Airlines and hotels change prices with demand as a matter of routine. Bars and restaurants mostly don't, because there a price that moves with demand is read as opportunism rather than inventory control. I argue that the fairness reaction depends mainly on attribution and transparency: what the customer believes caused the price to move, and whether the way it moves was predictable. The size of the move matters less. So the same price path can feel like an insult in one venue and like a game in another, depending on how it is governed. The article ends with the conditions under which a venue-level price exchange is a legitimate mechanic, and the conditions under which it turns into manipulation.
Revenue management and the hospitality exception
Talluri and van Ryzin (2004) describe revenue management as allocating perishable capacity across heterogeneous demand. The goal is a better match between what is available and who is willing to pay for it, which does not have to mean a higher price. The classical preconditions are fixed, perishable inventory, demand that varies a lot over time, willingness to pay that differs between customers, a marginal cost that is small relative to price, and a fence that keeps the segments apart in a way that can be defended. The fence turns out to be the one that matters here.
A bar meets four of those five conditions unusually well. A seat at 22:00 on a Friday is as perishable as an airline seat, demand swings across the week far more than in most retail formats, the marginal cost of a poured drink is a small fraction of its price, and willingness to pay visibly differs within a single room.
What hospitality lacks is the fence. Airline revenue management works partly because its segmentation devices (advance purchase, refundability, stay restrictions) are easy to understand, and partly because each price is quoted privately, with no public anchor to compare it to. In a venue the purchase is immediate, the relationship is repeated, and the price sits on a printed menu that everyone reads at the same time. Any price difference is between people standing next to each other, and nobody can say why it exists. The sector's resistance is often put down to conservatism. I think it has two concrete causes: there is no fence, and there is a public reference price.
The fairness constraint and dual entitlement
Kahneman, Knetsch and Thaler (1986) supply the mechanism. In their dual-entitlement account, a transaction is judged against a reference transaction, and both parties hold entitlements relative to it: the firm to its reference profit, the customer to the reference terms. A price increase that protects the firm's reference profit against a cost increase is judged acceptable. An increase that captures surplus from a demand shock is judged unfair, even when the arithmetic and the resulting price are identical.
This is the part of the finding worth building on. People judge fairness by the cause they attribute to the increase, much more than by the number. A €2 increase explained by a supplier price rise passes where the same €2 explained by a full room does not.
The printed menu as a reference price
Thaler (1985) splits the value of a purchase into acquisition utility (the good's value relative to obtaining it elsewhere) and transaction utility, the gap between the price paid and a reference price the buyer holds. A printed menu is an unusually strong reference, because it is public and explicit and customers read it again at every order. A price above it produces negative transaction utility even where acquisition utility is unchanged. That is why the same drink at the same price feels different depending on whether the anchor was ever stated.
Loss aversion and asymmetric price movement
Prospect theory (Kahneman & Tversky, 1979) adds that people evaluate outcomes as gains and losses against a reference point, and that losses weigh more than equivalent gains. So symmetric volatility around a mean is not experienced symmetrically. Upward moves count for more than downward moves of the same size. A mechanic whose price only ever rises is simply a surcharge shown on a chart. Even a symmetric one needs price drops that happen often and visibly enough for customers to act on them, and a line in the terms of service saying that prices may fall does not meet that bar.
Rule-governed pricing as a reframing mechanism
A customer who sees a price above the anchor can explain it in one of two ways. Either someone decided to charge them more, or a rule they already knew about produced this number from conditions they can observe. The first is a dual-entitlement violation. The second is treated more like the weather, as something nobody in particular chose.
Experimental economics showed that market outcomes depend on the institution, meaning the specific rules of exchange, and not on preferences alone (Smith, 1962). Vickrey (1961) is the standard demonstration that a published rule set with known incentive properties is itself the thing being designed. I cite both for the concept only: rules of exchange can be designed and analysed, and participants reason about them.
Hunicke, LeBlanc and Zubek (2004) supply the vocabulary for the rest. In the MDA framing, mechanics generate dynamics, which generate aesthetics. A price rule is a mechanic. The behaviour it produces, such as timing a round, watching for a dip or coordinating a table, is the dynamic. The aesthetic is what customers actually report on, and here it is either challenge and discovery or coercion. The same chain decides whether a group mechanic invites participation or excludes people, which is the design problem I work through in participation architecture for party games. Better presentation alone won't reframe anything. The reframing holds only if the mechanic is playable, and that comes with conditions you can test. The rule has to be published before the session, not worked out afterwards. Its output must follow entirely from inputs the customer can observe. Everyone in the room gets the same rule, with no per-customer personalisation. And the price has to be able to move in both directions, within bounds, so the worst case is known in advance.
A price rule for a venue-level drink exchange
Our own work here is BrewX, a venue operating system in development whose core mechanic is a live drink exchange. Prices for each item move with actual order flow and are shown as OHLC candles, with flash crashes and bull runs as named events. The venue sets the algorithm's parameters. Ordering runs through QR and settlement through a digital wallet, and four AI agents cover promotion and hype, churn prediction, fraud detection, and atmosphere scoring. The system replaces roughly ten separate venue tools, which is the commercial case for it. The pricing mechanic is where the ethical questions sit.
A minimal price rule of this family can be written compactly:
p_t = clamp( p_ref * (1 + k * (d_t - d_bar) / d_bar), p_min, p_max )
subject to |p_t - p_{t-1}| <= delta
Here d_t is observed demand for the item in the current window, d_bar a trailing baseline, k the sensitivity, delta a per-tick move cap, and p_min and p_max hard bounds expressed as fractions of p_ref, the published menu reference. Three properties do the fairness work. The per-tick cap delta bounds volatility, so no single move is large enough to feel aimed at the person ordering. The floor and cap bound the range, which turns the worst case into a published number. And mean reversion to p_ref keeps the expected session price at the anchor instead of letting it drift upward.
Two implementation details matter a great deal. First, the fairness instrument is the floor. A cap alone still leaves the price distribution biased upward relative to the anchor, which is the surcharge case again. Second, the price has to settle at order time and be shown before the customer confirms. If it moves between the tap and the charge, the construction fails, because the customer's action no longer determines what they pay.
Safety limits of gamified pricing in alcohol service
Alcohol changes the analysis and adds three constraints that a venue cannot treat as optional.
The first is consumption rate. A price crash is a discount with a countdown, and that structure speeds up drinking. Having a published rule behind the crash doesn't change this. So discounts should be limited by quantity, for example the first N units at the crash price, and not by a time window. Non-alcoholic items must be eligible for the same mechanics, and per-customer purchase caps must exist regardless of price.
Targeting is the second. Once a discount is aimed at the individual who responds most to discounts, dual entitlement is violated in its most direct form. The mechanic then discriminates against the customer with the least price resistance, and in a bar that is often the most vulnerable person in the room. Agent outputs therefore have to stay at aggregate level. Atmosphere scoring describes the room as a whole. Churn prediction can inform what the venue does, but never what one customer is charged. Enforcing that scope is also an orchestration constraint, since it decides which agent may write to which surface. I cover that separately in orchestration patterns for multi-agent workflows.
The third constraint is about whether the display is honest. If the chart shows fabricated volatility, staged crashes or invented scarcity in place of the real order-driven series, the legitimacy argument falls apart. The published rule was the only thing separating the mechanic from a hidden surge model. The gambling boundary sits close by. Price movement must depend on observable demand and not on chance, and no stake may be at risk. If the price moved randomly, each order would become a wager, and gambling licensing would then be the first question, ahead of the ethical one. The wallet and QR flow also produce per-person alcohol purchase histories. Purpose limitation and data minimisation under Regulation (EU) 2016/679 and, in Turkey, KVKK Law No. 6698 constrain those more tightly than ordinary retail telemetry. The architectural answer is the one I argue for in privacy by design for on-device AI: keep the sensitive record where it is generated, and send the model only what the feature needs.
Operator controls and the authorship question
Someone has to set k, delta and the bounds. We think the venue should. Local norms, licensing conditions and liability all sit with the venue, and a curve imposed by the platform would be both wrong and unaccountable. But once the venue sets the parameters, the original problem comes back: a venue can configure a pure upward surcharge and keep the chart as decoration.
The way out is for the platform to limit which parameter sets are allowed and leave the choice within those limits to the venue. A platform can require that the floor sit strictly below the anchor, that delta be finite, that upward and downward sensitivity be equal, and that the active parameters be visible to customers instead of buried. Stock exchanges work the same way. Participants trade freely, and the exchange sets and publishes circuit breakers and limit rules.
Measurement and regulatory exposure
The primary metric is margin per session rather than margin per drink, because a working mechanic can raise volume while lowering the average price, and the per-unit view reads that as failure. Fairness perception should be measured directly with a short post-checkout question, separately from satisfaction with the venue. Repeat rate at seven and twenty-eight days is the closest available proxy for damage to the relationship. A consumption-safety metric belongs on the same dashboard and should be able to veto a parameter set: drinks per person per hour, read as a distribution with attention to the upper tail, never as a mean.
Experiment design has one constraint that is easy to miss. Randomising price rules between customers in the same room destroys the uniformity that makes the rule fair, so the unit of randomisation has to be the venue-night. Regulatory exposure depends on the jurisdiction and has to be checked before any defaults are chosen. The relevant rules include posted-price obligations, consumer protection rules on advertised prices, and alcohol promotion restrictions. Several jurisdictions prohibit time-limited drink discounts outright, which would rule out the crash mechanic however it is presented.
Limitations
BrewX is still in development, so nothing above has been measured. We have no data yet on how a rule-governed exchange affects margin, perceived fairness or consumption. The dual-entitlement results come from surveys and experiments mostly outside hospitality, and none of them tested an exchange format. If people accept cost-driven increases, it does not follow that they will accept rule-driven ones. That transfer is a hypothesis. The formula above is a set of constraints, not a demand model, and theory cannot tell you what k, delta and the bounds should be. We think the safety constraints are necessary, but we have not shown that they are sufficient. The weakest point is the premise itself. The whole design assumes a customer who reasons about a published rule, and alcohol weakens that ability in the very setting where the mechanic runs.
References
- Hunicke, R., LeBlanc, M., & Zubek, R. (2004). MDA: A formal approach to game design and game research.
- Kahneman, D., Knetsch, J. L., & Thaler, R. (1986). Fairness as a constraint on profit seeking: Entitlements in the market.
- Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk.
- Smith, V. L. (1962). An experimental study of competitive market behavior.
- Talluri, K. T., & van Ryzin, G. J. (2004). The theory and practice of revenue management.
- Thaler, R. (1985). Mental accounting and consumer choice.
- Vickrey, W. (1961). Counterspeculation, auctions, and competitive sealed tenders.
- European Parliament and Council of the European Union (2016). Regulation (EU) 2016/679 (General Data Protection Regulation).
- Republic of Turkey (2016). Law No. 6698 on the Protection of Personal Data.