Augmenting contact centres through Gen AI & automation

Businesses are increasingly exploring generative AI (Gen AI) initiatives. The advantages of Gen AI projects include the ability to extract value from previously inaccessible data and the potential to revolutionize existing processes for added value.

Executives expect a 7-9% improvement in productivity from Gen AI adoption in the next three years¹ and 71% believe it will enable them to create more interactive experiences for their customers.

My article explores the proper utilization of technology and the correct approach to implementing it. It emphasizes the necessity of integrating both Automation and Gen AI to generate substantial business value using Capgemini’s framework for developing business and technology solutions enriched with Gen AI and Automation, illustrated by a case study in the Contact Center domain. It proposes a solution and architecture for a business issue related to customer interactions, addresses some shortcomings of Gen AI, and outlines strategies to mitigate these challenges

From Process Automation to Process Augmentation with Gen AI

How did automation solutions evolve in both the front-and back office and what were some of the advancements and short-comings of these technologies? What is Gen AI and how is it different?

1. Back-office evolution: a glimpse into the evolution of automation solutions

Most of the companies I work with are well advanced in their Automation journey. Initially, they started with rule-based systems like RPA. RPA is great at automating repetitive tasks. The technology was seen as an enabler to realize cost reduction ambitions. Now it is used to enhance revenue and do more with less capacity thanks to evolutions in complementary but more advanced technologies such as Intelligent Document Processing. These solutions have capabilities to deal with unstructured data as well, something a more traditional RPA solution could not do. Clients have reaped the benefits, saving thousands of hours, freeing up staff from mindless repetitive work, so they can focus on higher value tasks.

2. Front-end evolution: a glimpse into the evolution of chatbots

Initially, companies implemented chatbots with menu-driven interfaces, which were efficient for routine tasks but fell short when handling complex inquiries due to their limited options and absence of free text input. As technology progressed, rule-based chatbots emerged, equipped with conditional logic to manage predefined questions effectively. However, their capabilities were confined to the content they were programmed with, making them still inadequate for complex questions and subtle distinctions. Conversational AI chatbots marked a significant advancement. These bots utilize Natural Language Processing (NLP) to discern questions and answers, and Machine Learning (ML) to tackle more intricate queries, thereby enhancing the bot’s accuracy and performance. They possess the ability to comprehend user intent and sustain a conversation. A notable limitation of this technology was its tendency to trap users in repetitive loops, asking the same questions repeatedly.

3. How Gen AI enhances chatbots and redefines customer experience

Gen AI redefines customer experience as it has some attributes which makes it different from other Chatbot technologies:

1. Enhancing accessibility and inclusivity: are configured to communicate multiple languages, breaking down barriers for customers who prefer communication in their native language.
2. Immediate answer based on contextual understanding: process complex language structures, such as context, intent, and sentiment, allowing for more natural and relevant interactions tailored to the specific needs of the customer.
3. Advanced security: are used to monitor transaction patterns and identify unusual activities. By analyzing the customer’s behavior and spending habits over time, the chatbot can establish a baseline and identify anomalies more accurately.

Gen AI and Automation work great together, here is how it works:

1. Efficiency and productivity: create content, designs, code and more, which can then be automatically deployed, managed, or utilized by automation systems. This combination enhances efficiency and productivity by reducing manual intervention.
2. Enhance Decision-Making: analyze large datasets to produce insights, forecasts, and recommendations. Automation can then implement these recommendations quickly and consistently, improving decision-making processes.
3. Scalability: Automation enables the scaling of AI-generated output without requiring additional human resources.
4. Customization and Personality: create personalized content or solutions tailored to individual needs, while automation ensures these personalized outputs are delivered efficiently and accurately.
5. Continuous Improvement: handle repetitive tasks and feedback loops, allowing Gen AI to learn and improve its outputs through iterative processes.

Imagine the power of Automation technologies like RPA combined with Gen AI, enhancing customer engagement, decision-making and enabling efficiency and productivity gains, from front to back office, going beyond customer conversations towards an agentless enterprise and paving the way for more innovative, resilient, and agile organizations.

However, for a lot of us Gen AI, still it is like a high-performance sports car—many people discuss it, but only a few can drive it to its full potential. To make it a success you need a system and comprehensive solutions to cover a broad range of business requirements. There are also some challenges and hurdles with Gen AI that need to be overcome.

In order to do it right, Capgemini has a proven framework to build secure, privacy protecting and reliable high-scale generative solutions. The different elements of this framework are illustrated through a business problem.

The business problem of the illustrative case explained

Companies have communications with their customers, suppliers, and employees. These are typically the operational conversations required within sales, support, and service. Those conversations can be enriched, personalized, and customized with Automation and Gen AI.

Our imaginative client, Hotel Incognito has a lot of these conversations. Today, Hotel Incognito handles these customer transactions mostly through e-mail or support tickets. Then a clerk looks into the inquiry. This is a very time-consuming process. Clerks spend a lot of time searching for the right information in many different applications. Hotel Incognito simply does not have enough human capacity to respond to the customer in a reasonable time. Their clerks are already overwhelmed with handling the many guest inquiries and reservations. As a result, their customers have a poor experience. Did you know that Agents spend 35% of their time searching for information and 15% just transferring data between apps? They also need up to 20 systems to resolve 1 issue.²

Hotel Incognito wants to enhance the hotel clerk by leveraging the capabilities of Gen AI. The objective is to accelerate answering guest inquiries about the hotel, guests and concerns and providing information about hotel amenities and services. After all, hotel clerks play an important role in ensuring that guests have a positive experience.

Hotel Incognito needs its clerks to leverage a holistic knowledge base to handle highly complex queries that are normally routed to agents. The solution must be able to understand the context and intent, as well as find and generate answers, and decide on the next-best-action.

The solution they are looking for is a chatbot with next-generation Conversational AI capabilities which has realistic human-like interactions. It includes a Large Language Model (LLM) interface and a RAG that allows to retrieve information from a holistic database so that a clerk has access to all information to provide an answer to any customer inquiry.

As a results, the clerks will have quick access to customer data and customer insights to reduce handling and wait times. And their workload is reduced.

As an example, a hotel guest could wonder if airport transfer is included with his room.

Solution implementation:

1. The experience is key to drive adoption  

Hotel Incognito has decided to make the everyday task of a clerk handling customer queries easier and decided to build a specialized Gen AI that can personalize the conversations they have with their customers.

To start with, the customer input is captured in a form. The clerk receives the customer inquiry through a Software Agent and will be prompted. What this means is that the clerk gets an interactive application that is dynamic and responsive.

In our case, the customer prompts the question “I’m traveling by plane. Is the airport transportation to the hotel included?”.

The customer inquiry is sent to the Large Language Model (LLM) and the Gen AI solution will then propose an answer to the clerk that he can then validate and sent back to the customer.

There are some important considerations to make for the user interface that the clerk gets:

• Personification to manage user expectations and tolerance for errors
• Query template to provide users with quick access to a library of pre-made powerful queries
• User guides to help users apply good practices to maximize their results
• Users should be able to log issues if hallucination happens

2. Model to learn and adapt

What information might the model need to formulate a response and fulfill the service request? To capture the context, a prompt is used. This is a quick and easy way. Indeed, the customer’s name is available from the user input. But many more data elements might be needed to understand the full context. If the model does not have this data, it will start to generate content that is irrelevant, made-up, or inconsistent. In other words, the Gen AI solution will start to “hallucinate”. For the LLM to give a relevant and accurante response, the model needs to understand the full context of the customer, so information needs to be enriched. We want to minimize hallucinations at all cost and get a human-like response. Moreover, we also want to restrict the use of LLM when the prompt is likely to produce errors. This can be done by instructing the prompt to stay in its domain of validity.

The solution to overcome hallucination, bias and ensuring that the model answers only from verifiable sources from Hotel Incognito is done through RAG (Retrieval-Augmented Generation). RAG optimizes the output of an LLM by referencing an external knowledge base to ground LLMs. It basically gives an LLM a ground truth to start from when it tries to define an answer to a query. As Hotel Incognito data sets are continuously updated, the external knowledge base that the LLM uses must also be updated and this requires a good MLOps pipeline to ingest new data.

The data needs to be collected and gathered across the company, from the appropriate applications in the ecosystem of Hotel Incognito. First, the customer’s loyalty level is needed to understand the rules that can be played by. For this, the solution needs to interact with the System of Action.

3. System of Action to act based on the input

Hotel Incognito has a loyalty platform, but the system has no API. Luckily, a robotic task, through the use of a software agent, can extract the data, and this feature is available in Hotel Incognito’s Automation platform, part of the System of Action. It turns out John Doe has already stayed in Hotel Incognito’s hotels before. So now we know John is a premium customer, eligible for free pickup and drop-off from the airport.

The service that John Doe requested is available in the booking system. Luckily, an API is available and can be integrated through the Automation platform. With that, all the other relevant customer’s booking data is extracted as well (such as the hotel’s address, Antwerp, his history with Hotel Incognito), again through the booking’s system API.

One crucial input that is missing is the Airport location. The LLM asks the user for this input. John Doe responds he will land in Brussels, Zaventem at 1 PM on next week Friday. Now the model has all the information it needs.

All the relevant information is sent to the LLM via RAG embeddings so that the full context is captured and more human like responses are given.

The conversation is understood, and again the model interacts with the System of Action. First, the routing API of Google Maps is called to see what time it takes to collect the customer. Then Hotel Incognito’s transportation system is checked to see if there are any driver’s available as well as a car to pick up John on the date he arrives. Luckily this is the case. The Automation platform sends back the information to the Gen AI model. The message that the Gen AI generates and responds to the customer, includes a next-best-action recommendation as the model knows John has booked massages before:

Hi John,

Glad to hear you are planning your travel! Rest assured, we will make your life convenient and have booked a taxi to pick you up next week Friday at 1PM. The travel will take approximately 1 hour and then we welcome you!
Since you might have had a busy week, could we please you with a stay at our Spa? Massages are now at a discount of 15%, so don’t wait to let us know.
Hope you will enjoy your stay with us and have a safe travel ahead!
Greetings,
Hotel Incognito

How this works all together is explained in the picture below.

4. Guardrails to leverage the promise of AI in safe manner

The open-ended nature of the LLMs can produce outputs that may not align with an organization’s requirements, policies, or ethical guidelines. Capgemini’s framework foresees guardrails to protect sensitive information, manage organizational risk, enable growth, and increase governance.

5. Testing & Trust layer

The Testing & Trust layer ensures the solution is trustable with reliable outputs. Our technical solutions address data and model usage controls.

Conclusion: augmented Contact Center through Gen AI powered by Automation

By leveraging Gen AI, Hotel Incognito’s contact center can be augmented: omni-channel, context aware and with unified data.

The outcomes for Hotel Incognito were multi-fold as they were able to:

• Reduce average handling time (AHT) and reduced manual effort
• Increase first contact resolution
• Improve upsell and cross-sell offer conversion rates
• Increase customer satisfaction (CSAT) and customer loyalty

The real magic happens when generative AI-Powered Intelligent Automation executes. Gen AI and Intelligent Automation work better together, and allow for greater scope, bigger impact, and better results.

Integrating AI into your organization can be more straightforward than you think. At Capgemini, we are ready to guide you through the complexity and support the customization involved in prompt engineering and LLM configuration to achieve the desired accuracy and performance.

---

_{¹ Harnessing the value of generative AI, Capgemini Research Institute}

_{² Research from Forrester: Agent Desktops: The Silent – And Costly – Tax on Your Agent’s Time and Energy}

Interested to know how Capgemini’s Gen AI solutions are driving value for our customers?