If you’ve ever looked at product reviews on Amazon, you’ll notice that most consumer products have a J-shaped rating distribution. That means more people will leave 1-star and 5-star reviews than 2- or 3-star reviews. Why is that? In this business study from Temple University, the authors attribute this phenomenon to two types of self-selection bias: purchasing bias and under-reporting bias. People only purchase products when they expect them to deliver value, and they typically only leave reviews if their expectations are either greatly exceeded or sorely missed.
The average star rating is a poor measure of product quality. However, we learn something unique about consumers through verified reviews: their shift in sentiment from pre-purchase to post-purchase. The greater the shift, the more likely they will leave a highly positive or overtly negative review. The contextual clues in the reviews will indicate why the shift in sentiment occurred, whether the issue is due to specific product attributes, or improper expectations due to false marketing.
Natural language processing platforms like Native AI specialize in identifying consumer trends from public data such as product reviews, interviews, and surveys. “It’s important that brands and researchers recognize that what these star ratings tell you is why someone has become a promoter or detractor for a brand,” says Native AI CEO Frank Pica. “Word of mouth is one of the best marketing channels. People are unlikely to proactively recommend a product to their friends unless they are truly wowed by the experience.”
The data is clear: according to a Hubspot report, 55% of consumers report learning about products via word of mouth, and 40% made purchases based on these recommendations. Native AI believes that the key to predicting future purchase behavior at scale is understanding how to close the gap between consumer expectations and reality. “Receiving a 5-star rating often means that the consumer was pleasantly surprised by the positive experience, and ironically, that means that the brand has room for improvement regarding highlighting their best features through packaging and marketing,” says Native AI Head of Data Science Enes Gokce. “An abundance of 5-star reviews that cite a specific feature or theme indicates that product sales have yet to reach an optimal level.”
However, it’s not enough to identify common keywords. Contextual clues such as the tone, language, and circumstance make a huge difference in interpreting the weight of the sentiment. Native AI has created a solution to probe deeper into the overall themes. Using generative AI, Native AI clients can create Digital Twins for their brand’s customers. Using a proprietary relevance scoring system, the Digital Twins can answer almost any question with strong predictive accuracy, capturing not only the opinions but also the tone and weight of the statements.
The ability to ask hypothetical questions is important because sentiment is never black-and-white. According to Native AI, the average star rating for reviews containing “neutral” sentiment statements is typically around 4 out of 5 stars. The average star rating for reviews containing “negative” sentiment statements is typically between 2 and 3 stars. This means that many people who rated a product quite high, including 3 and 4 stars, express some degree of negative sentiment within their reviews. Negative sentiment within a 4- or 5-star review is almost always tied to a specific factor, whereas negative sentiment within 1-star reviews is often generalized and less actionable.
One way that natural language processing has advanced significantly in recent years is the ability to detect opportunities to shift future behavior. For example, if a consumer says, “I don’t often buy this product,” the machine learning algorithm can learn from the surrounding context how likely it is for that consumer to change their purchase frequency if a given factor differs. In the past, Data Scientists used labor-intensive approaches to train classification algorithms to decide whether specific factors such as price are important. An efficient way to get the answer you need would be to simply ask, “What would make you purchase more of this product?”
Digital Twins can provide descriptive and predictive responses, meaning they can surface direct quotes from the source data or make plausible assumptions based on available context. Native AI has created the Synthetic Output Slider, which allows clients to toggle between high-fidelity, Balanced, and High Creativity responses. When the toggle is set to High Fidelity, Digital Twins only provide responses based on known data with 100% certainty. In higher-inference settings, however, Digital Twins will respond if there is sufficient detail within the source data to make a prediction.
Predicting future purchase behaviors at scale is challenging because it requires a deep understanding of consumer purchase intent and expectations. Star ratings fall short in capturing consumer perception of product quality, but recent natural language processing advancements make these data sets more actionable. To learn more about how Native AI can help you enhance your market research studies with consumer trends and predictive consumer behavior analytics, visit gonative.ai.
Published by: Martin De Juan
