Your growth team plugs a repricing tool into the marketplace. It watches three competitors, matches their moves within minutes, and margins stabilise across the whole category. Nobody agreed anything with anyone. Six months later, a competition authority asks why four “rivals” have moved prices in lockstep for two quarters. Welcome to the algorithmic pricing problem; the moment where optimisation becomes coordination.
The legal line: parallel conduct vs. concerted practice
Competition law does not ban watching competitors or even matching them; conscious parallelism is lawful. It bans agreement and concerted practice. Algorithms blur the line in three recognised patterns: the messenger (humans collude, software executes; plainly illegal, as in the classic poster-seller cases); the hub-and-spoke (rivals independently adopt the same vendor’s pricing engine, which aligns them through a common brain); and autonomous tacit collusion (learning agents discover that punishment strategies sustain high prices without any human ever intending it). The first two fit existing doctrine; the third is where enforcement is heading.
Why this is now a Türkiye issue, not a seminar topic
The Turkish Competition Authority has put algorithmic pricing squarely on its 2026 agenda: it is building AI-based detection using deep-learning techniques to spot coordination patterns proactively, and its leadership has said learning algorithms that coordinate prices without human intervention are now a mandatory supervision area. Practically, that means the defence “the algorithm did it” will be tested in Ankara, not just in Brussels, and Turkish marketplaces, delivery platforms and e-commerce sellers using third-party repricers are the natural first sample.
The five-question self-test
Ask these before deploying any pricing automation. One: does the tool use competitor-specific non-public data from a shared vendor? Two: do several of your competitors use the same engine with the same objective function? Three: can the tool implement punishment logic (undercut responses, tit-for-tat)? Four: is there a human review point for price changes above a threshold? Five: could you explain, with logs, why a given price was set? “No” on four or five (or “yes” on one to three) is your signal to involve counsel before launch, not after the information request.
The three patterns, mapped to liability
| Pattern | How it happens | Legal status | Early warning sign |
|---|---|---|---|
| Messenger | Humans agree; software implements the agreed prices | Cartel; clearly illegal | Any competitor communication about pricing “logic” or tools |
| Hub-and-spoke | Rivals independently adopt the same vendor’s engine or data pool | Concerted practice risk, growing case law | Vendor markets “market-level optimisation”; rivals’ prices converge after onboarding |
| Autonomous tacit collusion | Learning agents discover reward in mutual restraint | Frontier, but deployer bears the conduct | Your own margin “stabilises” category-wide without a demand-side explanation |
The clause set for your repricing vendor
Four provisions turn a risky tool into a defensible one. A data-segregation warranty: your strategies and non-public data are not used to train or inform outputs served to competitors. An explainability commitment: on request, the vendor reconstructs why a given price was set, with logs; this is also your evidence file if the authority ever asks. A compliance-by-design representation that the engine does not implement retaliation or signalling strategies. And an audit-and-exit right if the vendor onboards your direct competitors onto a shared model. If a vendor resists all four, that resistance is itself due-diligence information.
Is using the same repricing SaaS as competitors illegal by itself?
Not by itself, but shared-vendor alignment is the fact pattern authorities probe first. Diligence the vendor: whose data trains the model, and are customers’ strategies segregated?
Who is liable when a self-learning agent colludes?
The undertaking deploying it. Competition law attributes the algorithm’s conduct to the company; “we didn’t know” is a mitigation argument at best. Design-stage guardrails are the only real defence.
This week’s homework
Inventory every automated pricing touchpoint, ask each vendor the two diligence questions above in writing, and set a human-approval threshold for price moves. If you run a marketplace, do the same exercise one level up: your sellers’ common tools are your exposure too.
Related reading: Turkish merger control thresholds 2026 · AI Compliance Hub.
Author
If this is on your desk
Templates and checklists are free in the Founder Academy; for a specific situation, book a 30-minute intro call.
Founder AcademyBook an intro callMore from Vircon Insights
AI Washing: When "AI-Powered" Becomes a Legal Claim in Advertising, Competition and the Fundraising Deck
August 3, 2026Will Insurance Pay When the Model Is Wrong? E&O, Cyber and Product Liability for AI Companies
August 28, 2026Whose Voice Is It Anyway? Voice Cloning Under Turkish Personality Rights, KVKK Biometrics and FSEK
August 26, 2026Synthetic Data Is Not a Legal Loophole: When "Fake" Data Is Still Personal Data
August 20, 2026The Turkish Signature Circular in 2026: Abolished, Still Requested, and What Replaced It
August 4, 2026When the AI Acts for You: Agents, Authority and Who Bears the Mistake
July 31, 2026Related Practice Areas
Competition Law
Merger control filings, abuse of dominance, cartels, authority investigations.
View service →Corporate Law
Share transfers, capital increases, board structuring, governance.
View service →Mergers & Acquisitions
End-to-end M&A: due diligence, structuring, documentation, negotiation.
View service →