Customer reviews are becoming an increasingly common part of online shopping, whether it’s about reading them or writing them. A large survey by PowerReviews revealed that 60% of shoppers would leave a product rating and/or a review multiple times a month (without being incentivised) in 2023, compared with 52% in 2021. In a different survey by the same company, reviews were an essential resource when making purchase decisions for 98% of consumers in 2021—up from 89% in 2018.
Reviews directly help potential customers, but retailers only occasionally glance through them—perhaps looking for glaring product issues or enthusiastic praise. Beyond this, the majority of this valuable data is lost in user-generated content too voluminous for a team of humans to read, leave alone synthesise and remember the opinions they contain. AI, as you’d expect, is perfectly suited to such a task.
The use of AI—especially the subfield known as natural language processing (NLP)—for analysing reviews is part of Voice of Customer Analytics, or VoC. It is a relatively recent development, and has focused on areas such as gauging customer satisfaction and predicting sales and customer behaviour. But over the past few years, we’ve seen companies realising its value and potential:
- Microsoft, in 2023, announced it would use gen AI to summarise customer reviews of apps.
- They later rolled out their “Review Summary” feature for the Edge browser, which—for product pages on many retail sites—presents pros, cons, and a summary of the most relevant feedback for the product.
- Amazon’s product pages have begun featuring AI-generated summaries of user reviews, which distil thousands of individual reviews into concise statements that highlight what customers liked (“Many customers praised the durability and ease of installation”) and disliked (“Some criticised the remote control’s responsiveness”).
- When users begin to write reviews on Amazon, AI dynamically suggests aspects or features that other reviewers touched upon—such as design or battery life—implicitly soliciting feedback that will be relevant for future customers.
- From Ford to Airbnb to Coca-Cola, AI sentiment analysis—which we’ll soon look at—applied to customer feedback has been helping companies understand customers and improve CX.
Clearly, this application of AI in e-commerce is of immense, immediate value. The following sections will delve into the processes involved in AI-powered customer feedback analysis as well as the benefits to businesses.
What AI Does When It Reads Reviews
AI that reads customer reviews essentially performs unstructured data processing: It extracts meaning from data that doesn’t have a defined format. AI sentiment analysis, the core NLP application here, assesses the overall feeling behind each review. Expressions of opinion are categorised into positive, negative, and neutral statements.
Natural language processing models can detect—with precision that increases by the day—sarcasm, irony, and nuanced language, which goes beyond basic AI sentiment analysis. For instance, “Their amazing customer service really helped me by hanging up!” would be correctly identified by today’s models as negative, not positive.
In a review that explores different aspects of a product, the AI needs to pinpoint what the reviewer was being positive or negative about. This is where the system breaks down reviews into themes such as “battery performance,” “ease of setup,” or “after-sales support.” The process requires natural language processing techniques for topic modelling, which clusters mentions of similar topics. For instance, if multiple reviewers mention slow customer support response, those mentions form a distinct cluster that signals a common pain point.
Here are examples of practical insights that might arise from the analyses we’ve described:
- Identification of changing customer opinions, over time, about a product might reveal increasing dissatisfaction related to battery life after a software update—which can prompt the company to roll back the update.
- If AI identifies a consistent discrepancy between customers’ experiences and their expectations based on product descriptions or marketing materials, the business can revise how they present their products online (and thereby reduce returns).
- Spikes in the number of instances of certain words (“flimsy” or “fragile,” for instance), in reviews for a single product, can alert a manufacturer to a subtle change in material quality or a recent lapse in quality control. Such granular analysis allows companies to spot problems early; it enables proactive responses that promote brand reputation and customer loyalty.
Additionally, AI can perform keyphrase extraction and summary generation to pull out the most salient phrases and create concise review summaries.
Review Weighting Before Deep Processing
Not all reviews are created equal. AI that reads customer reviews evaluates them for weighting and prioritisation before deep processing: It determines how useful a review is likely to be from the point of view of the business.
Review Length: While micro-reviews like “It works well for me!” contribute to overall sentiment, an AI that reads customer reviews typically wouldn’t accord them the same importance as detailed ones. They might be flagged, before deeper analysis, as less important.
Apparent Reviewer Engagement: Reviews with an excessive number of typos, nonsensical phrasing and such can indicate lower engagement—or they might be spam. AI systems analyse linguistic patterns, review length, and reviewer history to deprioritise or flag reviews that seem less than genuine. Conversely, reviews that are well-structured and detailed are given a higher weighting in the analysis. Some systems might even use a “review helpfulness” score—perhaps derived from user upvotes—to prioritise analysis of more useful content.
Level of Objective Detail: Beyond AI sentiment analysis and topic modelling, there is a need to prioritise reviews that offer the most constructive feedback for product development, service improvement, and marketing adjustments. As one example of “more valuable” and “less valuable” reviews, consider star ratings:
- Five-star reviews signal overall satisfaction, but the customer’s satisfaction itself might veer the details away from neutral and objective. Four-star reviews tend to also explain why the product wasn’t perfect, pointing out minor flaws or unmet expectations. AI can extract specific “buts” and “ifs,” which are goldmines for product teams looking to refine.
- One-star reviews indicate that a product suffers from major issues, but the content often emphasises complaint over detail. Two-star reviews, too, indicate major issues—but they are more likely to explain why the product was a failure (and not a total failure). They might mention specific pain points, or what combination of minor issues led to dissatisfaction. This level of detail is crucial for diagnosing complex product issues.
AI That Reads Customer Reviews: How Do Manufacturers Benefit?
Manufacturers have historically gathered customer feedback from formal surveys, focus groups, and direct customer service interactions. These methods have their limitations: Small sample sizes, potential selection biases, and customer reluctance to articulate nuanced experiences directly to a company.
A paper titled Using AI to Track How Customers Feel—In Real Time spells out that last issue. To paraphrase: “Qualitative surveys miss critically important feedback. Customers often reveal their true thoughts and feelings in the open-ended comment boxes typically provided at the end of surveys; AI can help companies make use of this valuable data.”
Reviews on platforms like Amazon, in contrast to survey responses, are spontaneous, unsolicited, candid opinions from millions of actual customers. AI-powered review sentiment analysis offers manufacturers direct, scalable access to these authentic consumer perceptions. A prime example of effective applications of NLP in a business context, it provides a clear view into what customers genuinely value and what truly frustrates them—which can inform product development and quality control.
If there has been a consistent pattern of criticism regarding a product’s “awkward user interface,” for instance, or “flimsy build quality,” product teams don’t just get a vague or unsubstantiated idea; they receive quantified feedback tied to specific product aspects. The insight can trigger immediate internal investigation: Design engineers, UX specialists, and quality control teams can collaborate to identify the causes of the relevant manufacturer shortcomings. The result could be a component redesign, a simplification of the user flow, or a review of manufacturing processes to ensure material consistency.
Consider a concrete example of the power of AI that reads customer reviews to drive tangible product improvements: Say AI-generated insights help a smartphone accessory manufacturer identify consistent criticism of the durability of a certain charging cable’s connectors. The company can quickly analyse whether the criticism correlates with particular production batches or suppliers. If a link is established, the manufacturer might switch to a more robust supplier—and tighten quality control protocols for that component.
Voice of Customer Analytics: An Imperative for Online Retailers
Online retailers are at the sharp end of customer interaction, which makes reviews a direct barometer of their operational efficiency and the extent to which products are accurately represented on their platforms. Voice-of-customer analytics—or, specific to the current context, AI that reads customer reviews—helps retailers with crucial service-level insights.
If, for instance, a good proportion of customers criticise slow response times from customer support, retailers can optimise staffing levels during peak complaint hours, streamline support processes, or implement targeted retraining for their teams.
Similarly, AI analysis might highlight persistent issues with product descriptions, or other website content that confuses customers or sets unrealistic expectations. By identifying recurring confusion stemming from misleading product images, for instance, a retailer can revise and enhance its product listings. This might result in fewer product returns, greater customer trust, and improved repeat business.
As a practical illustration of AI in e-commerce for improving operational efficiency, say AI analyses reveal that customers consistently express disappointment upon finding that a portable speaker set is smaller than the product page suggests. The retailer can use this information to update product imagery—perhaps including scale references (placing the product alongside a common object), or add contextual descriptions (“ideal for travel”). These actions would help manage customer expectations, reduce returns, and enhance trust.
Better Marketing Using AI Sentiment Analysis
Marketers’ success in crafting compelling brand narratives and driving engagement often hinges on an understanding of authentic customer perceptions. AI analyses of customer reviews offer marketers an unparalleled, unfiltered view into how consumers truly perceive their products or brands. They bypass the potential biases inherent in traditional market research methods such as surveys.
Unsolicited reviews can reveal subtle emotional associations, highlight unexpected product strengths, or expose unsuspected weaknesses—all of which marketers can strategically integrate into their communications. This is a powerful application of AI sentiment analysis beyond simple “positive/negative” categorisation.
Furthermore, marketers gain an enormous competitive advantage through AI-driven competitive analysis. By systematically analysing reviews of rival products, AI can clearly identify specific competitor weaknesses—such as “inferior battery life” or “a confusing interface”—that marketers can leverage in their own advertising and positioning strategies. Conversely, understanding competitors’ strengths helps marketers refine their unique selling propositions and effectively differentiate their offerings.
For a real-world scenario that highlights the benefits of AI that reads customer reviews, consider a home-appliance brand whose marketers are shown a consistent pattern of mentions of noisy operation in a competitor’s dishwashers. Recognising this pain point, the brand’s marketing team might position their own appliance as “whisper-quiet,” prominently featuring this benefit across marketing channels from product descriptions to social media campaigns.
In Sum
The sheer volume of customer reviews on e-commerce sites represents a wealth of hidden data that is impossible for human teams to process at scale.
Companies can use AI to transform all of this unstructured text into actionable intelligence. This process requires natural language processing techniques including sentiment analysis and topic modelling—and sophisticated processing of review quality and nuance. It gives manufacturers insights for product refinement, empowers online retailers to enhance the customer journey, and equips marketers with authentic perceptions for impactful campaigns.
As the relevant AI techniques become capable of discerning even more subtle emotional cues, identifying emerging trends through predictive analytics, and integrating with other business systems, they will empower companies to not just react to feedback but proactively anticipate customer needs.
References
- Artificial Intelligence in Marketing: Two Decades Review (Sage Journals)
- How Amazon continues to improve the customer reviews experience with generative AI (Amazon)
- Microsoft Store will use AI to summarize app reviews by customers (GeekWire)
- Review Summary (Microsoft)
- Survey: The Ever-Growing Power of Reviews (PowerReviews)
- Survey: The Ever-Growing Power of Reviews (2023 Edition) (PowerReviews)
- The role of AI in content sentiment analysis for customer feedback (AIContentfy)
- Using AI to Track How Customers Feel — In Real Time (Harvard Business Review)
- Voice of customer analytics (Qualtrics)
- What Motivates Shoppers to Write Reviews in 2023? (PowerReviews)
Table of Contents
