SELF-LEARNING PRICE OPTIMISATION SYSTEMS

1. Introduction

Self-learning price optimisation systems are algorithmic systems that use artificial intelligence, machine learning, historical market information, real-time demand, supply conditions, consumer behaviour, competitor prices, and operational constraints to determine or recommend prices. Unlike traditional pricing software based on fixed instructions, a self-learning system continuously updates its strategy according to observed market outcomes.

In electricity markets, such technology may be used for wholesale bidding, dynamic retail tariffs, battery dispatch, demand-response pricing, electric-vehicle charging, flexibility services, and optimisation of generation portfolios. These systems can improve efficiency, but they also create significant questions concerning competition law, market manipulation, transparency, discrimination, consumer protection, and regulatory accountability.

2. Operation in Electricity Markets

A self-learning system may analyse electricity demand, weather, renewable generation, congestion, fuel costs, competitor bids, imbalance prices, and historical market behaviour. It then learns which pricing or bidding strategy maximises a chosen objective.

Learning algorithms differ from simple adaptive systems because they may discover profitable strategies without every response being expressly programmed in advance. Research concerning electricity markets has specifically examined learning-based bidding and pricing strategies and their ability to adapt dynamically to market conditions.

This creates regulatory concern where several autonomous systems independently learn that maintaining higher prices is more profitable than aggressive competition.

3. Competition-Law Risks

The principal legal risk is algorithmic collusion. Traditional cartel law ordinarily requires some form of agreement or concerted practice between competitors. Self-learning algorithms, however, may produce parallel or coordinated pricing without conventional human communication.

South Africa's Competition Act 89 of 1998, particularly provisions dealing with restrictive horizontal practices, abuse of dominance, and price discrimination, provides the principal competition-law framework. South African scholarship considers existing competition rules potentially capable of addressing several forms of algorithm-assisted collusion, although uncertainty remains where autonomous algorithms generate coordination without identifiable human agreement.

Electricity regulators must therefore distinguish legitimate optimisation from conduct that suppresses competitive bidding or manipulates wholesale-market outcomes.

4. Transparency and Accountability

Self-learning pricing also creates an explainability problem. An electricity supplier or generator cannot necessarily avoid legal responsibility merely by arguing that an algorithm independently selected the challenged price.

Regulatory safeguards may include algorithmic audits, record-keeping, human oversight, market-surveillance systems, disclosure of relevant pricing parameters, testing for discriminatory outcomes, and requirements allowing regulators to reconstruct significant automated decisions.

These safeguards are particularly important where automated prices affect essential electricity services or vulnerable consumers.

5. Case Law

Samir Agrawal v Competition Commission of India (Supreme Court of India, 2020)

Facts: The applicant alleged that Ola and Uber used algorithmic pricing systems that effectively fixed fares for participating drivers, preventing drivers from independently negotiating prices.

Legal Issue: Whether algorithmically determined fares constituted unlawful price fixing or a hub-and-spoke cartel among drivers.

Judgment: The competition authorities found insufficient evidence of an agreement or collusion between drivers, and the Supreme Court upheld the dismissal of the competition complaint.

Legal Principle/Ratio: Similar prices generated through a common algorithm do not automatically establish unlawful cartel conduct; competition law ordinarily requires evidence satisfying the legal requirements for agreement or concerted action.

Significance: The case demonstrates the difficulty of applying conventional cartel concepts to automated pricing and is highly relevant to electricity bidding algorithms.

CMA Online Sales of Posters and Frames Investigation

Facts: UK competition authorities investigated competing online sellers that had agreed not to undercut each other's prices and used automated repricing software to implement that arrangement.

Legal Issue: Whether the use of pricing software altered the illegality of an underlying price-fixing agreement.

Judgment: The conduct was treated as unlawful horizontal price coordination.

Legal Principle/Ratio: Using an algorithm to implement a cartel does not shield businesses from competition-law responsibility; software may become the mechanism through which an unlawful agreement operates.

Significance: Electricity firms deploying automated bidding technology remain responsible where algorithms implement or reinforce prohibited coordination.

6. Electricity-Sector Regulatory Implications

Self-learning pricing should therefore be governed through competition surveillance, licence obligations, market-manipulation rules, algorithm testing, consumer safeguards, cybersecurity controls, and regulatory audit access. South Africa's electricity pricing framework is itself undergoing modernisation, illustrating the importance of ensuring that emerging pricing technologies remain consistent with transparent and regulated tariff principles.

7. Conclusion

Self-learning price optimisation can make electricity markets more responsive and efficient, but autonomous pricing also creates risks of collusion, discriminatory pricing, opacity, market manipulation, and weakened accountability. The central legal principle is that automation does not remove responsibility. Regulators must ensure that learning systems operate within competition law, electricity-market rules, consumer-protection requirements, and transparent governance structures.