Finding the sweet spot for online product recommendation sales

Research shows how online retailers can boost sales by controlling product information displayed in recommendation systems

A US online retailer with annual sales of around US$100 million faced a problem that resonates with many e-commerce businesses. Despite investing heavily in its proprietary recommendation algorithm, the company’s recommendation-driven sales remained stubbornly below the industry average of 12%. The retailer’s product recommendations on detail pages contributed far less revenue than expected, prompting managers to explore whether the issue lay not in what products were recommended, but in how they were displayed.

This challenge reflects a broader dilemma facing online retailers. With limited screen space, businesses must decide how much product information to display in their recommendation lists. Should they provide comprehensive details to help customers make informed decisions, or limit information to encourage exploration? The answer, according to research, defies conventional wisdom about providing more information to customers.

Researchers examined this question through a randomised controlled field experiment. Published in the Journal of Marketing, the study, Too Many or Too Few? Information Cues in Recommender Systems and Consequences for Search and Purchase Behaviour, was co-authored by Dr SunAh Kim, Senior Lecturer in the School of Marketing from UNSW Business School together with Dr Xing Fang from Royal Holloway University of London and Professor Pradeep K. Chintagunta from the University of Chicago.

“The catalyst for this research emerged from a puzzling real-world problem we encountered with our partner retailer,” said Dr Kim. “Despite investing heavily in sophisticated recommendation algorithms, they were seeing disappointing results – their recommendation-driven sales remained stubbornly below industry benchmarks.”

The research team recognised a growing disparity in the e-commerce landscape. “We know that major platforms like Amazon have achieved remarkable success with its recommendation systems, with recent industry analysis showing that 35% of what consumers purchase on Amazon comes from product recommendations,” said Dr Kim. “Meanwhile, Netflix estimates savings of more than US$1 billion due to its recommender system. However, these tech giants have access to massive data pools and sophisticated machine learning algorithms that most smaller retailers can hardly match.”

This reality has become even more challenging with evolving data privacy regulations. “The introduction of GDPR in Europe and similar privacy legislation worldwide has significantly raised the bar for data collection and personalisation,” Dr Kim said. “Smaller e-commerce players are finding it increasingly difficult to gather the extensive customer data needed for highly accurate algorithmic recommendations, while also navigating complex compliance requirements.”

The research team wanted to explore whether there was an alternative path forward, and Dr Kim said the team was particularly interested in whether online retailers could improve recommendation performance and sales through strategic information design in their recommendation thumbnails. ”Our hypothesis was that even businesses with underperforming algorithms could enhance their effectiveness by carefully controlling how much product information they display. This seemed like a more accessible and practical solution for the majority of e-commerce players that aren’t Amazon or Netflix.”

This approach addresses a fundamental design challenge: given space constraints, how can retailers determine the appropriate amount of information to display in recommendation lists to enhance customer engagement and sales?

When more information means fewer sales

The research team worked with the struggling US retailer to test four different information designs across 47,996 customers from October to December 2019. They manipulated the amount of information shown in product recommendations, creating conditions where customers saw either no cues (product name only), single cues (price or review), or dual cues (both price and review). The experiment tracked detailed customer behaviour across 71,807 shopping sessions, capturing both search patterns and purchase outcomes.

The findings challenged the prevailing assumption that more information leads to better outcomes. The research revealed an inverted U-shaped relationship between information cues and sales. Single-cue conditions, whether displaying price or review information alone, generated the highest sales compared with both more information (dual cues) and less information (no cues).

Learn more: How ChatGPT bias affects product recommendations

This non-linear effect stems from the interplay between search intensity and efficiency, the researchers explained. When customers saw recommendations with both price and review information, they could quickly assess and dismiss products that didn’t match their needs. While this efficiency might seem beneficial, it actually discouraged further exploration. Customers made snap judgements and moved on without engaging deeply with the recommendations.

Conversely, when no information was provided beyond product names, customers faced excessive friction. The researchers observed that consumers in the no-cue condition exhibited significantly higher total product clicks and longer session durations compared with other conditions. This inefficient search process often led to frustration and session abandonment.

The “goldilocks zone” of product information

Single-cue conditions emerged as the most effective approach. When retailers displayed either price or review information (but not both), they created what the researchers described as moderate friction. “Single cues strike a balance, offering just enough information to aid product evaluation while maintaining high search intensity across both recommender and non-recommender tools,” the study found.

Learn more: GMO food labelling: what works (and what doesn’t) for consumers?

The research revealed that single-cue displays increased not only engagement with recommended products but also had positive spillover effects. Customers who engaged with single-cue recommendations were more likely to use other search tools on the site, browsing more extensively and ultimately making more purchases.

This finding held true whether retailers displayed price information alone or review ratings alone, though subtle differences existed between the two approaches. The key insight was that the number of cues mattered more than the specific type of information displayed.

From algorithm obsession to design solutions

Perhaps the most striking implication emerged from the context of the experiment. The retailer’s recommendation algorithm was admittedly underperforming, yet simple changes to information display significantly boosted sales without any algorithmic improvements. “Improving recommender algorithms is typically complex and costly, which can be prohibitive for smaller retailers,” the researchers noted.

The study revealed something that should change how every online retailer thinks about their website design. “We discovered that when customers shop online, they don’t just look at your recommendations and stop there,” Dr Kim explained. “They bounce between your recommended products, your search bar, your category pages, and back again. It’s like shopping in a physical store – customers might start in one section, wander to another, then circle back with new ideas.”

This has huge implications for Australian e-commerce businesses, from major players like Kogan and The Iconic down to smaller specialty retailers. “What we found is that changes to your product recommendations don’t just affect those recommendation clicks – they influence how customers use your entire website,” she said. “Show too much information in recommendations, and customers make quick decisions without exploring further. Show just one key piece of information, and customers stay more engaged across your entire site.”

The practical takeaway requires thinking holistically about your website design: “Instead of improving each feature separately – recommendations here, search bar there, category navigation somewhere else – think of them as an integrated shopping experience,” Dr Kim suggested. “Pick either price or reviews in your recommendations, but not both. This creates just enough curiosity to keep customers exploring not just your recommended products, but your entire website ecosystem.”

This discovery offers hope for businesses struggling with recommendation performance. Rather than investing heavily in sophisticated algorithms or third-party recommendation services, retailers can achieve meaningful improvements through thoughtful design choices. The study demonstrated that non-technical adjustments to recommender systems, like modifying information cues, can substantially boost sales without the high expenses linked to algorithmic changes.

Takeaways for e-commerce professionals

The research provides actionable insights for online retailers. First, resist the temptation to display all available information in recommendation thumbnails. Creating moderate friction through selective information display can increase customer engagement and sales.

Second, consider the interplay between different search tools on your platform. The study showed that recommendation design affects not just direct clicks but also how customers use search boxes, category navigation, and other discovery features. Platform managers should view these tools as an integrated system rather than isolated features.

Third, test information display strategies within your specific context. While single cues performed best in this study, the researchers cautioned that results may vary based on factors such as algorithm accuracy and customer base.

Finally, remember that technical sophistication isn’t always the answer. In an era of AI-driven recommendations and complex algorithms, this research demonstrates that simple design decisions can have profound impacts on business performance. For retailers struggling with recommendation effectiveness, adjusting information display offers a cost-effective starting point for improvement.

Subscribe to BusinessThink for the latest research, analysis and insights from UNSW Business School

“What excites me most about these findings is their immediate relevance across different digital platforms,” said Dr Kim. “Whether you’re managing millions of product recommendations like Kogan, curating fashion like The Iconic, improving grocery suggestions for Coles Online, or enhancing restaurant discovery experiences like DoorDash, the core principle remains valuable: strategic information design can transform how customers explore and engage with your platform.”

The research opens up compelling questions for different digital platforms. “DoorDash’s restaurant recommendations currently display multiple information types simultaneously – but does this efficiency-focused approach inadvertently reduce discovery behaviour? Similarly, The Iconic’s fashion recommendations need to balance inspiration versus information: showing prices might help decision-making, but could it also limit exploration of aspirational brands?” Dr Kim asked.

As digital platforms become increasingly competitive, these seemingly small design decisions about what information to show – and what to hold back – could make the difference between customers who browse extensively and purchase, versus those who make quick decisions and leave.

Interested in understanding how these insights might apply to your platform? Dr Kim is interested in collaborative research opportunities with Australian digital businesses – contact here.