Skip to content

Personalisation: Price Discrimination

tags: case study

1. Summary

2. Case Study Details

Case Study Description

An online travel booking platform decides to implement a flight aggregator that shows customers different prices for the same flight depending on information inferred from their browsing habits and digital fingerprints (e.g. location from IP address, household income based on data extracted from broswer cookies).

The platform groups users into different buckets, which it uses to estimate their 'willingness to pay'—an inferred category. Customers that have visited sites associated with luxury brands, or who are accessing the platform from more affluent postcodes, are charged a premium compared to other users.

Customers are, typically, not aware or informed that this is happening, but the platform does have a permissive privacy policy in place that can be accessed by users. Furthermore, because the prices of the goods change frequently, consumers do not have a straightforward way of comparing prices among different customers.1 The online retailer is engaging in price discrimination, in particular what is known as targeted pricing.

Category Definition: What is Personalisation?

In a general sense, personalisation is the targeting or adapting of some behaviour or outcome to a specific person. In the context of AI, personalisation occurs when an algorithm learns about an individual's profile in order to tailor the services or products offered to that person.

A range of AI techniques can be employed in personalisation, including learning algorithms that create and refine a model of a specific user over time based on the user's behaviour and interaction with the algorithmic system.

Recommendation systems are a well-known application of personalisation. For instance, an algorithm can provide recommendations on what content to watch or listen to based on a consumer's previous browsing or purchase patterns.

Background

Key terms include:

Key Information

  • SIC Classification
    • Section: G
    • Class: All codes which involve retail sales. (list?)
  • Stakeholders and Other Affected Individuals:
    • Online shoppers
    • Retailers
    • System Developer
  • Data Types:
    • Digital Fingerptints (e.g. IP Address)
    • Browser Cookies
  • Possible Algorithmic Techniques Employed:
    • Unsupervised Learning (e.g. clustering) to segment consumers during market research
    • Multi-class classification to predict which group a specific user is most closely aligned to
    • Reinforcement learning to identify and optimise personalised 'willingness to pay'

3. Regulatory Considerations

Specific Issues

Existing and Proposed UK Regulation or Legislation

Gaps in Current Framework

Socioeconomic Value Chains

Harms and Benefits

  1. Travel sites have been known to do a version of this. They charge travelers from some countries a premium when they book flights or hotel rooms from specific countries (Rose & Rahman, 2015). Additionally, the prices for flights and hotel rooms fluctuate with relative high frequency, which makes it hard for individual connsumers to keep track of them and compare. ↩