Pricing Algorithms and Competition Law: Why the Enforcement Lull Is a Trap
When rivals feed prices into the same third-party model and act on its output, the result can look like a cartel while the law still files it under data services. A quieter enforcement phase narrows that risk on paper, not in fact.
Here is the uncomfortable version. If your competitors have stopped meeting in hotel bars and started feeding their prices into the same third-party model, then adopting whatever number it returns, the market outcome can be hard to tell apart from a cartel. Nobody agreed anything out loud. The coordination lives inside a vendor's software, and both the law and the sales deck file it under data services. That gap is where pricing algorithms and competition law now collide, and it is wider than most UK boards assume.
The mechanism is not exotic. A shared optimisation tool ingests non-public information from competing firms, models the market, and hands each customer a recommendation that tends to point everyone the same way: up. Each firm acts alone. The aggregate can behave like a coordinated one. Economists call the result algorithmic collusion, and the label is doing real work.
Does UK competition law already reach shared pricing software?
Yes, further than the quiet news cycle suggests. The relevant prohibition is not new. Chapter I of the Competition Act 1998 bans agreements and concerted practices that prevent, restrict or distort competition, and it does not require a signed deal. A concerted practice can arise from coordination that substitutes practical cooperation for the risks of competing. Exchanging commercially sensitive, non-public information through a common intermediary is a recognised route to exactly that. The Competition and Markets Authority set this out in its research on how algorithms can reduce competition and harm consumers, which flags shared pricing tools and hub-and-spoke information flows as live risks rather than hypotheticals.
The UK also has form. In 2016 the CMA found that two sellers of posters and frames had used automated repricing software on Amazon's marketplace to avoid undercutting each other, an infringement of Chapter I. That matter, the online sales of posters and frames case, ended in a fine and a criminal conviction for one director. Software did the enforcing; the law still called it a cartel. The technology was cruder than today's optimisation engines. The principle transfers.
What the US cases do and don't establish
Across the Atlantic, US enforcers have pushed the same theory into court. In September 2023 the Department of Justice sued the agricultural data firm Agri Stats, alleging it ran an information-exchange scheme across meat processors. In August 2024 the DOJ, joined by a group of state attorneys general, sued the rental-pricing firm RealPage, alleging its software helped landlords coordinate rents. In January 2025 the government amended that complaint to add six large landlords as defendants and, on the same day, announced a proposed settlement with one of them, Cortland Management, which agreed to stop using competitors' non-public data to train its pricing models. That is the shape to watch: individual defendants peel off through settlement while the case against the software model itself grinds on. No US court has yet ruled on the merits that a shared-algorithm pricing model is unlawful in itself, and a consent decree with one landlord sets no such precedent for the next.
Why a quieter enforcement phase is the risk, not the relief
Here is my thesis, with the evidence that would flip it. A rational firm reads a settlement like Cortland's as the storm passing. I read it as the barometer falling. A consent decree resolves one defendant's exposure while leaving the underlying model on sale to the next hundred buyers, and it does so without the precedent a trial would set. A behavioural remedy or fine smaller than the gain is priced as a cost of doing business, not a deterrent. So the perceived risk of adopting pricing software can fall at the very moment its latent legal exposure is climbing. What would change my mind is a settlement that dismantled the data-sharing architecture across the market, or a merits ruling that named the shared-algorithm model unlawful. Absent those, the base rate favours continuity.
What should UK procurement and compliance actually do?
Start by resisting the universal claim. Not every pricing tool is a problem. A model trained only on a firm's own data, public list prices, or genuinely aggregated and anonymised market statistics sits in very different territory from one that pools rivals' live, non-public quotes and feeds back a number everyone follows. The exposure concentrates where two things are both true: sensitive competitor data goes in, and the output is adopted with little independent judgement. That is the pattern to audit for, on both sides of the trade.
If you sell in a concentrated input market, the question for compliance is not whether the tool is legal to buy, but how your inputs and pricing would look reconstructed from vendor logs in a future disclosure request. If you buy, assume nothing is arriving to rescue you on a useful timescale, and build the checks in before signing. This is the sort of thing a serious technical strategy review should surface before a vendor contract, not after a claim lands. The wider lesson generalises past pricing: any system where a third party's model quietly sets outcomes for competing parties needs the same scrutiny we bring to keeping humans in genuine control of automated decisions. A recommendation that is always accepted starts to look like a delegation, and delegations can carry liability.
One correction to the comfortable story: the vendor is not safe by default. The firm holding the algorithm can itself be the facilitator a regulator names first, which is exactly the position a data intermediary now finds itself in on the other side of the Atlantic. Buyers can inherit coordinated pricing they never chose; sellers can build a latent claim they cannot yet see on the books; and the vendor sits between the two, holding the evidence. The audit that feels like paranoia now is the one that looks like foresight when the cycle turns.
Questions people ask
Does UK competition law need proof that competitors agreed to fix prices?
No. The Chapter I prohibition in the Competition Act 1998 covers concerted practices, not just formal agreements. Coordination that replaces the normal risks of competing, including the exchange of non-public commercial information through a shared tool, can fall within it without anyone signing a deal. The US Agri Stats and RealPage cases are testing a similar information-exchange theory, but both remain allegations rather than settled law.
If a pricing tool is legal to buy, is the buyer protected from liability?
Not necessarily. The legality of a product is separate from how it is used. If your data inputs and pricing behaviour form part of a coordinated pattern with competitors, the tool being on general sale is unlikely to be a defence. Both the buyer's conduct and the vendor's role in facilitating any information exchange can attract scrutiny.
How should a UK firm assess algorithmic pricing risk under competition law?
Start with the information flows. Identify what non-public competitor data, if any, enters the model, whether recommendations are followed mechanically, and whether the aggregate effect softens competition. Record the human judgement applied to each decision, because an unexamined auto-accept is the hardest position to defend.
Related
- Your Cloud Provider Is Financing Its Own AI Revenue. Do the Financial Due Diligence
- The Announcement Is Not the Deal: AI Infrastructure's Announced vs Contracted Gap
- The Target-Less Number: Glucose Sensors Are Turning Fitness Trackers into Health Anxiety Machines
- Data & Strategy
Written by an AI editorial persona of Abyshire's proprietary editorial system and reviewed by our team.