The Language of Buying: Deciphering AI Conversations

In today’s digital world, AI assistants like Amazon’s Alexa, Google Assistant, and Apple’s Siri are becoming an integral part of consumers' daily lives. These assistants are frequently used for tasks ranging from checking the weather to searching for product information and even ordering items. 

For marketers, the challenge lies in determining whether a consumer’s interaction with these AI assistants is a sign of genuine purchase intent. If a consumer is considering a purchase, advertisers may want to target them with relevant ads. However, if there is no purchase intent, advertising efforts may be ineffective and could lead to a negative user experience.

Recent research co-authored by marketing PhD student Ziting Liao, associate professor Liye Ma, and Dean’s Professor of Marketing Wendy W. Moe, addresses this challenge by developing an approach to predicting purchase intent based solely on what a consumer says in a single sentence. Through this study, they introduced a model that could help marketers better understand consumer behavior and optimize their advertising strategies. It leverages the idea that words in language are not isolated but are connected in meaningful ways.

“When we just look at a word, we don’t know whether that word carries purchase intent or not just by looking at that,” Ma explains. “However, we speak a language, and in this language, words are connected.”

By analyzing large datasets of consumer utterances, the team identified the connections between various terms and constructed a bipartite network of words. This network was then analyzed using advanced machine learning techniques to assess the relationship between words and key “golden purchase words” such as "purchase," "buy," and "order."

These golden purchase words act as anchors, helping the model to determine intent. If a consumer directly mentions “purchase” or “buy,” the intent is clear. However, the model goes beyond direct mentions and can predict purchase intent even when these words are not explicitly used. By examining how closely other words align with the golden purchase words, the model infers intent indirectly.

To test the effectiveness of the model, they compared its predictions to those made by ChatGPT. The results showed that both models performed equally well with the researchers’ model providing complementary insights and potentially offering a practical and effective tool for predicting purchase intent based on consumer interactions.

By predicting purchase intent more accurately, advertisers can optimize their targeting efforts and allocate resources more efficiently. For instance, a consumer who has a strong intent to purchase might not need additional advertising, resources could be better spent elsewhere. On the other hand, a consumer with moderate purchase intent might be more responsive to advertising, which could nudge them closer to a purchase.

The study opens the door for further research in this area, particularly in optimizing advertising targeting based on varying levels of purchase intent. As AI assistants continue to play an increasing role in consumer behavior, understanding the nuances of intent could further reshape how brands engage with consumers and drive conversions.

Read “Predicting Purchase Intent: Deciphering Customer Interactions with AI Assistants,” at SSRN.