Lesson #4 Revisited: Text Analytics from Machine Learning to LLMs

Explore how large language models (LLMs) are transforming Voice of the Customer (VoC) programs with advanced insights, dynamic topic discovery, and actionable feedback summaries, powered by AI-human collaboration.

Sean McDade, PhD

Sean McDade, PhD

Founder & CEO, PeopleMetrics

When I wrote Listen or Die, text analytics was already emerging as the backbone of Voice of the Customer (VoC) programs. Even in 2017, machine learning (a form of AI) was recognized as essential to making sense of unstructured customer feedback—those open-ended comments that tell you the "why" behind your scores. Machine learning allowed businesses to analyze thousands (or even millions) of comments, uncover trends, and act.

Fast forward to 2025, and we’ve entered a new era of text analytics. Large language models (LLMs) like ChatGPT and Claude have taken the foundational work of machine learning to the next level. These advanced systems have transformed how we process, interpret, and act on unstructured feedback.

Let’s explore three key differences between traditional machine learning and LLMs—and why they matter for VoC programs.

---

1. Deeper Understanding of Context and Nuance

In 2017, machine learning excelled at identifying patterns and classifying feedback. But it had its limitations. These models often relied on keyword-based analysis, which could misinterpret the intent behind a customer’s words. For example, a comment like “The Wi-Fi is killing it” might be flagged as negative simply because of the word “killing.”

Another challenge was accessibility. Machine learning tools required technical expertise, often confined to teams with programming knowledge. This made them less user-friendly for broader CX or market research teams.

How LLMs Improve This:

LLMs, trained on massive datasets, have revolutionized how we understand unstructured feedback. These models grasp the nuance and context of entire sentences and paragraphs, including idiomatic expressions, mixed emotions, and complex sentiments. What’s more, they offer user-friendly interfaces, enabling anyone—regardless of technical skill—to unlock their power.

Example:

A restaurant analyzing customer feedback receives these comments:

  • “The appetizers were amazing, but the main course was a letdown.”
  • “The main course was fine, but the dessert was to die for.”

Unlike traditional models, an LLM understands that the first comment reflects mixed sentiment, while the second is overwhelmingly positive. This level of precision helps businesses address specific issues more effectively.

2. Dynamic Topic Discovery

In 2017, machine learning models could detect patterns and group data using techniques like clustering and unsupervised learning. However, these models often required manual updates to track emerging topics. If customer concerns evolved, someone had to step in to refine categories or themes. This limited the ability of these systems to adapt to rapidly changing customer needs.

How LLMs Improve This:

LLMs take topic discovery to a new level. These models dynamically identify and group emerging themes without requiring predefined labels. They can uncover subtle patterns in customer feedback, making them highly adaptable to evolving trends and preferences.

Example:

Imagine an airline using an LLM-powered VoC platform to analyze feedback. The system notices a spike in comments about seat recline issues—something that wasn’t being actively tracked. The LLM flags this as a new topic, prompting the airline to investigate. They discover a mechanical issue affecting newer aircraft and resolve it before it escalates into widespread dissatisfaction.

3. Summarization at Scale

In 2017, text analytics tools often produced raw data like word clouds, sentiment scores, and theme counts. While useful, these outputs required CX professionals to synthesize insights manually—a time-consuming process, especially for large datasets.

How LLMs Improve This:

LLMs can generate clear, human-like summaries of unstructured feedback, condensing vast amounts of data into actionable narratives. Instead of just listing issues, they provide context and clarity.

Example:

A retail chain receives over 10,000 survey responses in a month. Rather than wading through word clouds and sentiment charts, the CX team gets an AI-generated summary: “Customers love the variety of products, especially in the home goods section. However, 22% of respondents mentioned long checkout lines as a recurring issue, particularly during weekends. Many suggested adding self-checkout kiosks to improve the experience.”

This level of insight empowers decision-makers to act quickly and effectively on behalf of their customers.

The Big Picture: A Collaborative Future

LLMs are in the process of revolutionizing text analytics. Their ability to process unstructured feedback at scale, uncover trends, and adapt to evolving customer language is nothing short of remarkable.

But let’s be clear: they aren’t magic.

Even the most advanced systems still benefit from human expertise to guide and refine their outputs. It’s this collaboration between humans and AI that delivers the best results.

Here’s how humans remain essential to the process:

  • Train and Fine-Tune Models: 
    Modern AI systems can self-train and adapt over time, but their initial training requires human input. People teach these models to recognize industry-specific language and terminology, ensuring the AI understands the nuances of your business. Periodic oversight helps correct errors and align outputs with business goals.
  • Interpret Insights: 
    LLMs can identify trends and highlight patterns, but they can’t decide which actions to take. Humans assess the insights, weigh their implications, and determine priorities. For example, if feedback points to dissatisfaction with a product, a human team must decide whether to improve it, replace it, or phase it out entirely.
  • Build Relationships: 
    AI can analyze feedback and identify pain points, but delivering personalized, empathetic responses is something only humans can do. A thoughtful follow-up call or tailored response can turn a dissatisfied customer into a loyal one.

The Bottom Line

Text analytics has come a long way since 2017.

Machine learning introduced scalability and efficiency, but large language models have taken things to the next level, enabling deeper understanding, dynamic insights, and real-time action.

However, the most successful VoC programs recognize the importance of collaboration. LLMs provide the tools to listen smarter, but it’s up to humans to act thoughtfully and effectively.

Because one truth remains constant: listening to customers is just the beginning. Acting on their feedback is what sets great companies apart.

Latest Articles

Lesson #16 Revisited: Do You Need a Survey Tool or a True VoC Partner?

Lesson #16 Revisited: Do You Need a Survey Tool or a True VoC Partner?

AI is revolutionizing VoC, but human expertise remains essential. Discover how AI-powered platforms enhance insights—and why a strategic Vo...

Lesson #15 Revisited: You Can’t Do This Alone – Building the Right Internal Team for VoC in the Age of AI

Lesson #15 Revisited: You Can’t Do This Alone – Building the Right Internal Team for VoC in the Age of AI

Building a world-class VoC program requires internal alignment—IT, BI, Legal, and Procurement all play a role. Learn how AI is streamlining...

Lesson #14 Revisited: Voice of Customer (VoC) Communications in the Age of AI - A 2025 Perspective

Lesson #14 Revisited: Voice of Customer (VoC) Communications in the Age of AI - A 2025 Perspective

AI is revolutionizing VoC communication. While the fundamentals of clear, consistent messaging remain, AI now amplifies and personalizes ho...