American Marketer


Upping the retail experience with conversational search

June 22, 2023

The emergence of generative large language models presents a transformative opportunity to accelerate the development of conversational search experiences. Image credit: Shutterstock The emergence of generative large language models presents a transformative opportunity to accelerate the development of conversational search experiences. Image credit: Shutterstock


By Eric Redman

In today's ecommerce environment, key decision-makers in marketing, retail, media and digital positions face a monumental challenge: capture and retain the attention of the modern consumer in a world saturated with information and choices.

The solution lies in leveraging the power of conversational search to revolutionize the retail experience. Here, I explore the significance of generative language models and conversational search platforms, empowering decision-makers with insights to drive customer engagement, enhance personalization, and achieve exceptional retail success.

Evolution of search
The product discovery and knowledge management industry has witnessed remarkable advancements in integrating natural language features into the search experience.

However, I still observe that most ecommerce users often enter queries consisting of three words or less, leading to potential ambiguities and suboptimal search outcomes.

Moreover, search platforms heavily rely on implicit signals to predict users' goals, making it challenging to accurately discern their intent and transition between different stages of the customer journey.

The emergence of generative large language models presents a transformative opportunity to accelerate the development of conversational search experiences.

By enabling search platforms to behave as intelligent assistants, we can guide users through natural language interactions, asking relevant questions and providing tailored suggestions.

This approach shifts the paradigm from a traditional search box-based model to a conversational framework that fosters efficient information exchange and deepens user engagement.

Eric Redman Eric Redman

Building conversational search experiences
To achieve successful conversational search, it is essential to incorporate key elements of a good conversation.

Here are four principles that can guide decision-makers in creating exceptional conversational search experiences:

Active listening: Active listening goes beyond mere question repetition. It involves asking clarifying questions, offering suggestions and retaining contextual information to avoid redundant interactions.

Non-verbal cues: Although search platforms cannot perceive users' body language, they can still pay attention to implicit signals. By integrating implicit signal analysis, conversational search can leverage additional context to provide more accurate and personalized results.

Staying on topic: Conversational search should prioritize relevance and maintain a coherent conversation flow. Questions and responses should be aligned with the user's current context, including personalized recommendations.

Building trust: Trust is paramount in any conversation. Search platforms must strive to be consistent, truthful, respectful and willing to acknowledge mistakes. This approach enhances user confidence and fosters long-term engagement.

Implementing conversational search
The most effective strategy for conversational search revolves around leveraging generative language models in a task-specific manner while adhering to the principles of good conversation I mentioned above. These are the key steps and tasks to consider:

Hybrid search = Building trust + Staying on topic

Use generative language models to improve query understanding, rewrite queries for better recall and relevance ranking, and train semantic encoder models for hybrid search.

A hybrid search approach combines semantic and lexical techniques to cast a wide net and retrieve the most relevant documents.

Additionally, enable the generative language model to reflect on its previous output, prompting itself to generate improved responses.

Incorporate these memory features so that the responses can benefit from prior reflections.

Generative language models can also monitor user and model responses for respectful content, aligning with ethical considerations.

Descriptions = Active listening + Building trust

Use additional prompts to request the generative language model to describe individual documents in the results list based on the ongoing search conversation.

Enable users to ask questions and provide further instructions, allowing the search platform to respond by asking clarifying questions and refining its understanding of user goals.

Implicit signals prompting = Pay attention to non-verbal cues

Leverage implicit signals from user interactions, such as clicks and filters, to prompt the generative language model and enhance search relevance.

Adopt AI-driven search to remain competitive
Extensive research has unequivocally demonstrated that artificial intelligence (AI), particularly generative AI powered by large language models (LLMs), is an indispensable component of the ecommerce landscape moving forward.

Furthermore, industry experts unanimously concur that AI has the potential to enhance search practitioners' capabilities and drive innovation.

In fact, a recent search relevance survey revealed that 100 percent of respondents considered search relevance to be of utmost importance, with an overwhelming 96 percent acknowledging the challenges in delivering it effectively.

A remarkable 88 percent firmly believe that AI will play a crucial role in delivering relevant search results in the future.

IN THE EVER-EVOLVING landscape of marketing, retail, media and digital domains, the power of conversational search cannot be underestimated.

By embracing generative language models and implementing conversational search platforms, decision-makers can unlock unprecedented opportunities for customer engagement, personalization and retail success.

The future of search lies in harnessing the essence of human conversation, actively listening to users, paying attention to implicit signals, staying on topic and building trust.

Eric Redman is Lake Oswego, Oregon-based senior director for product – data science and analytics at Lucidworks.