How LLMs and Conversational Analytics Are Transforming Customer Service

The digital age has had its share of pros and cons, depending on who you ask. However, one sector that certainly isn’t complaining about is marketing. The advent of Large Language Models (LLMs) has created more touchpoints for collecting and analyzing customer data. This data can be used to understand customer behaviors and patterns, improving companies' service.  

If we break the customer journey into three stages — pre-purchase, purchase, and post-purchase — we can identify various channels to reach customers at each stage. The opportunity lies in how AI and LLMs can collect and use data to enhance customer experience over time.  

Pre-purchase cycle: Attracting and educating customers  

The pre-purchase cycle is when most marketing companies try to attract customers by showcasing their products and explaining the benefits. In this stage, customers are typically looking for product information and trying to assess its relevance. This is where LLMs can play an important role.  

A common use case is social media platforms, where businesses run marketing pages to display products. AI-driven chatbots are used to answer questions and engage with potential customers. These bots also gather valuable insights about customer needs and preferences. For example, after COVID-19, Chipotle monitored social media conversations and found that safety and food preparation precautions were key customer concerns. As a result, they adjusted their marketing content to focus on safety, which resonated strongly with their audience. 

LLM-powered chatbots can directly interact with customers, improving the overall pre-purchase experience. A recent cross-industry study found that websites using AI chatbots saw a 23% increase in conversion rates compared to those without chatbots, providing instant responses, clearing up doubts, and helping customers find the information they need to make a decision.  

Furthermore, LLMs can assist in capturing engagement metrics from social media platforms and websites. By analyzing these metrics, businesses can identify which parts of their website attract the most attention or which products spark the most interest. Other marketing studies suggest that seemingly minor details—such as website background images or the font used in marketing messages—can influence customer behavior. 

Around 95% of users say typography is a critical aspect of web design, and pages with well-chosen, legible fonts have about 15% higher user retention than pages with poor text design. 38% of users will stop engaging with a website if they find the content presentation unattractive. This allows companies to optimize their website and marketing efforts based on data-driven insights. 

Personalization is another powerful benefit of customer analytics. Businesses can tailor promotions and even adapt chatbot responses by tracking individual customer journeys. For instance, a chatbot can recommend a product based on a customer’s previous interactions, providing a more personalized experience and increasing the chances of conversion.  

Purchase cycle: Converting interest into action  

The purchase cycle is the stage where conversions finally happen. Due to the focus on closing a sale, companies often prefer direct interactions, particularly phone calls, over other communication methods. Within this phase, there are significant opportunities to leverage customer analytics to enhance how businesses engage with customers.  

One such opportunity lies in analyzing recorded customer calls. LLMs can analyze thousands of hours of conversations between customer representatives and customers, some of which resulted in a sale and some did not. By feeding this data into a trained LLM model, patterns can emerge, revealing the tactics or formulas that lead to a successful sale. A recent market study found that this analysis can lead to a 7% increase in conversion rates on bts.com. This analysis can help businesses identify which keywords and phrases are most effective and which should be avoided. 

LLMs can also learn how specific phrases impact customer engagement. Research has shown that using clear, precise language—such as referring to a "red sweater" rather than "that item"—generates more positive customer engagement. Another valuable insight is the impact of using a customer's name during a conversation. Personalized interactions, especially when a customer’s name is emphasized, result in higher customer satisfaction and more positive outcomes.  

In addition to phone calls, chatbots represent another key point of contact during the purchase phase. LLMs can learn from previous interactions by identifying common questions and customer issues. They can also identify patterns where customers have trouble with specific products and provide valuable insights to the marketing team, guiding them on how to describe those products on the website better. 

Furthermore, LLMs can assess the sentiment of interactions, recognizing whether a customer is having a positive or negative experience. In cases where frustration is detected, LLMs can prompt chatbots to escalate the interaction to a human representative or even suggest calling the customer directly to resolve their concerns.  

Post-purchase cycle: Enhancing customer support with LLMs  

The final stage of the customer journey is the post-purchase cycle, where customer service typically focuses on addressing complaints or issues related to the product. This is where Large Language Models (LLMs) truly shine. By combining LLMs with customer analytics, businesses can identify common pain points and predict where customers will likely encounter difficulties with the product. 

An important aspect to consider in this phase is the attributes of the customer service agents and their communication style. Research has shown that customers often prefer agents of the same gender and find it easier to connect with those who are more agreeable. 

Interestingly, studies have also found that conflicts are more easily resolved when agents speak with a tone of agreement. These insights stem from analyzing hours of customer service interactions using AI and ML models. Additionally, empathy has been shown to play a significant role in improving customer satisfaction during these interactions.  

Conclusion 

Beyond agent characteristics, regional factors influence customer behavior during the post-purchase phase. For instance, customers in China tend to be more open to AI-powered chatbots as part of their customer journey, while in some developing countries, there is more skepticism surrounding AI. 

In these regions, customers may fear AI replacing human jobs, which can lead to dissatisfaction with chatbot interactions. These cultural and regional considerations are critical for companies to understand, as they directly impact how customer service is perceived and delivered.  

These findings highlight the broader psychological and societal effects that AI-powered customer service can have. As the customer service industry evolves, LLMs and customer analytics empower companies to adjust their strategies to meet customer expectations, providing a more personalized and effective experience across all customer journey stages.