Improving GenAI Accuracy with Master Data Management

ABSTRACT: Discover how master data management (MDM)  provides language models with high-quality enterprise data to improve their response accuracy.

Sponsored by Semarchy.

Since generative AI (GenAI) burst onto the scene, it’s demonstrated the potential to revolutionize many aspects of business. Whether applied to customer service, product management, or data analysis, GenAI can be transformative in sparking creativity and helping people get things done. However, GenAI has also shown us that it doesn't always get things right. This presents a significant risk that companies must mitigate before they can trust the technology and fully realize its benefits. 

In this article we'll look into how data quality impacts accuracy in language models--the branch of GenAI that focuses on understanding and generating human-like text. We'll explore how master data management (MDM) is an essential technique for providing language models with high-quality enterprise data. 

Language Models

Let's start with a brief discussion of large language models (LLM) and small language models (SLM). The primary difference between LLMs and SLMs is their scale and capabilities. LLMs are trained on massive datasets, which enables them to understand and generate text with complexity and nuance. This makes LLMs good at a wide range of language tasks, such as natural language conversation and content creation.

SLMs, on the other hand, are trained, fine-tuned, or prompted with a smaller volume of domain-specific information that's often customized with an enterprise's operational data. Their smaller footprint makes SLMs easier to secure, protecting sensitive and proprietary information. It also allows them to perform faster than LLMs, which is important in customer-facing applications. And with their domain focus, SLMs can produce more accurate and contextually appropriate responses for fields such as healthcare, banking, and customer service. 

MDM's Importance to SLMs

MDM is a natural companion to GenAI. It’s a method of managing the quality of an organization's key data domains, such as information about customers, suppliers, or products. It reconciles and standardizes a business’s data from multiple operational and analytical systems. Companies that implement MDM, have unified, consistent, and accurate views of business data from across the organization. 

The learning process of an SLM and the accuracy of its responses depend directly on the quality of the data it's trained on. MDM helps increase an SLM's accuracy by improving the quality of the critical business data it uses. It provides definitive facts for an SLM to incorporate into its responses. 

An Implementation Example

Let's delve into how MDM improves responses by looking at an example use case. Imagine a credit card company that wants to improve customer service. They just deployed a new GenAI application to help human agents handle inquiries through chat. The objective of the application is to reduce the time agents spend resolving customer issues and improve overall customer satisfaction and loyalty. 

The application's SLM is fine tuned to categorize customer inquiries and analyze sentiment during chats to help the agent understand the customer's emotional state. The key feature, however, is its ability to suggest customized, contextually focused responses to the agent. The application was trained on detailed customer data including the products they have and their customer service history, and a knowledgebase of support problems and resolutions. It bases its recommended responses on these details to help the agent provide fast, personalized service.

The Customer Service Assistant in Action

In this scenario, a customer contacts the service center through chat about a persistent billing issue. Although the customer doesn’t explicitly state his frustration, the assistant uses the customer’s service history and subtle cues in the current chat to provide the agent with a head start before they interact directly with the customer. The assistant’s analysis and guidance helps the agent quickly identify and resolve the customer’s issue, which remained open after two previous interactions.

Inquiry Initiation and Historical Analysis

Customer: "Hello, I'm contacting you again about a problem with my bill that hasn't been fixed yet."

Personalized Acknowledgment

GenAI App Assistant: To start, the app analyzes the chat text and scans the customer's interaction history. Noting this is the third inquiry about the same problem, the app infers potential frustration and urgency, and categorizes the inquiry accordingly.

The app suggests that the agent acknowledge the repeat issue upfront, signaling to the customer that the agent is informed and attentive

Agent: "Hello. I see you've contacted us about this issue before. Let’s get to the bottom of this today. Could you please confirm your account number for me?"

Customer: "Sure, it's 123456. I really hope it can be sorted out this time."

Accurate Problem Identification and Solution Proposal

GenAI App Assistant: Leveraging data on common billing errors and resolutions, the app guides the agent to the likely issue and optimal solution. It also suggests acknowledging the customer's long tenure and possible injured trust level. 

Agent: "Thank you for your patience. It looks like a promotional interest rate was not applied to your account as it should have been. I'll correct this immediately, and ensure your bill is accurate going forward. We truly appreciate your business over the years and want to earn back your trust."

Continuous Sentiment Monitoring and Assurance

Customer: "I appreciate your help. How can I be sure this won't recur?"

GenAI App Assistant: The Assistant recommends that the agent provide specific assurances and steps taken to prevent future issues. It recommends sending a detailed confirmation email.

Agent: "All set. I've added a note to your account for an extra check on your next bill to confirm that the promotion rate is applied. Plus, you'll receive a confirmation email shortly. Is there anything more I can assist you with?"

Customer: “Thanks again for your help. I don’t need anything else right now. Bye.”

Post-Interaction Productivity Boost

GenAI App Assistant: After concluding the chat, the app provides a summary of the interaction that the agent can update and add to the customer's account, saving them time in closing the issue.

The Assistant Without MDM

Integrating MDM with the assistant ensures that the data fueling its sophisticated capabilities is accurate and comprehensive. Consider how this conversation could have gone without an MDM-improved view of the customer's account history. 

For example, this company, like most, has conflicting customer information in different operational systems. MDM helps them reconcile discrepancies that the assistant would otherwise use. Without MDM, duplication in the source data could cause the assistant to attribute the wrong service history to the current interaction or associate the wrong credit card products to the customer. The assistant would then relay inaccurate information to the agent. These preventable data issues would cost the agent more time to sort out while the customer grows more frustrated and dissatisfied.

A Strategic Advantage

Integrating MDM with GenAI offers a strategic advantage that extends beyond mere operational efficiency. This synergy enhances the accuracy and reliability of GenAI applications, leading to improved results. For customer service that means increased customer satisfaction and loyalty. In other contexts it means improved healthcare, less stressful travel, or tailored insurance quotes.

Looking ahead, the importance of integrating MDM with GenAI technologies will grow. It's not just about keeping up with technological advancements but about setting new standards for service and operational excellence in any industry. 

Jay Piscioneri

Jay has over 25 years of experience in data technologies including data warehousing, business intelligence, data quality, and data governance. He's worked with organizations in a wide variety of industries...

More About Jay Piscioneri