Big Data and Competition Policy Market power, Personalised pricing and advertising CERRE Executive Seminar, Brussels, 16 February 2017 Prof. Marc Bourreau, CERRE and Telecom ParisTech Prof. Alexandre de Streel, CERRE and University of Namur Dr. Inge Graef, Tilburg University and KU Leuven
1. Big Data Value chain 2. Regulatory Landscape Outline of the Report 1. Big Data Value chain 2. Regulatory Landscape 3. Data and Market Power 4. Personalised Prices 5. Target Advertising
1. Big Data Value Chain 3 main parts Data collection Data storage Directly from users (often against ‘free’ services) or from machines Indirectly: data brokers and data marketplaces Data storage Large data centres But fixed costs can be made variable with cloud computing Data analytics Applications and algorithms (more and more self-learning): 4 V’s: volume, variety, velocity and value Need data, skills, capital Relationship between data-driven innovation and productivity
Big Data Value Chain Value chain evolves and is shaped Mainly by the technological progress Progress in processing and computing technologies made 4 V’s possible Impact of AI remains to be better understood And by customers preferences: understanding, behaviour and trust But also by rules on data collection and use (the ‘data regulations’)
2. EU rules on data collection and use Typologies of data Personal/non-personal, User-generated/machine-generated, Raw data/processed data Rules applicable to all data Intellectual property and trade secrets Competition rules Specific rules mandating data sharing, esp. for publicly-owned data Consumer protection Rules applicable to personal data General Data Protection Regulation Telecom specific data protection rules, which are stricter Human right of privacy Those rules can increase or lower the barriers for data collection or use
3. Data and Market Power Analytical framework 3 principles Data is one input - important but not unique - to develop applications and algorithms Big data value chain - and the broader data ecosystem - exhibit many network effects ‘Holistic’ competition analysis Each big data application/algorithm is different Case-by-case analysis 2 questions
Data and Market Power Data availability: Costs of data collection Direct collection Data is non-rival and may be everywhere, but technical, legal, contractual barriers may increase collection costs Data can be collected against direct payment or against ‘free’ services Data can be collected for their own sake or as by-product Indirect collection Antitrust assessment Merger cases: use of data for online advertising Concentrations of user data are not anti-competitive Abuse cases: data cross-subsidisation between monopolised and competitive services Customers lists are not replicable
Data and Market Power Data value for the analysis Volume of data What is the relationship between data quantity and application/algorithm quality? Variety of data What is the relationship between data variety and application/algorithm quality? Depreciation value of data Transient and less transient data For less transient data, how can they be ‘capitalised’ in application development or improvement? Impact of Artificial Intelligence on the experience curve of algorithmic development
Data and Market Power Relationship between data collection and data analysis Feedback loops may decrease costs of collection Existence of the feedback loop ? Effects of the feedback loop ?
Data and Market Power Governance framework for competition agencies Improve understanding of the big data value chain Structured dialogue and possibly sector enquiry (which should not create an unreasonable burden) Strengthen expertise in computer and data science Close cooperation with agencies in charge of consumer protection, data protection and intellectual property Understand common problems, understand the effects of the ‘data regulations’ Achieve consistency in decisions But no legal fusion as the different legal instruments are complement and not substitute
4. Personalised Prices Personalised prices: individual prices (first-degree price discrimination) or group pricing (third-degree price discrimination) with small targeted groups Presumption that more data on consumers’ characteristics or behaviour should make personalised prices more prevalent Technically possible (according to OFT report) But little empirical evidence (CNIL-DGCCRF/OFT reports) Fear of consumer backlash and reputation concerns Indirect ways to personalise prices: personalised discounts, steering
Impact on Firms and Consumers Monopoly: a priori beneficial for the firm (but not necessarily true with repeated interactions) Less clear under imperfect competition: the introduction of personalized prices by all firms can intensify competition Asymmetric access to data can strengthen competitive advantage: the firm with access to data introduces personalized prices, to the detriment of its rivals Impact on consumer surplus: Under monopoly (group pricing): ambiguous effects Appropriation vs. market expansion Under imperfect competition: ambiguous too Personalized pricing can reduce consumers’ trust in online markets, in particular if realized in a non-transparent way
Recommendations No rationale for banning personalised prices per se Ambiguous effects on consumer surplus and welfare Reputation concerns for large players Transparency to ensure consumer trust in online markets Risk that firms engaging in personalised pricing in non-transparent ways could undermine consumers’ trust in online markets Monitoring of online prices (rather than audit of price algorithms) Could be handled by consumer protection agencies upon complaints
5. Targeted advertising Advertising: a keystone for many business models in digital markets Targeted advertising: when firms place ads that target a specific audience based on their estimated personal characteristics/interests Targeting for display advertising: widespread now and quickly growing Different levels of targeting First-party targeting (from publisher) Third-party targeting (ad networks, ad exchange) Automated, real-time third party targeting (RTB…)
Impact on firms and consumers For advertisers: presumption that targeted advertising is more effective than regular advertising, but not always true (details matter) For publishers: potential trade-off between effectiveness and exposure Impact on consumers: For consumers, trade-off between receiving more relevant ads that match their interests and nuisance from intrusive ads Potential concern with steering: targeted ads as a way to price discriminate
Recommendations Online advertising: a dynamic and innovative market Firms in the online advertising ecosystem compete for the best algorithms, the best advertising technologies, to improve advertising effectiveness and limit consumer reactance Complexity of the ecosystem may call for some monitoring But anything beyond monitoring is not warranted as long as no evidence of abuse of market power This ‘light’ approach works as long as no dominant player can block innovation by rivals (e.g., entrants)