When Every Seller Uses the Same Pricing Software, Nobody Has to Agree to Anything
Algorithms that set prices automatically raise a question the law was not written for. Competitors using a shared pricing tool can end up moving together without any of them ever communicating.
What Algorithmic Pricing Does
Rather than setting a price and revisiting it periodically, a firm can let software adjust continuously based on demand, inventory, competitor prices, time remaining, and customer segment.
The commercial logic is sound. Airlines and hotels have done versions of this for decades under the name revenue management, because they sell a perishable fixed capacity where an unsold seat or room is worth nothing after departure. Getting the price right day by day is worth a great deal.
Why It Spread
What changed is that the same approach became available to businesses without airline scale. Online retail made price changes free to implement, competitor prices observable, and demand response measurable almost immediately.
The result is that prices in many categories now move constantly, and the pricing decision has shifted from a periodic management judgment to a continuously running system that few people inside the company fully understand.
Pricing moved from something a team decided to something a system does, and oversight did not automatically follow it.
The Competition Problem
Competition law prohibits agreements between competitors to fix prices. The central concept is agreement, whether explicit or inferred from conduct.
Algorithms create situations that produce similar outcomes without one. Consider several competing firms that each independently license the same pricing software, and each supplies its own confidential transaction data to the vendor. The vendor recommends prices informed by all of it. No firm communicates with any other. Prices nonetheless move in a coordinated way, informed by pooled data none of them could lawfully have shared directly.
| Scenario | Traditional legal view |
|---|---|
| Executives agree on price | Clearly unlawful |
| Firms independently watch public prices | Generally lawful |
| Shared tool fed by confidential data | Contested and actively litigated |
| Independent systems that learn to coordinate | Largely unresolved |
The Harder Version
The genuinely difficult case involves no shared vendor at all. Systems that adjust prices in response to observed competitor behaviour can, in principle, settle into a pattern where each refrains from cutting prices because it learns that cuts are matched and everyone ends up worse off.
That is the economic result competition law aims to prevent, arrived at without communication, without agreement, and arguably without intent. Whether existing law reaches it is unsettled, because the legal machinery was built around proving an agreement that in this case does not exist.
Personalised Pricing
A separate concern is charging different customers different prices for the same item based on inferred willingness to pay, using signals such as location, device, or browsing history.
Economically this is price discrimination, which is common and often legal. Airlines and universities do versions of it openly. The objections are about transparency and about which signals are used, since inferences correlating with protected characteristics can produce discriminatory outcomes regardless of intent, and the customer cannot see that a different price exists.
Where Responsibility Sits
The direction of regulatory thinking is that firms remain responsible for what their systems do. Delegating pricing to software does not delegate the legal obligation, and not understanding the tool is not a defence.
Practically that means firms need to know what data their pricing vendor pools, whether competitor confidential information is part of it, and whether their system behaviour would be lawful if a person had done the same thing deliberately. That last test is the most useful one available.
The Bottom Line
Algorithmic pricing is a real efficiency, matching prices to demand far better than periodic human decisions can. It also creates coordination that existing law struggles to characterise, because the law asks whether firms agreed and the software makes agreement unnecessary. The unresolved question is not whether the outcome is harmful but whether the current legal test can reach it.