There is a real step between "find me something that works" and "if something meets these conditions, choose it for me." This study is about the second one: when consumers stopped needing to know which exact unit they were getting, what they were paid for it, and what protected them when the system was technically right and they were unhappy anyway. The harder question is whether people truly prefer that delegation, or simply fail to reverse it once it becomes the default.
The obvious way to read this question is as a trust curve — people try delegation, it goes well, they delegate more. No case in this set behaves that way. What the durable cases have in common instead is that the operator kept narrowing the range of things that could turn up until the difference between them stopped mattering much to the buyer. INFERENCE
The six mechanisms that do this recur across the successful cases: define a class where "or better" has an unambiguous meaning, narrow the set of qualifying units, state a floor the system cannot go below, give the consumer one clear moment of consent, provide a remedy for a bad result, and pay for the surrender in price, access or time. None of them is persuasion. All of them are constraints on the outcome. Where the constraints were weak, either the price of the surrender ran very high or consumers took the selection back.
The second thing the record makes obvious is that in a lot of these categories delegated selection is not the experiment at all. It is the default, and has been for decades. Airlines assign your seat for free at check-in. Rental companies guarantee a class and hand you whatever is in the row. Four-fifths of new retirement plan participants end up holding whatever fund their employer chose. REPORTED
What these markets then discovered is that the veto itself is a product. Eight US airlines generated $4.2 billion from assigned seat revenue in 2022; United alone booked $1.3 billion in seat fees in 2023. REPORTED Hertz gates exact-car selection behind loyalty tiers. Nobody in these categories is trying to talk consumers into delegation. They are charging for the exit from it, and the price they can charge is a direct read on how consequential the remaining differences still feel.
So the question is not whether people will give up choice in general. It is how narrow the outcome has to get before they stop paying to keep the veto — and that has a measurable answer in at least one category, which is where this study starts.
Lodging is the only category where the discount required to surrender the identity of the item has been measured against matched transparent bookings — same hotel, same dates, same booking window. Cossa and Tappata matched opaque bookings on two platforms to their transparent equivalents over October 2011 to June 2012.
The same study found consumers in the sample required an unconditional 44% discount to make an opaque booking at all. The 40, 44 and 49 figures are matched-transaction measurements. The 50 is an advertised ceiling and is not a comparable average; it sits here to show where the product moved, not to be differenced against the rows above it.
Two things follow. The discount for full opacity is enormous — roughly two-fifths to half of retail — which is a direct read on how expensive it is to buy the veto out of someone. And the direction of product development has been to narrow the envelope rather than widen it, which is what you would expect if the discount is the cost and the operator wants to pay less of it.
System selects, consumer approves before money moves.
System commits, consumer can undo cheaply inside a window.
Parameters set in advance, transaction completes, no second approval and no undo.
Products do not climb this list. Across the case set there is no example of one that started at Structure 1, earned trust, and graduated to Structure 3. The movement that does exist runs the other way: grocery moved from picker-decides toward customer-approves, and the styling-box category watched a conventional storefront take share from the curated box inside the same company.
Where Structure 3 does exist at scale, it is almost never because consumers chose it. In opaque travel, irrevocability is the supplier's condition for the discount — a hotel releases inventory below its published rate only if the booking cannot be cancelled and the brand is not named, because that is what stops the discount cannibalising the transparent channel. In default retirement investing, the delegation was installed by an employer under a regulatory safe harbour and the consumer simply did not opt out. Neither is a consumer preference. INFERENCE
The largest delegated-selection market in consumer travel, and the clearest proof that people will pay real money to take selection back.
Route, date, fare class. Nothing about the seat.
Row, side, window or aisle, proximity to travelling companions.
Yes, priced. Free auto-assignment at check-in, or pay to choose in advance.
$4.2B across eight US carriers in 2022. More than half of travellers buy add-ons including assigned seating. REPORTED
The specification is technically satisfied by any seat on the aircraft. Consumers disagree, in the billions, every year. Note the direction of travel too: Southwest abandoned open seating in 2024 in favour of assigned seating with premium tiers. REPORTED Where this category has moved recently, it moved toward more specificity, not less.
The oldest successful manufactured-fungibility scheme in consumer commerce. "Toyota Camry or similar" has been standard for decades and provokes almost no resistance.
Class: size, seats, luggage capacity, doors.
Make, model, trim, features, colour — stated in the booking terms. FACT
At the counter, and only upward. A class shortfall is a remedy event.
Hertz Gold Choice and Ultimate Choice, National's Emerald Aisle: pick your own car, gated behind loyalty status. FACT
Why the class works: it is defined on attributes that are ordinal and that the consumer already optimises for. More seats is more. More luggage space is more. One class up is unambiguously better. A consumer can reason about "or better" without being told what will actually turn up, and that is the entire trick.
Target-date funds became the standard default investment after the 2006 Pension Protection Act created a safe harbour. The participant specifies nothing except, often, not opting out.
In the pure default case, only an expected retirement date, inferred from age.
Every security held, the glide path, active or passive, the manager.
Total TDF assets exceeded $4.8 trillion in 2025; 87.2% of plans with a default used a TDF as that default in 2024. REPORTED
84% of participants were 100% invested in the default after one year; acceptance runs about 80% initially, decaying to about 70% after five years. REPORTED
Two things this case establishes and one it does not. It establishes that delegated selection can reach almost total penetration, and that the mechanism is durable over years. It also establishes something less comfortable: acceptance decays with tenure. The longer a person holds a system-selected outcome, the more likely they are to take selection back. What it does not establish is willingness, because nobody chose this. They failed to reverse it, which is a different behaviour with the same footprint in the data.
The most instructive design in the study. The operator stopped hiding the hotel and started naming the candidates.
Destination, dates, guests, amenity filters. Then approves a specific set of three.
Only which of the three. Every property is named, photographed, and reviewed above 7.0. FACT
None after purchase. Non-refundable, non-changeable.
The company's stated rationale: the average traveller views three hotels before booking. REPORTED
That last line is the design rule the whole study points at. The envelope was sized to the consideration set the buyer would have built anyway — not to what the system could technically fill, and not to what the seller wanted to move. The consumer is not being asked to accept an unknown. They are being asked to stop discriminating between three things they had already screened in.
Full opacity with a quality floor. Twenty-five years old, still operating, never became the way most people book a hotel.
Neighbourhood, star rating, guest review score, amenities, price.
Brand, exact address, room specifics.
Star rating and review-score floor. No refund, no change, no veto.
An entire cottage industry exists to defeat the opacity before booking by cross-referencing amenities and map pins against known inventory. REPORTED
The countermeasure is the finding. Handed a Structure 3 product with a wide envelope, a meaningful share of buyers do unpaid detective work to convert it back into Structure 1 before committing. They want the discount and the certainty, and they will spend twenty minutes to get both. That is a revealed preference about the veto that no survey would surface.
By volume the largest delegated-selection problem in consumer retail, and the one where the industry deliberately walked autonomy back.
The substitution satisfies the order and the customer hates it. 30–45% express dissatisfaction with the substitution offered. REPORTED
Ask first. After deploying pre-emptive approval prompts, Walmart's customer acceptance of substitutions rose above 95%. REPORTED
Per-item instructions set in advance — specific replacement, best match, or refund — saved and reused next time. FACT
1 and 2 in combination. Never 3.
That acceptance number is the most important single data point in the study. Same algorithm, same substitute, same customer. The only variable is whether the customer was asked. Acceptance turns out to be a property of the consent step, not of the selection. A cheap approval tap converted a category-wide grievance into a mostly solved problem.
The best available evidence on repeat behaviour after a system picked something the consumer had never seen.
A style profile of 85-plus data points, sizes, price bands, occasion notes.
Free, post-delivery, three days. Physical inspection before paying for anything kept.
70% of clients returned for a second Fix within 90 days; 83% and 86% of revenue from repeat clients in 2016 and 2017. REPORTED
Active clients fell from 4.18M in late 2021 to 2.3M by early 2026, while revenue per active client reached a record $577. REPORTED
Separate the mechanism from the business. The delegated-selection mechanic retained well and the per-client economics kept improving. What eroded was the audience, alongside the launch of Freestyle, a conventional a la carte storefront that let clients pick items directly and which management acknowledged was competing with the box. REPORTED When the same company offered delegation and choice side by side, choice took share. That is close to a natural experiment, though price and assortment were not held constant. UNRESOLVED
Enormous numbers of consumers accept an assigned provider and vehicle daily, with no meaningful selection at all.
Origin, destination, service tier. Price quoted before the match.
Driver, vehicle, exact arrival, route.
Free for roughly two minutes after matching, then a fee. Grace periods vary by product and city. FACT
Rating floors, with deactivation for providers below the city minimum. FACT
The pattern worth naming: a decaying free veto. Cheap to exercise immediately, expensive after a short window. It is the most consistent design across successful Structure 2 products, and it exists because the counterparty's cost of the reversal starts at zero and rises.
Randomised collectibles sold sealed. The consumer pays a premium rather than receiving a discount, and buys repeatedly. On the surface this refutes everything above. It does not, and the reason matters.
Series and price tier. Nothing about the figure.
None. There is no discount. Uncertainty is priced in, not out.
Roughly 70% of collectible-toy consumers buy three or more boxes trying to get the style they want. REPORTED
China restricted blind-box sales to children under 8 and requires parental consent for older children. REPORTED
The repeat purchasing is caused by the buyer not getting what they wanted. That is the opposite of delegation, where a bad fill is a failure. Here the uncertainty is the product being consumed, and disappointment is the engine. Two very different mechanisms produce identical transaction footprints: indifference between candidates, and appetite for the draw itself. Any analysis that counts them together will reach the wrong conclusion about both.
Routinely cited as evidence for autonomous purchasing. It is not evidence for this study's question, and saying why is part of the finding.
Timing. The item is fixed by the consumer in advance and never changes.
Selection. Nothing ever chooses between non-identical alternatives.
Amazon reported roughly 6,000 Dash button orders per day before removing the device fee; the physical buttons were discontinued in 2019. REPORTED
Delegated timing and delegated selection are different problems with different failure modes. Reordering the detergent you already chose cannot produce something you dislike. Excluded from the pattern analysis.
Every durable scheme here defines the envelope on attributes that are ordinal and that consumers already optimise for: car class by size and seats, hotel by star tier and review score, cabin by service level. "Or better" is usable because better is defined. Where the attributes are not ordinal — where consumers disagree about which direction is up — no class definition in this set has held. INFERENCE
Pricebreakers set the envelope at three because the operator's data said travellers look at three hotels before booking. The consumer is not asked to accept an unknown; they are asked to stop discriminating among options they already screened in. The distance between "somewhere in this neighbourhood" and "one of these three" is worth roughly ten points of discount. INFERENCE
Walmart's grocery substitution acceptance rose above 95% after a pre-emptive approval prompt, against a category baseline of 30–45% dissatisfaction. REPORTED The substitute did not get better. The asking did. Any system that can afford a cheap approval step should assume it is the highest-leverage element in the design, ahead of the matching logic.
Opaque hotels have a star tier and a review-score minimum. Rideshare has rating floors with deactivation. Rental cars have a class guarantee and an upgrade at the counter. Curated boxes have a free return. In this case set there is no example of delegated selection operating with neither a veto nor an enforceable floor. INFERENCE
No case shows a product earning its way from recommend to autonomous. The observed migrations run the other way — grocery toward approval, Stitch Fix toward a direct storefront. Structure 3 exists where a supplier requires irrevocability as the price of the discount, or where a third party installed it as a default, not where consumers grew comfortable over time. INFERENCE
Roughly 80% of retirement plan participants initially accept the default fund; about 70% still hold it after five years, and 84% are fully in the default after one year. REPORTED The drift is slow and it is one-directional. Given enough time and enough stake, people take selection back. Nothing in the set shows the reverse drift.
70% of Stitch Fix clients returned within 90 days; 83–86% of revenue came from repeat clients. REPORTED That is the strongest available answer to "did they come back after a system picked for them," and it is a yes. It is also a yes under Structure 2, with a free physical veto after delivery. No comparable repeat measurement exists for any voluntary Structure 3 consumer product in this set. UNRESOLVED
Indifference between candidates and appetite for the draw itself both produce a consumer accepting a system-chosen item and coming back. They behave oppositely. In the first, a disliked outcome is a defect to be minimised. In the second, it is the thing generating the next purchase. Blind-box commerce is the clean case of the second, and it is the only case in the set where consumers pay a premium rather than receive a discount. INFERENCE
One. Consumers pay billions annually to reverse a free delegated seat assignment. $4.2B across eight carriers in a single year is not a preference at the margin.
Two. Every mature partially-fungible market that invested in product over the last fifteen years invested in more precise choice: aisle selection for rental cars, per-item substitution instructions, richer filters and previews.
Three. Given a Structure 3 product, a real share of buyers do manual detective work to defeat the opacity and convert it back into an informed decision before committing.
Four. When one company ran delegation and choice side by side, choice took share, and management said so on the record.
Five. The one place Structure 3 achieved near-total penetration, it was installed by an employer under a regulatory safe harbour, and acceptance still decays year on year.
Six. The price of full opacity is roughly 40–49% of retail. That is what it costs to buy the veto out of a consumer, and almost no market can pay it.
What the opposing case does not survive: it cannot explain why anyone books a Pricebreaker, or accepts an assigned rental car without complaint, or takes an assigned driver at one in the morning. The reconciliation is that these are not the same ask. Consumers resist surrendering choice between options they can distinguish and rank. They do not resist surrendering choice between options they have already judged equivalent. Star tier plus review floor plus neighbourhood does that for hotels. Class definition does it for cars. The entire design problem is whether a specification can get a consumer to indifference before the money moves.
Every case that worked satisfies most of these. Every case that stalled or reversed violates at least two.
Class, tier, rating, score. "Or better" must have an unambiguous meaning. Where consumers disagree on the ranking, the class fails.
Three hotels. One car class. Not "anything in this postcode." The consumer should recognise the set as the shortlist they would have built themselves.
Review score above 7.0. Star rating. Class guarantee. Provider rating minimum. This is what replaces the veto and it is never absent.
Free return, counter upgrade, refund, credit, rebooking. Where no remedy exists, the required discount runs to two-fifths of retail or more.
The consumer reviews and approves the envelope rather than the fill. Approval at declaration costs almost nothing. Approval at fill costs the transaction.
Scarcity, a short window, or high search effort. Absent time pressure, consumers browse. This is what makes delegation attractive rather than merely tolerable.
Price, access, or time. In lodging the measured price of full opacity is 40–49%. As the envelope narrows the required payment falls, but in the voluntary cases it never reaches zero.
Consumers allow a system to choose the exact unit when they have already been brought to indifference between the candidates, when a stated floor prevents the worst outcome, when something covers a bad fill, and when they are visibly paid for the surrender in money, access or time. Delegation is not a trust milestone that gets unlocked after enough good experiences. It is a trade that gets priced, transaction by transaction, and the price is legible. The work in the durable cases went into shrinking what the consumer was giving up, not into making them comfortable giving up more.
By naming an ordinal class with a floor, and by sizing the class to the consumer's own consideration set. Rental car classes and hotel star-plus-review tiers both work this way and both are decades stable. The mechanism is not concealment. The successful direction of travel in opaque commerce has been away from concealment and toward a named, reviewed, explicit candidate set — which is the same move, made deliberately, with the envelope shrunk to what a person would have looked at anyway.
Ranked by how consistently they appear: a floor that cannot be breached, a remedy for a bad fill, a payment large enough to notice, and one cheap moment of consent at declaration time. The payment is the least interesting of the four, because it is the one that scales down as the other three get stronger. Speed and scarcity matter, but as conditions that make delegation worth considering rather than as compensation for it.
No. It relocates. In every durable case it moves to one of two places: forward, to approving the envelope before anything can be bought, or backward, to a cheap remedy after the fill. Where it was removed from both ends, the required discount ran to roughly 44% and consumers built their own tools to smuggle it back in. Structure 3 with no floor, no remedy and a wide envelope has no successful voluntary consumer example in this set.
Search and recommendation work almost everywhere. Delegated selection does not, and the failures cluster in recognisable places.
The load-bearing failure. Class systems work because everyone agrees a full-size car is bigger than a compact. Where the good is a bundle of attributes that consumers rank differently and inconsistently — taste, aesthetics, fit beyond size, personal association — no specification produces indifference, because there is no shared direction of "better" for the specification to point at. Apparel sits right on this boundary, which is why the Stitch Fix model needed a free physical return to function at all.
Every case in this set selects one unit: one room, one car, one provider, one substitute jar of sauce. Nothing here involves a requirement that binds across several units at once, where each individual selection satisfies its parameters and the combination still fails. That is the most likely source of a technically-compliant outcome the consumer rejects, and the record is silent on it. UNRESOLVED
Free return, counter upgrade, refund, rebooking — these carry the weight in every case where the veto was weak. Where the good is dated, perishable and non-substitutable, there is nothing to offer after a bad fill, and the required discount rises to the level where most operators cannot pay it.
Delegation is easier when nobody sees what you got. A hotel room and a rental car are largely private. Where the item functions as a signal to other people, the consumer is optimising against a standard the specification does not contain. This is consistent with the evidence rather than demonstrated by it. INFERENCE
Several cases assume the person knows what they want well enough to write it down. Where preferences are discovered by looking rather than known in advance, the specification is not a compression of the decision, it is a guess at it, and the system inherits the guess.
The record cannot separate willingness to delegate from failure to reverse a default.
This is the central weakness of the whole evidence base and it runs through the strongest cases. Where Structure 3 reached near-total penetration, in default retirement investing, nobody chose it — an employer installed it and the participant did not opt out. Where consumers did actively choose Structure 3, in opaque travel, adoption stayed permanently secondary and the price was two-fifths of retail. The seat assignment case sits in between: the delegated outcome is the free default and the alternative costs money, so the $4.2 billion measures the veto's value but the people who did not pay it are a mix of the indifferent and the unbothered.
Those two populations behave identically in transaction data and completely differently in every other respect. The indifferent will accept a wider envelope over time. The unbothered will not — they will drift out slowly as stakes rise, which is exactly the shape of the five-year decay in default acceptance.
What would resolve it. A market where delegated selection and individual selection are offered to the same buyer, at the same price, with the same friction, and where repeat behaviour after a system-chosen fill is measured against repeat behaviour after a self-chosen one. No case in this set is that experiment. Every real instance confounds the comparison with a discount, a default, or a friction gap, and usually all three. Stitch Fix's Freestyle storefront running alongside its curated box is the closest thing that exists, and price and assortment differed.
Until that comparison is run somewhere, the honest position is that history establishes the conditions under which delegated selection survives, and establishes that consumers pay real money to escape it, but does not establish how much of the observed acceptance is a preference at all.
Cases were selected on structural match: the consumer sets parameters, more than one non-identical item could satisfy them, and a system picks the specific one. Cases were assigned to Structure 1, 2 or 3 on the basis of where the consumer's last opportunity to reject sat relative to payment. Failures and reversals were sought at the same weight as successes, and two cases in the set are included specifically because they cut against the pattern.
Automated replenishment is recorded and then excluded, because it delegates timing rather than selection. Including it would have inflated the apparent evidence for autonomous purchasing by counting cases structurally incapable of producing the failure mode this study is about. Blind-box commerce is included but firewalled, because the mechanism producing repeat purchase there is appetite for uncertainty rather than indifference between candidates.
FACT — company documentation, filing, or peer-reviewed measurement. REPORTED — credible secondary reporting or company statements in trade press. INFERENCE — conclusion drawn across cases, labelled as such. UNRESOLVED — recorded but not established.
The 40% and 49% opaque discounts and the 44% required discount come from a single matched-transaction study covering October 2011 to June 2012 in US lodging. They are the best measurement of the price of opacity in existence and they are one market, one period, one category. The whole of section 02 rests on them and would not survive their being wrong.
The grocery substitution acceptance figure above 95% comes from trade reporting rather than a retailer disclosure and is not independently verified. It is simultaneously the strongest datum in the study and one of the more weakly sourced, which is stated here rather than smoothed over. The 30–45% dissatisfaction baseline is separately sourced from consumer research and is more robust.
Priceline does not disclose Express Deals or Pricebreakers volume, so the discount ladder shows what each envelope width is worth per booking and says nothing about how many people buy at each width. The Stitch Fix repeat figures are revenue-based, from 2016–17, and the client base has since fallen by roughly half from its 2021 peak; no per-box keep rate is disclosed. Blind-box repeat statistics are drawn from company-commissioned research and academic secondary analysis rather than audited disclosure, and are treated as directional.
It does not establish that any of the seven conditions is individually necessary; it establishes that they co-occur in the durable cases and that the reversals violate several. It does not claim delegated selection is generally acceptable or generally resisted — the evidence supports neither, and section 10 explains why the question is harder to answer than the adoption figures suggest.
PastBehavior Research. Compiled August 2026 from public sources.