A Little Randomness in Prices Could Save You Money

A pricing page can quietly walk its price up until it hits the most you'd still pay. A recommendation feed can hide the good stuff behind a sponsored item you'll barely tolerate. A hiring model can shift its cutoff once applicants change their resumes. In each case, the software learns how you react, then picks the option that squeezes the most out of you. Five researchers just published a paper on arXiv explaining why — and what breaks the pattern.

Their answer is a bit surprising: the platform should deliberately be a little unpredictable. Instead of always offering its single best option, it should randomly mix among several near-best options that a normal person would still happily accept. The randomness acts like a promise the platform can't take back. Anyone studying its behavior can still learn the general pattern, but can't reliably predict the next move — so they can't target the exact moment you'd give in. In pricing tests, buyers kept more money, and sellers barely lost anything. The paper calls these "unpredictable near-optimal policies."

There are real limits. The trick stops working when options are ranked by quality rather than price, when only one good option exists, when the platform's guess about your preferences is wrong, or when it only cares about the next few minutes rather than the long run. It also backfires if the other side is genuinely trying to help you — say, a hospital scheduling your surgery — because there's nothing to defend against.

So what does this mean day to day? It's a theory paper, not a product launch, so nothing changes tomorrow. But it explains a nagging feeling many of us have: quote prices wobble between sessions, and it's not always a glitch. It also gives regulators and app designers a concrete recipe — small, honest randomness among good options — that protects consumers without costing companies much. Expect this idea to show up in debates about how airlines, insurers, and shopping apps set your price.