Netflix, the dominant force in the video-on-demand streaming industry, has long been celebrated for
its user satisfaction and strong customer retention rates, largely attributed to its robust personalization
and recommendation technologies (Gomez-Uribe and Hunt 2015, Steck et al. 2021). However, faced
with a recent growth slowdown, Netflix is undertaking significant strategic shifts that promise to
impact both consumer finances and attention resources.

One notable change is the introduction of an ad-supported subscription tier, which aims to lower
subscription costs but also introduces the monetization of user attention and data. While hyper-
personalization in user experiences may offer time and effort savings and improve content discovery,
it comes at the cost of increased data usage, raising concerns about privacy and data protection.
Surveys indicate that 80% of online users feel they lack control over their personal data, with 60%
expressing deep concerns (European Commission 2019). Despite these concerns, Netflix continues to
invest heavily in data-driven recommender systems (Netflix 2021, Biddle 2021).

Our research addresses the critical issue of whether consumers are aware of the attention-related
strategies employed by digital entertainment companies and how they assess the associated costs and
benefits. This dilemma, known as the personalization-privacy paradox, underscores the tension
between individuals' desire for personalized experiences and their concerns about privacy (Aguirre et
al. 2016). Despite expressing concerns, individuals often continue to engage with personalized
services, creating a paradox where privacy is valued but willingly compromised for benefits such as
time and effort savings, improved user experience, and enhanced content discovery.

Our study focuses on the preferences of Polish Netflix users concerning data protection and in-
platform user experiences. It employs two sets of split-sample choice experiments through a custom
web application that simulates a Netflix-like environment. This simulated Netflix environment presents
respondents with choice scenarios involving "depersonalized" recommendations and varying levels of
personalized advertisements, allowing one group to experience potential disutility resulting from
hypothetical changes.

In Series 1, participants choose between their current subscription plans and discounted plans with
personalized or non-personalized ads of different lengths per hour of content. Additionally,
respondents are asked for consent, hypothetically tied to new EU regulations, for specific types of
personal data processing unrelated to essential service provision. More consents result in further
subscription discounts.

Series 2 delves into the personalization-privacy trade-off, focusing on users' willingness to compromise
their data for enhanced convenience in user experience. It presents varying levels of personal data
usage and the corresponding degree of recommendation personalization, describing improvements in
terms of time and effort savings during content selection. Types of personal data include user ratings,
browsing history, gender and age, geo-location, and out-platform data from external sources. Notably,
the monetary aspect is excluded.

The study, conducted via the CAWI method in May-June 2023, involved 1686 participants, about half
of whom interacted with the mock Netflix environment. Participants were required to be active Netflix
users, with at least one household member contributing financially to the subscription fee. Quotas
were applied to ensure demographic diversity.

To analyze choice responses, we utilized a standard mixed logit approach and a 3-class latent class
model to account for unobserved preference heterogeneity. We also incorporated interaction terms
to address the experiential impact of simulated disutility on preferences. Furthermore, we controlled
for participants' attitudes toward personalization in digital entertainment services and their privacy
concerns.

The findings reveal a notable gap between user responses and the perspective of Netflix executives.
Users express a limited reliance on personalization and are unwilling to share personal data beyond
ratings and browsing history to enhance in-platform user experiences. This aligns with the mixed logit
model results, showing that consent to share in-platform activity data positively influences users'
utilities. However, users remain highly sensitive to sharing other types of personal data, especially
external data sources like publicly available social media posts and publicly accessible databases
associating IP addresses with internet service providers. This resistance to data sharing holds true for
both scenarios involving privacy compromises for financial incentives and those for improved user
experiences.

As expected, increased subscription discounts in Series 1 have a positive impact on users' willingness
to share data, demonstrating the effectiveness of discounts in incentivizing data sharing. Regarding
the introduction of ads, the majority of respondents prefer to retain their current ad-free subscription
plans. Polish Netflix users are particularly averse to personalized ads and show lower marginal utility
for each additional minute of ads per hour of content.
In Series 2, substantial standard deviation terms indicate significant unobserved preference
heterogeneity among users, suggesting varied opinions on the personalization-privacy trade-off.

Latent class analysis identifies three distinct user groups: one favoring the current recommender
system (47%), another willing to embrace hyper-personalization at the expense of privacy (18%), and
a third strongly opposing increased data sharing and the recommender system's deactivation (35%).
Notably, our research highlights the impact of experiential presentation formats on preferences, with
respondents who experienced simulated disutility exhibiting greater sensitivity, particularly in terms
of their time budget. This aligns with recent evidence demonstrating the efficacy of innovative
presentation formats in enhancing respondent engagement and understanding of choice scenarios
(Bateman et al. 2009, Matthews et al. 2017).

In summary, our study represents a pioneering exploration of individuals' valuations of privacy in the
context of VOD services using choice experiments. It contributes to a better understanding of whether
consumers recognize the attention-related strategies of digital entertainment companies and how
they evaluate the associated costs and benefits. Moreover, it underscores the potential for alternative
display formats to improve valuation methods, particularly in contexts where attention budget
constraints significantly influence decision-making. Although the importance of experiential impacts
on stated preferences for cultural consumption was noticed early (Mazzanti, 2003), no valuation study
to date has leveraged recent technological advances.