailoring responses to individual user preferences, understanding emotions and past interactions. 

• Improved User Experience:

By generating more engaging, natural, and relevant content, GenAI can improve the overall user experience of conversational AI systems. 

• Conversational AI Engineering:

Co-pilot features to build bots based on textual description, develop text cases, training and testing data.

Examples of GenAI in Conversational AI:

• ChatGPT: A prime example of a chatbot that leverages both conversational and generative AI. 
• Google Dialogflow with GenAI features: Utilizes LLMs and Google Enterprise Search to provide natural language interactions and context-aware responses. 

Next Step the way forward

Most Financial institutes currently use traditional methods to reach customers and, customer do not have flexibility to reach for Business need. Organizations are evaluating conversational AI technology to bring more synergy with the Business priorities, strategies and competition. Hence, adoption is increasing due to maturity in NLP technologies, machine learning, data availability & LLM. Also, it brings a lot of flexibility and openness to integration which improves overall demand in markets for adoption.

To protect the investments already made in contact center technologies and to meet growing business demands while staying competitive, here is a phase-wise recommendation roadmap to add AI based advance features like conversational AI, Gen AI, predictive engagement etc

1. Evaluate current technology stack, conduct a gap analysis to understand performance, ROI & user adoption. Optimize workflows & IVR. Simplify call flows, reduce transfer rates, and use data to improve self-service options.
2. Decide the right AI/conversational platforms, as per Business use cases, expected TCO & organization priorities.
3. Create enterprise level technical & functional architecture with integration touch points. Develop detailed implementation roadmap & risk assessments. Develop conversational design prototypes.
4. Implement low priority Business use cases & monitor the realization and benefits
5. On successful transition of low priorities (existing & new) business call types. Deploy next level business use cases & other AI features into existing contact center platforms
6. Leverage LLM capabilities with necessary Guardrails & securities in place
7. Monitor the realization, benefits & operational KPIs for contact center along with the expected Business outcome, ROI.
8. Monitor and continue the enhancement

References:

https://www.cxtoday.com/contact-centre/what-is-conversational-ai/