Making Sense of the AI & Automation Landscape

This article is meant to provide an overview to help you begin to navigate the AI and automation landscape, especially with a focus on Generative AI.

What's covered:

• Perspective: The 5th industrial revolution; this new chapter in history where the pace of exponential change in technological capabilities is fueling the AI hype cycle
• What we have been doing: The role of RPA, Predictive AI, and Conversational AI we currently use to improve the customer/employee experience
• How we can use RPA, Predictive AI, Conversational AI, and Generational AI: Practical use cases
• Underlying similarities and differences: A look at RPA, Predictive AI, Conversational AI, and Generational AI technologies and the resulting in differences in business outcomes
• Steps to consider: What to consider when you are thinking about implementing Generational AI in your organization.

Making Sense of the AI and Automation Landscape

The article is by no means a comprehensive overview and everything you need to know about AI and automation. The intention is to provide some basic building blocks so you can sort through the marketing hype vs. true technological differences and understand the connection between the technology and the business outcomes. This way, when you and your team are evaluating what you have, you have a basic understanding to ask the right questions in your organization and of vendors to move forward as you consider how to implement Generative AI.

With all the news about GenAI [Generative Artificial Intelligence], I wanted to address a couple of things I see in the marketplace:

1. First, there is a lot of “GenAI white-washing.” Nearly every tech company is saying they offer GenAI which may or may not be true.
2. Second, there is a great deal of buyer hesitancy to move forward with AI projects because they are overwhelmed and confused.
3. Third, GenAI marketing needs to transform into education to help potential customers see what they can’t see. ?

AI Overview: Let’s take a step back to put the field of automation and AI into perspective through the lens of customer and employee experience technology. With perspective, leaders can see a path forward. And, taking action is key because the field of GenAI is moving quickly, quicker than any tech revolution to date – see the exponential effect technology evolution has had on our lives below.

We are now in the 5th industrial revolution, as described in the book Empathy to Action , where the realization that “for profit” and “for benefit” don’t need to be conflicting business models, permitting us to put our focus on humanity and personalization through deep, multi-level cooperation among people, technology, and businesses. The question remains, what will we do with this technology? Some speculate it is the end of humanity and others see a bright future for all.

Hype Cycles: Part of the confusion around Generative AI is because we are in the hype cycle. Hype cycles, first introduced by Gartner analysts back in 1995, happen when any new tech emerges. They go all the way back to the beginnings of each tech revolution…. From CRM/ERP, Cloud, Social /Digital Media, Big Data… Now people are getting caught up in the magical mysteries of generative models, large language models (LLMs), and transformers.

Part of the hype is due to how quickly the underlying Generative AI technology is advancing. And part of it is there is more focus on the hype than there is on the practical application and use cases. So, let’s dig in and get a little more practical.

The Role of Automation Technology in Customer / Employee Experience

I’ve chosen to define four areas of automation innovation to show the evolution of capabilities and use. The areas are RPA, Predictive AI, Conversational AI, and Generative AI. If you have been working in customer experience, customer service, contact centers, or employee experience/future of work, you have likely been using RPA (Robotic Process Automation) or some form of Predictive AI and/or Conversational AI for the last 10+ years. I point this out because we can get swept away with the hype and not realize we already using many of these technologies.

Getting Real: Starting with RPA and the evolution to Predictive and Conversational AI, these technologies are why companies were able to improve the customer/employee experience. From self-service bots and virtual assistants to predictive routing to agent/employee-assist applications, they have all gradually made experiences better. (If you want to read more, Chapter 9, in Empathy in Action , is all about the application of AI and automation technology to the customer and employee experience.)

Getting Personal: We’ve all used AI-based technologies in our personal lives. Have you been watching a movie or shopping on a website, and it shows you things you might like based on previous purchases or browsing history? That experience is based on using recommendation engines and those are typically powered by a type of AI technology known as machine learning.

In addition, if you use a smart personal assistant, like as Apple's Siri, Amazon's Alexa, or Google Assistant, you are using some form of AI. When integrated with smart home devices, AI assistants can manage your home thermostats, control lights, or manage security systems based on your habits and preferences. For example, your assistant might learn to dim the lights as it gets closer to your usual bedtime.

Wearable devices, integrated with AI, can monitor vital signs like heart rate, sleep quality, and activity levels. They can predict potential health issues based on trends and anomalies in the data. For instance, an unusual heart rate pattern might prompt the device to advise the user to take it easy or even consult a doctor.

Getting Practical: While the capabilities and applications of Generative AI are emerging and are quite significant, many companies haven’t even taken full advantage of what RPA or Predictive and Conversational AI can do. And, the combination of PRA, Predictive, Conversational, and Generative AI will provide a competitive advantage to first movers.

Basic Definitions

RPA: Robotic process automation is a form of business process automation technology that is based on software robots (bots) or artificial intelligence (AI) agents. It’s software that uses computer programs to?automate repetitive, rule-based tasks. Example: Automating filling in forms, reentering data, copying, and pasting, data entry, and data analysis.

Predictive AI: Predictive AI is about understanding and predicting future events based on historical data to provide insights, make informed decisions, and personalize interactions based on predicted future behavior. Example: If you like that, you’ll like this type of scenario.

Conversational AI: Conversational AI is about facilitating natural and effective communication between humans and machines by understanding, processing, and responding to human language in a way that is natural and intuitive. Example: Chatting with a chatbot or responding to an SMS.

Generative AI: Generative AI?is about taking input and creating and generating new content or automating responses, generating creative solutions, and handling a wide range of queries without the need for predefined scripts. Example: Inputting information and receiving content or images based on the input.

The AI-Automation Combo Deal: The improvements in customer and employee experience accelerated when we went from RPA to Predictive / Conversational AI. The next wave of exponential change will be from the application of Generative AI, along with RPA, Predictive AI, and Conversational AI.

Digging Into the Practical Applications Practical Examples / Use Cases of Predictive AI

Predictive AI enables proactive customer service by anticipating customer needs or issues and enhances workforce optimization and resource/people planning. It is primarily focused on identifying patterns, trends, and relationships in data to forecast what might happen next and to anticipate needs, personalize interactions, identify issues before they escalate, and streamline operations based on customer/employee behavior predictions.

In practical terms, this could look like predictive routing, i.e., identifying aspects of the customer’s needs or history and the employee/agent’s skills/knowledge/expertise and routing the customer to the person best suited to help them at that moment.

In addition, when employees have data-driven recommendations or next-best-action suggestions, it improves the decision-making process and helps them to know the next-best actions to take – whether they are looking to enhance their sales performance capabilities (think of a co-pilot) or help the employee know what next to do for the customer, so the experience is better and then the experience drives higher customer satisfaction and loyalty.

In addition, Predictive AI can help determine which customers are at risk of churning to enable proactive measures to retain them, enhancing customer satisfaction and loyalty. And, in terms of staffing, it can analyze historical data to identify trends, forecast demand, anticipate peak interaction times, and plan for resource allocation and management.

Case Study: Harley-Davidson, using their Predictive AI tool they called Albert, drove in-store traffic by generating leads from customers who expressed interest in speaking to a salesperson by filling out a form on the dealership’s website.[i] Harley-Davidson could identify, by using predictive analytics, which customers resembled previous high-value customers and were more likely to make a purchase. By the third month, the dealership’s leads had increased 2930%, with 50% of them being lookalikes.

Case Study: Sephora creates personalized profiles for each customer based on their purchase history and preferences. Predictive analytics then sorts through that data to predict the products customers need and puts them in a customized “Recommended for You” section on its home page.[ii] Sephora sends targeted rewards and marketing messages, resulting in 80% retention of customers.

Practical Examples / Use Cases of Conversational AI

Conversational AI can enhance the customer experience by providing 24/7 support bots and instant responses through messaging platforms or voice calls and reduce the workload on employees by handling routine inquiries and questions. Other examples are voice IVRs, voice chatbots, Amazon Alexa, Apple Siri, Google Assist, and Microsoft Cortana.

Case Study: TechStyle Fashion Group’s Global Member Services, which supports brands like Fabletics, Savage X Fenty, JustFab, ShoeDazzle, and FabKids found many customers were making the same type of simple transactional requests. It didn’t make sense to have them wait in a queue for a live agent. At the same time, employees were experiencing frustration, repeatedly answering the same “simple” questions.

To reduce member/customer and employee frustration, they used Predictive AI to personalized experiences to listen, understand, and predict the intent of the request. It divided order requests into four main intents: skip, cancel, WISMO (where is my order), and update my profile info. They then used Conversational AI to answer the simple transactional requests with self-service (AI voice and chatbots) allowing members to get what they needed with less time, friction, effort, and hassle, and employees were freed up to use their skills to solve more complex problems.[iii]

Practical Examples / Use Cases of Cases of Generative AI

The transformers (the “T” part of ChatGPTs, BrandGPTs, or BERTs) are part of what gives rise to the landmark developments in the accuracy, performance, and usability of numerous natural language tasks which enhance capabilities in understanding text, answering questions, text generation, and performing sentiment analysis. As a result, Generative AI can produce high-quality, customized emails, proposals, blog posts, or social media updates. It can ingest a Word document and using the Transform Command automatically convert it to a PowerPoint presentation.

Generative AI can improve the response time and consistency as well as enhance personalization in customer interactions by automatically generating contextually relevant and personalized responses, reports, and meeting summaries with task lists. The content that gets generated sounds like it was created by a human, but the employee didn’t have to spend time writing it (though they need to review it.) It can also be used for simulated dialogues or customer scenarios to train employees to prepare for real-world interactions.

Case Study: A large bank looked at their self-service capabilities, built using Conversational AI chatbots whose underlying technology was intent-based models that leveraged NLU. With the brand’s ever-changing products and services, it took a long time to add new information and train the models. The conversation flows required a developer to imagine all the things the customer might ask, collect hundreds of training utterances, and conduct careful optimization and curation to achieve acceptable accuracy. And in banking, you have to be precise.

The time, cost, and level of effort were very high, but at the time Conversational NLU-based chat bots were the only choice available. While it was better than what they had before, the limitations of the technology meant that the number of chats transferred to agents was high as customers either didn’t trust what the bot told them, or they couldn’t get what they needed.

To fix this, the bank moved its chat capabilities to a chatbot platform built on an LLM. However, they didn’t use an LLM that sourced answers and information from the whole internet, like ChatGPT. Instead, it used what is referred to as a BrandGPT, where the content for the interactions comes from internal brand sources like knowledge bases, content management systems, the brand’s URL, and emails, and is contained in a wall-garden or private instance.

Using a chatbot built on a Generative-based (BrandGPT) drastically changed the customer’s perception and satisfaction with virtual agents with CSAT increasing by 15% and transfers to live agents reduced by 20%.

To understand the difference between the experiences created we can compare the previous-generation NLU-based chatbots to an infant with no language understanding. In contrast, a Generative-based, BrandGPT-based bot would be like talking to an adult level with comprehensive language development. GenAI is allowing brands to begin to replicate an in-person experience in a digital environment allowing them to rethink websites, and mobile apps that have been constrained by the underlying technology they were built on.

?Case Study: A financial services company was looking to transform its sales staff. They implemented GenAI tools and found productivity gains, i.e., substantial time and effort savings including:

• Meeting Assistance- Summaries of meetings and generation of next steps lists and reminders
• RFP Responses – Deal strategy, pricing, and tailoring customer-specific responses guidance
• Deal Support- Guides for deal and pricing strategy
• Sales Collateral – Generates tailored content, call scripts, emails, and objection responses
• Knowledge Management – Provides specific information about products and how to sell them
• Live Coaching – Provides real-time, personalized recommendations

The employees were not resistant to the adoption of GenAI because the leadership team ensured the staff the goal was not to reduce headcount. Using a solid organizational change management program (including stakeholder readiness, skills gap and job workflow assessment, training, and clear communication planning) alongside the technology deployment, the employees trusted this new process. The result was an increase in productivity and the extra time employees gained was used to serve the increase in net new customers and additional business from current customers, increasing the firm’s overall revenue by 20%.

The Technology Underlying Predictive, Conversational, and Generative AI

Each type of automation and AI can provide something different in terms of capabilities and business benefits because it’s based on different types of underlying technology. While you may not really want to know what’s under the hood, it’s good to have a basic understanding.

Here’s a general overview so when you are asking vendors about their solutions, you at the very least have an understanding that there are differences and even ask them what how their applications work and put them into perspective.

While different types of AI technologies have some overlapping underlying technologies and can complement each other in applications, they also have distinct characteristics and use cases. Let’s look at some examples.

Similarities of Predictive, Conversational and Generative AI

• AI-driven: All three are underpinned by advanced AI and ML technologies.
• Data-reliant: They rely on large amounts of data for training and fine-tuning.
• Continuously Evolving: As they are exposed to more data, the algorithms are refined, and they improve with time.
• Enhance Decision-making: Whether through predicting future trends, facilitating communication, or generating new content they all ultimately aim to support better decision-making.

Differences between Predictive, Conversational, and Generative AI

• Primary Focus: Predictive AI focuses on forecasting and risk assessment, Conversational AI on understanding and generating human-like dialogue, and Generative AI on creating new content or data or information.
• Underlying Technology: While there's overlap (e.g., NLP is a component of both Conversational and Generative AI), each area has distinct technological frameworks and models.
• Output: Predictive AI provides insights or forecasts, Conversational AI provides responses or interactions, and Generative AI provides new, original content.

The Underlying Technologies of Predictive AI

Predictive AI primarily relies on a blend of Machine Learning (ML), Natural Language Processing (NLP), Computational Analysis, and Speech Recognition, with each playing a crucial role in enabling systems to analyze historical data, identify patterns, and make predictions about future events or behaviors.

• Natural Language Processing (NLP) helps to comprehend and process human language using both Natural Language Understanding (NLU) and Natural Language Generation (NLG).
• Machine Learning (ML) provides predictive capabilities by learning from data.
• Computational Analysis supports these technologies by offering the tools and methodologies to effectively process and analyze large datasets.
• Speech Recognition converts spoken language into text, enabling the AI to process and respond to voice interactions.
• Together, these technologies form the backbone of predictive AI, enabling it to deliver accurate predictions and valuable insights.

The Underlying Technologies of Conversational AI

Conversational AI is made up of technologies which include NLP, NLU, NLG, and Speech Recognition.

• NLP is a field of AI that gives machines the ability to read, understand, and derive meaning from human language and involves tasks like speech recognition, language translation, sentiment analysis, and understanding the nuances, intentions and implied meanings in a conversation. NLP uses machine learning algorithms, rule-based systems, and statistical methods to process language at various levels (phonetic, syntactic, semantic, pragmatic.)
• NLU is a subfield of NLP that focuses on the machine's understanding of the input given in natural language. It deals with the comprehension of intent and context to understand the nuances, intentions, and implied meanings in a conversation using parsing, semantic analysis, and entity recognition to interpret the input accurately.
• NLG is responsible for creating a natural language response from the system's internal knowledge or data using templates, machine learning, and linguistically rich databases to generate human-like text.?
• Speech Recognition, also known as automatic speech recognition (ASR), computer speech recognition, or speech-to-text, is the process that enables a computer to identify and respond to the sounds produced in human speech. It essentially converts spoken words to text, enabling voice input for conversational AI using acoustic modeling, language modeling, and signal processing to process and understand human speech.

The Underlying Generative AI Technologies

Generative AI, especially in the context of creating text, images, or other forms of content, predominantly relies on advanced machine learning architectures known as Generative Models. It uses large language models (LLM) which are a type of AI that is trained to take large amounts of data to understand, generate, and interact with human language to generate new content that is similar to the structure and theme of the input data but results in something new.

These models are called "large” because they consist of a massive number of parameters (the parts of the model that learn from data) and are trained on extensive collections of text data. In particular, NLG is used to generate human-like text based on the context. The Generative Models, like GPT-3, GPT-4, or custom models are trained on specific types of data sets so that they can generate content specific to a particular use case, i.e., customer service scenarios.

Examples of companies working on different types of generative models and LLMs (and major investors at the time of this writing) are:

• OpenAI (Microsoft)
• PaLM (Google)
• Apple (Ajax)
• Claude (Amazon/Anthropic/Google)
• Llama (Meta)[iv] and [v].

Note not all models are the same, nor are their outputs. It’s important to know what you want the model to do so you can pick the right one(s.)

The Deep Learning Models include neural network architectures and transformer-based models like:

• GPT (Generative Pre-trained Transformer)
• BERT (Bidirectional Encoder Representations from Transformers) for text generation and
• DALL-E for image generation.

The transformers allow the model to better understand the meaning of the text by using attention weights assigned with more importance to the most significant parts of the input and less importance to the less relevant parts. The transformer also adds positional encoding when encoding each word. This is why they can understand, interpret, and generate human language in a way that's more accurate than ever before.

What is Prompt Engineering?

Prompt engineering is the science of inputting a text request into a generative model and receiving the output. That output could be text or an image. When I first started to experiment with prompt engineering it reminded me of my days as a management consultant, where we learned it wasn’t always what we knew, but how to ask the right questions.

I created the images at the top of this article by providing ChatGPT with a text description of RPA, Predictive, Conversational, and Generational AI and asking it to provide a visual depiction of each based on the provided description.

The images were generated by DALL-E, which is part of ChatGPT-4. It created four different diagrams, which I combined into one for this article. By strategically crafting and optimizing input prompts, I was able to effectively interact with the generative model to produce a cool image that I didn’t have the skills to create myself and it did it in less than 4 seconds. (Note I had asked it to include the titles of each section. While the images were really cool, the words were not spelled correctly, so I added them myself later.)

There is definitely a learning curve to get good at prompt engineering. It combines your human creativity, linguistic precision, and an understanding of the AI's mechanisms to harness its full potential. In this case, I experimented with how to guide the model more effectively, i.e., I didn’t get these diagrams on my first try. I played around with prompts and redirected ChatGPT-4 by refining what I was asking the model to do until I was able to get something close to what I was imagining.

In terms of an application for a contact center? Data scientists would use prompt engineering to design questions or prompts that could accurately understand and respond to a customer's intent. This involves crafting prompts that can extract meaningful information from the customer's input, ensuring that the AI correctly interprets the context of the conversation. Prompt engineering could also be used to refine the AI's responses, so they are not only accurate but also delivered in a manner that is empathetic, professional, and aligned with the company's brand voice. This involves training the AI to understand the nuances of human communication and respond in a way that enhances customer satisfaction.

What is RAG (Retrieval Augment Generation)?

You may also have heard of the term Retrieval-Augmented Generation or RAG. In a generative model, it’s a process used in NLP and ML that combines the retrieval of relevant documents or information with a generative model.

Here’s How RAG Works:

There are two steps. First, the model retrieves documents/information relevant to a query or context. In the second step, the retrieved documents/information is provided to a generative model, often an LLM like GPT-4 or a custom variant. It uses the context from the retrieved content to generate a coherent and contextually appropriate response or output.

Applying RAG

One application of RAG is to power sophisticated chatbots or virtual assistants/agents. For instance, when a customer asks a question, RAG retrieves relevant information from company documents, FAQs, or prior customer service interactions, and generates a response that is tailored to the customer's specific inquiry. You have control over what is produced based on the input the model is fed. The old adage of garbage in, garbage out applies here.

Let’s say we were answering a technical support question or something that requires deep, factual, and specific knowledge. RAG ensures the responses are not only contextually relevant but also factually accurate, drawing directly from the source input material. As you speak to vendors, you might want to ask about “walled gardens.” This refers to creating a specific instance/source from which the content is drawn from vs. the whole internet. This can help with the issue referred to as hallucinations, where the model might make stuff up because the source of the information could be everything on the internet. ?

In addition, RAG can generate responses that are highly personalized by retrieving and leveraging customer-specific information, while adhering to privacy standards.

RAG can also assist customer service agents by quickly retrieving information from a vast internal knowledge base, thereby reducing response times, and improving the accuracy of the information provided to the customer.

?And, as new data and customer interactions are incorporated into the knowledge base, the retrieval component of RAG can help the system stay up to date, ensuring that both the retrieved information and the generated content are current and relevant. This is especially important if you want to design a customer service chat bot to not only respond accurately but also use your brand voice.

Wondering What Your Next Steps on Your AI Journey Should Be?

As with any technology decision you have to know what you want the technology to accomplish. And that means we need to start with the business basics.

Step 1: Revisit your values, mission, and purpose.

• Why are you in business? Who are your customers?
• How is what you deliver, the products and services, in service to your customers?
• Is this the business you still want to be in? Or is it time to pivot?

Step 2: Choose a strategy that will best serve your value, mission, and purpose.

• Are you looking at headcount reduction or innovation, differentiation, and a competitive advantage?
• Many companies looking at AI want to use it to reduce headcount. However, how many employees are going to get on board for a project that may put them out of a job?
• If the strategy is to value your employees and NOT lay anyone off, you will find employees will have an open mind. With that gain in productivity, a business can get from AI, you will also find you can use it to do the things the company hasn’t been able to do to grow and innovate.

Step 3: Choose an AI vendor that can help you take that strategy and execute it into measurable goals and KPIs.

• If they can’t explain their technology in simple terms, then maybe they aren’t the right ones for you.
• Make sure you are clear on what you are buying and what its intended business outcomes are going to be—otherwise, you can expect to be disappointed.
• Also know that especially when it comes to GenAI, many of the vendors who are truly doing GenAI are at the very beginning of this journey. Don’t be surprised or alarmed if they tell you they don't know yet everything it can do. That is an honest answer.
• And know that the good ones have figured out many use cases that can be applied to customer and employee experiences.
• May want you to join the vendor in the exploration of applying GenAI by building special use cases for your business.
• Embrace the Age of AI as an opportunity to learn and grow; depending on your tolerance for risk and experimentation, start with a pilot project in a particular area or develop it in a skunkworks situation.

Step 4: Start.

• Commit to a learning and agile business mindset.
• Evaluate and pivot as necessary.
• Release your inner data scientist. No matter what level you are at – CEO to frontline worker, just do it.
• Begin to get a visceral experience of how powerful an LLM can be by experimenting with ChatGPT or Bard, Claude, or other user interfaces to LLMs and generative models

Step 5: Remember, your AI strategy and your culture come before AI technology. It's only then we can get the ROI from the tech...

?"We can imbue technology with our hopes and dreams for a future focused on bettering humanity. We just need to understand what we are optimizing for and why." – Dr. Natalie Petouhoff

??REFERENCES:

[i] https://hbr.org/2017/05/how-harley-davidson-used-predictive-analytics-to-increase-new-york-sales-leads-by-2930

[ii] https://d3.harvard.edu/platform-rctom/submission/beauty-in-the-age-of-individualism-sephoras-data-driven-approach/#_ftn1

[iii] https://www.customercontactmindxchange.com/ai-meets-global-fashion-brands-with-fantastic-results/

[iv] https://hackr.io/blog/best-large-language-models

[v] https://techmonitor.ai/technology/companies-large-language-models-llms-chatgpt-alternatives#h-anthropic-claude