Search More, Disclose Less

Chen Hajaj, Noam Hazon, David Sarne, Avshalom ElmalechAAAI 2013 · DOI

Introduction

Comparison Shopping Agents (CSAs), such as Google Shopping or PriceGrabber, have become indispensable tools for modern e-commerce. By aggregating prices from across the web, they ostensibly help consumers find the best deals. However, the business model of these agents—relying heavily on referral fees—creates a fundamental conflict of interest. If a user finds a price but continues searching on a rival platform, the original CSA loses its commission. This paper explores a counter-intuitive strategy for CSAs to maximize revenue: strategically hiding prices to manipulate user expectations and discourage further search.

Modeling the Search Problem

The authors frame the buyer's journey as an "optimal stopping problem." A rational buyer will continue searching as long as the expected gain from one additional search (the "critical cost") exceeds the actual cost of that search (e.g., time and cognitive effort). To maximize the probability that a user stops and clicks, a CSA must convince the user that the market is "tight"—that the current best price is unlikely to be surpassed by further searching.

Algorithmic Disclosing for Rational Agents

For a fully rational agent, the decision to stop depends on the perceived distribution of prices. The authors propose two primary algorithms for selecting a subset of prices to disclose:

  1. Monte-Carlo based Disclosing: This method involves random sampling of price subsets to find the one that minimizes the buyer's expected gain from further search.
  2. Interval Disclosing: This approach selects sequences of consecutive prices (intervals). By showing a dense cluster of prices near the minimum, the CSA signals that the probability of finding a significantly lower price elsewhere is negligible.
Comparison of Critical Cost reduction between Interval and Monte-Carlo methods across different subsets.

As shown in the performance analysis, the Interval Disclosing method is particularly effective. It outperformed the Monte-Carlo approach by approximately 7.78% in reducing the incentive for further search while requiring significantly less computational overhead.

The Human Element: Quantity as a Signal

While rational agents respond to the statistical shape of the distribution, human behavior reveals a paradox. In experiments conducted via Amazon Mechanical Turk, the authors found that humans often use the volume of results as a proxy for the CSA's thoroughness.

If a CSA shows only a few "perfect" prices, a human user may suspect the search was incomplete and feel compelled to check another site. Conversely, showing a larger number of prices—even if many are mediocre—convinces the human that the CSA has "scoured the market."

Termination probability increases as more prices are disclosed, suggesting humans value volume as a signal of search thoroughness.

Key Results and the "Minimum Prices" Method

The study utilized real-world price data for electronics like the Sony WX50 camera and HP LaserJet printers. The results confirmed that for human subjects, the "Minimum Prices Method"—disclosing the $k$ lowest prices—was the most effective strategy. By showing the top 30 prices instead of just 10, CSAs significantly increased the "termination probability" (the likelihood a user stops searching), even when the absolute minimum price remained the same.

Conclusion and Limitations

Strategic price disclosure offers a powerful psychological tool for CSAs to retain users. However, the authors note significant limitations, most notably reputation risk. If users eventually discover that a CSA is hiding superior deals, long-term trust will erode. Future research should consider how these disclosure strategies might be integrated with other factors like shipping costs, seller ratings, and personalized user profiles to create more sophisticated, yet potentially more deceptive, shopping environments.

Summary updated

BibTeX
@inproceedings{hajaj2013search,
  author = {Chen Hajaj and Noam Hazon and David Sarne and Avshalom Elmalech},
  title = {{Search More, Disclose Less}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2013},
  doi = {10.1609/aaai.v27i1.8607},
}

Loading related content...

Figure