PastBehavior · Fieldwork No. 037

When the buyer goes first

Marketplaces that let buyers declare what they want, at what price, and then waited for the other side. What happened to liquidity, who actually resisted, and the conditions under which people committed money before the transaction existed.

QuestionCan persistent buyer intent beat a listing marketplace
Cases examined14 across 9 industries
Period1995 – 2026
Evidence qualityMEDIUM–HIGH
OutcomeYES, conditionally
CompiledAugust 2026
01 / The finding, first

Buyer-declared intent wins where the unit is fungible. Everywhere else it becomes a lead.

In every market where "declare what you want and wait" became the dominant way transactions happen, the thing being bought had been reduced to a specification a machine could fill without judgment. A share of Apple. A size 10 Jordan 1 in the same colorway. A dry van from Memphis to Atlanta on Tuesday. An impression served to a user matching these attributes.

In markets where the unit could not be reduced that way — a mortgage, a job candidate, a hotel room of unknown brand, a specific seat at a specific show — buyer-declared intent survived, but as a request that generates seller interest, not as a mechanism that clears trades. LendingTree has processed over 35 million loan requests REPORTED and is a lead generation business. Lenders specify the borrower type they want and pay for the introduction. Nobody bids.

The second finding matters more for anyone building this. Buyer participation was almost never the thing that broke. In the clearest failure in the record, Priceline WebHouse Club had roughly two million grocery and gasoline customers and was adding about 100,000 a week when it shut down. FACT Demand was not the problem. The supply side would not fund the discount, so WebHouse paid the gap out of its own pocket until the money ran out, in under a year. That shape repeats: the buyers show up, and the question is whether sellers would rather answer a standing bid than post a price.

02 / The distinction that does the work

Three levels of delegation, and why they are not a ladder

These get collapsed constantly. Held apart, they explain almost every outcome in the case set.

Level 1
Monitor for me

Software watches. The buyer decides everything, including when to look.

  • Google Flights tracking
  • eBay saved searches
  • Price history charts
Level 2
Tell me when

The buyer states a condition. The system pushes a notification. The buyer still transacts by hand.

  • Fare drop alerts
  • Restock and back-in-stock
  • eBay Best Offer (buyer side)
Level 3
Act for me

The buyer's money is committed in advance. Software transacts without asking again.

  • Limit orders
  • StockX bids
  • DSP bidding
  • Priceline NYOP

The received view is that consumers climb this — get comfortable with alerts, then graduate to delegation. The record does not show that. Level 3 appears immediately and completely in markets that qualify for it, and never appears at all in markets that don't, regardless of how mature and heavily used the level 1 product is.

Google Flights has run price tracking at enormous scale for more than a decade and has added price history, a "cheapest time to book" window, and a price guarantee that refunds you after you buy. REPORTED It has never added "buy it when it hits my number." That is not a roadmap gap. Twelve years of the most-used flight search product on earth is enough time. The reason is that a flight is not a fungible unit: departure time, carrier, layovers, seat, bag policy and cabin all move independently, and a machine filling the order at $312 can be wrong in a way a machine filling a limit order at $312 cannot.

03 / What history shows

The cases

Equity limit orders
Dominant mechanism

The reference case, and the one that sets the bar for what "works" means. A limit order is a fully funded persistent bid with transaction authority and no further human involvement.

What the buyer declares

Security, quantity, maximum price, duration.

Persistence

Day, or good-till-cancelled.

Authority

Level 3. Fills without asking.

Measured outcome

Limit orders are 25.5% of retail orders and 29.2% of submitted shares across 27 million orders from 19 US retail brokers. FACT

The detail that gets left out: across a range of electronic markets, roughly 70–80% of limit orders end in cancellation rather than execution. REPORTED Retail limit orders sit open for over 20 minutes on average, and further from the touch, closer to an hour. FACT The most successful buyer-intent mechanism ever built has a low fill rate by design, and nobody calls that a conversion problem. Any consumer product that treats unfilled bids as failed sessions will strangle the mechanism in the first quarter.

Programmatic advertising
Dominant mechanism

Advertisers declare audience, context, frequency and price. The impression does not exist until a page loads. Machines resolve it in milliseconds.

Before

Insertion orders, negotiated placements, humans on both sides.

Persistence

Campaign-length standing instructions.

Authority

Level 3, uncontested.

Measured outcome

Roughly 91–92% of US digital display spending is now transacted programmatically, up from about 24% in 2013. REPORTED

This is the strongest evidence that buyer-declared intent can take an entire market. It is also the case with the least transferable buyer: professional, repeat, indifferent to any single unit, and evaluated on aggregate outcomes rather than on whether they liked the specific impression they got.

Truckload freight — load boards
Dominant mechanism

The buyer of capacity posts the requirement first. Carriers with idle equipment respond. This is the clearest large market where the demand side goes first and the supply side answers.

What the buyer declares

Lane, date, equipment type, weight, rate.

Persistence

Hours to days, then it decays.

Authority

Level 2 by default; level 3 where instant-book and automated acceptance are enabled.

Measured outcome

DAT reports brokers and shippers posting more than 722,500 loads per business day. VENDOR-SOURCED

Why sellers answer here: a truck rolling empty earns nothing and the deadhead cost is immediate and known. The responding side has decaying capacity and can compute the cost of not answering. That condition recurs in every case where sellers genuinely responded to latent demand.

StockX bid/ask
Dominant in category

The only mass-consumer marketplace that runs a real two-sided book. Bids are card-backed, visible to the whole market, persist for a chosen duration, and execute automatically when an ask meets them.

What made it possible

Collapse of the good to SKU plus size. Every unit at that spec is identical.

Trust mechanism

Central authentication. Counterparty identity is irrelevant to the buyer.

Persistence

Buyer sets it: 1 to 60 days, with expiry reminders. FACT

Evidence gap

No public split between bid-filled and Buy Now volume. UNRESOLVED

Two design decisions carry this case. Buyers can only purchase at the lowest ask, and sellers can only sell to the highest bid. FACT Removing the choice of counterparty is what makes a standing bid safe to leave unattended. It also means the buyer has pre-agreed that all units at a specification are the same — the exact agreement no ticket buyer has ever been asked to make.

Priceline Name Your Own Price
Retired

The most famous consumer attempt. A binding, card-backed, one-shot bid against opaque inventory. It worked for years and was then killed category by category, not by regulators or by suppliers, but by a better button.

What the buyer declared

Dates, route or city, quality tier, price. Not the supplier.

Persistence

None. Single bid, 24-hour lockout, accept or reject.

Authority

Level 3, binding and non-refundable.

Timeline

Flights discontinued September 2016; rental cars March 2018; replaced by fixed-price Express Deals. FACT

Priceline's own stated reason for the car rental shutdown was speed: bidding required consumers to bid and wait, and Express Deals eliminated both steps. REPORTED That is the whole finding. Academic work on the channel measured the frictional cost of placing a single NYOP bid at a median of roughly €3.50–6.00 depending on category. FACT A declaration costs the buyer something real, and it is charged per declaration, not per purchase. Buy Now costs nothing and resolves immediately.

Priceline WebHouse Club
Failed in 11 months

Name-your-price extended to groceries and gasoline. The cleanest natural experiment in the whole set, because buyer demand and seller response can be separated.

Buyer side

~2 million customers, ~100,000 new per week at shutdown. FACT

Distribution

7,200 grocery stores, 6,000 gas stations participating. FACT

What broke

CPG brands would not fund the discount. WebHouse paid the difference itself. REPORTED

Outcome

Wound down October 2000, less than a year after launch. Customers refunded. FACT

The good was not perishable. A tube of toothpaste unsold today is sold tomorrow at full price, so a manufacturer has no reason to answer a standing bid below list. In travel, the seat expires. That difference, not consumer enthusiasm, decided both outcomes.

eBay proxy bidding
Delegation offered and refused

The single most instructive case for anyone building agent-mediated buying. eBay gave every buyer a genuine level 3 agent — submit your maximum, the machine bids for you — and instructed them to use it exactly that way.

What buyers did instead

Bid at the last moment by hand. Sniping appeared in roughly 37% of eBay auctions. REPORTED

The control

Amazon ran the same second-price proxy system with a soft close. Last-minute bidding there ran under 1%. REPORTED

Direction of learning

On eBay, late bidding increased with bidder experience. On Amazon it decreased. FACT

Interpretation

The refusal was strategic, not psychological.

Buyers were not uncomfortable with automation. Many of them delegated to third-party sniping services instead, which placed the same bid seconds before close. REPORTED They refused to reveal a reservation price early into a market where revealing it invited someone to bid it up. Whether buyers delegate is set by the closing rule and the information structure, not by their comfort with software.

eBay Best Offer
Useful secondary mechanism

Buyer proposes a price on a fixed-price listing. Worth studying because of where the automation authority sits.

Buyer's side

Level 2. Offer expires after 48 hours if the seller does nothing. FACT

Seller's side

Level 3. Seller sets auto-accept and auto-decline thresholds. FACT

Category exclusions

Not available on auction listings in cell phones, motors, real estate, tablets — and tickets. FACT

Reciprocal feature

Sellers can send offers to buyers who watched or carted an item in the prior five days.

In the largest consumer marketplace, the software that decides is the seller's, and the buyer's declaration has a two-day half-life. eBay also chose to exclude tickets from the mechanism. The reason is not published UNRESOLVED but the choice is on the record.

B2B reverse auctions — FreeMarkets, Ariba
Useful secondary mechanism

The buyer publishes a specification and suppliers bid it down. Genuinely worked and genuinely never took over.

Measured savings

Around 15% reported in 1998; roughly 20% across 146 events in later academic work; Ariba cited about 20% per project. REPORTED

Persistence

None. Fixed-duration events.

Terminal state

FreeMarkets sold to Ariba for $493M in 2004; Ariba to SAP in 2012. A feature inside procurement suites. FACT

Constraint

Supplier relationship damage; savings overstated when misapplied. REPORTED

Buyer-declared intent produced real price improvement and still lost to relationship-based sourcing for anything other than commodity categories. Twenty percent is a large number and it was not enough.

Reverse recruiting — Hired, Vettery
Collapsed with the cycle

Candidates declare what they want and employers bid. Worked well enough to raise real money, then stopped working when the scarce side flipped.

Scale

Hired raised $133M and was valued near $500M. REPORTED

Outcome

Near wind-down in 2020; acquired by Vettery (Adecco) November 2020; folded into LHH June 2024. REPORTED

Why it broke

When employers had abundant applicants, they stopped paying a premium to bid for declared candidates. REPORTED

The mechanism only holds when the declaring side is the scarce side. Reverse marketplaces are a bet on which way scarcity points, and scarcity moves.

UK energy auto-switching
Collapsed

The purest consumer test of level 3 delegation to a third party with a commission attached. Standing instruction: move me whenever you can save me £50.

Peak adoption

Look After My Bills ~200,000 members; Flipper ~30,000, against roughly 28 million UK households. REPORTED

Failure trigger

Flipper closed September 2021 when wholesale prices spiked and cheap tariffs disappeared. FACT

Aftermath

A 2024 report to Ofgem alleged around 20,000 households moved without their knowledge, with commissions up to £90. REPORTED

Two lessons. Delegated transaction authority is only as durable as the spread it harvests. And when an agent transacts on commission, the agency problem eventually becomes a regulatory problem. Anyone building level 3 with a take rate is building this.

Songkick Detour
Did not scale

The closest historical analogue to declared intent in live events. Fans pledged a price for a show that did not exist, and Songkick assembled promoters against the demand.

Beta result

Over 1,000 users; ten concerts made to happen. REPORTED

Commitment level

Card on file, but withdrawable. Pledgers could decline if the final price or date did not suit. REPORTED

Resistance

Managers were reportedly receptive; booking agents were the most conservative stakeholder. REPORTED

Terminal state

Not present in Songkick's current product; no public shutdown data. UNRESOLVED

Detour never asked for a binding commitment, which makes it a weaker test of buyer intent than it looks. What it does establish is that the resisting party in live events was the layer that controls whether the event happens at all.

Hopper Price Freeze
Working consumer mechanism

Not a bid. An option. The buyer pays cash now to hold a price for a defined window, and separately decides later whether to buy.

Price of the option

Roughly $5–$40 per traveler. REPORTED

Scale

Fintech products including Price Freeze accounted for about 40% of $7.5B in 2024 bookings. REPORTED

What it proves

Consumers will pay real money before the purchase decision, when the payment is small, priced, immediate and bounded.

This is the most useful consumer datapoint in the set and it is routinely misread as evidence for persistent bids. It is evidence for the opposite: consumers paid to preserve the decision, not to hand it over.

LendingTree
Became a lead business

Borrower declares need and terms; lenders respond. The marketing promise was competition. The mechanism that survived was targeting.

Scale

Over 35 million loan requests since inception. REPORTED

How it actually clears

Lenders specify the borrower profile they want and what they will pay for the introduction. REPORTED

Why not an auction

Underwriting. The product is priced to the individual after assessment, so no supplier can quote against a spec alone.

04 / Findings

What the record actually establishes

Finding 01

Fungibility is the gate, and it is binary

Every dominant case reduced the good to a specification where all units at that spec are interchangeable: a share, a size-10 SKU, a lane-date-equipment triple, an impression matching attributes. Every case that stalled at level 2 involved a good that could not be reduced that way. There is no intermediate case in the set where partial fungibility produced partial adoption. INFERENCE

Finding 02

Sellers are the constraint, not buyers

WebHouse had two million customers and died of supplier refusal. Hired had candidates and died of employer indifference. Detour had fans and hit booking agents. Priceline NYOP had bidders and was replaced because a faster path converted better. In the entire case set there is no instance where a buyer-intent mechanism failed because buyers would not declare. FACT — across cases

Sellers answer standing bids under one condition that recurs everywhere: their inventory decays and they can price the cost of not answering. Empty truck. Empty seat. Unsold impression. Where the good keeps its value if unsold, sellers post prices and wait.

Finding 03

Level 3 is not the top of a ladder. It is a different market structure

Consumers did not progress from alerts to delegation anywhere in this set. Delegation was present from day one where the market qualified and absent permanently where it did not. The most heavily used level 1 products in the world — flight price tracking, restock alerts, watchlists — have not evolved into level 3 despite a decade or more of runway and obvious commercial incentive. INFERENCE

Finding 04

When buyers refused delegation, the cause was market design

The eBay and Amazon comparison is close to a controlled experiment: same auction format, same proxy agent, different closing rule, and an order-of-magnitude difference in whether buyers used the agent as intended. Buyers who refused eBay's agent frequently hired a different automated agent to snipe for them. Comfort with software was not the variable. FACT

Finding 05

Unfilled bids are the normal case and must be designed for

Roughly 70–80% of limit orders across studied electronic markets end in cancellation. The mechanism is still the backbone of the world's deepest markets, because resting bids are themselves the product: they are displayed liquidity, and they are information the other side pays attention to. REPORTED A consumer marketplace that measures buyer intent by conversion rate will conclude the feature failed within a quarter and be wrong.

Finding 06

Declaring has a per-declaration cost that Buy Now does not

Measured at a median of roughly €3.50–6.00 per NYOP bid. FACT Priceline retired the mechanism in two categories citing exactly this: bid and wait versus book now. Every buyer-intent product competes against an instant-purchase button and loses on speed, certainty, and cognitive load. It can only win on price, scarcity access, or the removal of search effort — and it has to win on one of those by a visible margin.

Finding 07

Consumers pay in advance to preserve a decision, not to surrender one

The only consumer mechanism in this set that reliably takes money before the transaction exists is the priced option: Hopper Price Freeze, at $5–$40, attached to a business where fintech products drove 40% of $7.5B in bookings. REPORTED The consumer bought time and optionality. This is evidence against the persistent-bid thesis at least as much as for it.

05 / The opposite case

The strongest version of "consumers just want better alerts"

Stated fairly, because it is well supported.

One. The largest and most durable consumer-side intent products in existence are all level 1 and 2. Google Flights tracking, watchlists, restock alerts, saved searches. None of them has been extended to transaction authority by companies with every incentive and resource to do so.

Two. Where a consumer marketplace does automate acceptance, it gives the authority to the seller. eBay's auto-accept sits on the seller's side; the buyer's offer expires in 48 hours.

Three. eBay's own auction data shows buyers actively working around a delegation tool they were told to use, and doing it more as they gained experience.

Four. The one consumer market where buyers were asked for a binding, funded, blind commitment — Priceline NYOP — was killed in the categories where an instant alternative existed, and Priceline said speed was the reason.

Five. Third-party delegated transacting in a consumer category produced a regulatory complaint about consumers being moved without their knowledge.

What the opposite case does not survive: StockX. A mainstream consumer audience, largely young, placing card-backed persistent bids that execute without a second confirmation, in a category with no urgency and no institutional buyers. If the thesis were "consumers won't delegate," StockX should not exist. The correct reading is narrower and more useful: consumers delegate readily once they have accepted that all units at a specification are the same. That acceptance, not the delegation, is the hard part.

06 / The pattern

Conditions under which buyers committed money before the transaction existed

Derived from the cases that worked, tested against the cases that didn't. Every dominant case satisfies all seven. Every failure violates at least two.

01

The unit is fully specified

The buyer can write down what they want completely enough that a machine filling it cannot be wrong. This is the gate. Nothing else matters if it fails.

02

Counterparty identity is removed from the decision

Escrow, authentication, or central clearing. StockX buyers can only buy the lowest ask; equity buyers never know who sold. A standing bid is only safe to leave unattended if who fills it cannot make it worse.

03

The commitment is bounded and priced up front

The buyer knows the maximum cash at risk before declaring. Price Freeze charges $5–$40. A limit order commits exactly the limit. Open-ended exposure kills participation faster than price does.

04

Waiting has a visible cost

Scarcity or volatility the buyer already believes in. Where the buyer thinks the price will hold, they wait and use level 1. Where they think it will move against them, they commit.

05

The seller's inventory decays and they can price the loss

The recurring condition on the answering side. Empty truck, empty seat, unsold impression. WebHouse failed on exactly this: toothpaste does not expire.

06

Declaring is cheap and withdrawing is not punished

Cancel freely, adjust freely, expire quietly. The measured frictional cost of a single Priceline bid was several euros because the bid was one-shot, binding, and locked the buyer out for 24 hours. Compare StockX: change or cancel any time before match.

07

Most declarations expiring unfilled is an acceptable outcome

Structurally, not just tolerated. The unfilled bids have to earn their keep as displayed liquidity or as information the sell side responds to. If they are dead weight, the economics don't work.

07 / Synthesis

What history suggests

A marketplace built around persistent buyer intent can create materially better liquidity than a listing marketplace, and has done so in equities, programmatic advertising, truckload freight and sneaker resale. In each case it did not improve the existing market. It replaced it.

But the improvement came from a specific trade, and it is the same trade every time. The buyer gives up the right to choose the individual unit, and receives in exchange price improvement, access to inventory that was never listed, and the removal of search. Fungibility is what makes that trade rational. Authentication or clearing is what makes it safe. Bounded, priced commitment is what makes it easy to say yes to. Decaying seller inventory is what makes anyone answer.

Where the buyer will not give up choice of unit, the mechanism does not become a smaller version of itself. It converts into something else entirely: a lead, an alert, or a paid option that hands the decision straight back. LendingTree, Google Flights and Hopper are all the same outcome in different clothes.

Consumers are not the obstacle. Consumers delegate immediately, in large numbers, with real money, once the specification problem is solved. The obstacle is that in most consumer markets the specification problem has never been solved, and the second obstacle is that sellers with non-decaying inventory have no reason to answer.

08 / Limits

Where the analogy breaks for consumer ticketing

A ticket is not a SKU

StockX collapsed sneakers to model plus size and every unit at that spec is identical. A ticket carries section, row, seat, quantity, adjacency, view obstruction, aisle, and delivery method — and buyers care about all of them, differently, per event. Collapsing tickets to a fillable spec means asking buyers to accept substitution risk they have never been asked to accept in this category. That is a genuinely new ask, not a smaller version of an old one.

The good expires at a known time and then is worth zero

Shares and sneakers persist. A resting bid on a stock is patient capital. A resting bid on a ticket has a hard wall, and the buyer's option value collapses at the same instant the seller's does. This cuts both ways: it strengthens condition 05 dramatically, because ticket inventory decays harder than almost anything, and it weakens conditions 04 and 07, because there is no next week.

The answering side is professional and already automated

In freight and programmatic, the responding side had idle capacity and an incentive to clear it. In ticket resale, a large share of the responding side is brokers running pricing software. A visible bid book in that environment publishes buyer reservation prices to counterparties with the tooling to price against them. In equities this is fine because bid competition is deep and units are identical. In a market where one seller holds the only pair of adjacent seats in section 112, it is not obviously fine. INFERENCE

The primary market sets the rules

In every dominant case, the marketplace controlled the clearing rules end to end. In ticketing, onsale timing, holds, transfer restrictions and resale permissions are set upstream by parties with their own economics. Detour's most conservative stakeholder was the booking agent, and that layer still exists.

The largest consumer marketplace excluded this category

eBay's Best Offer is unavailable on auction listings in tickets. The rationale is not published, but a company that owned StubHub for over a decade made that call.

Agentic commerce changes one input, not the gate

Software that can monitor continuously and act reduces the cost of declaring, which is condition 06, and that is a real improvement — it is roughly the Priceline frictional cost problem being solved. It does not touch conditions 01, 02 or 05. An agent that can transact still needs a specification it cannot get wrong, a counterparty structure that makes identity irrelevant, and a seller with a reason to answer.

09 / The open question

The question XP would still need to prove

Will a fan accept a seat they did not personally look at, and still be happy after they see it?

Everything else in this study transfers or can be engineered. Fungibility can be approximated with tight enough parameters. Trust can be bought with escrow and guarantees. Commitment can be bounded and priced. Seller response can be tested against decaying inventory, which ticketing has more of than almost any market.

What history cannot answer is whether a consumer who has spent twenty years choosing a specific seat from a map will accept a fill inside a stated envelope — section range, price ceiling, quantity, adjacency — without a final look. In every market where level 3 worked, the buyer had already conceded that all units at a specification are the same, and that concession pre-existed the marketplace. No ticket buyer has ever made it.

The test is not whether people will set up a bid. They will; declaring is the easy part and every case in this set confirms it. The test is post-fill satisfaction, and specifically the rate at which filled buyers would run the same bid again. That number is unknowable from history and knowable from about two hundred real fills.

Method

Cases were selected on structural match to the mechanism — buyer declares terms first, marketplace searches, matches, routes, or waits — rather than on outcome. Failures were sought deliberately and are represented at roughly the same weight as successes. Ticketing and agentic commerce were excluded from case selection and reintroduced only in the limits section, per the brief.

Three states were used for each case: dominant transaction method, useful secondary mechanism, or failed. Where a mechanism persisted in name but changed function, it is coded as secondary and the functional change is stated (LendingTree, B2B reverse auctions).

Evidence labels

FACT — company filing, regulatory document, company help documentation, or peer-reviewed measurement. REPORTED — credible contemporaneous or secondary reporting. VENDOR-SOURCED — published by a company selling the thing being measured; treated as directional. INFERENCE — conclusion drawn across cases, labeled as such. UNRESOLVED — recorded but not established.

Key evidence gaps

No public figure exists for the share of StockX volume that clears against resting bids versus Buy Now. This is the single most valuable missing number in the study, because StockX is the only mass-consumer case of level 3 buyer authority and the bid-versus-instant split would quantify the appetite directly. The finding that StockX proves consumer delegation is possible does not depend on it. Any claim about how much consumers prefer bidding would.

No adoption data was located for offer-based mechanisms on any major ticket resale platform. Songkick Detour's discontinuation is not documented with a date or a reason in any source found; its absence from the current product is observable, the cause is not. Amazon Subscribe & Save adoption figures range from 23% to 35% of customers depending on survey methodology, come from third parties rather than Amazon, and were judged too weakly sourced to carry a finding; the case is not included above.

The 70–80% limit order cancellation range is drawn from studies across several venues and asset classes over different periods and should be read as a range, not a single market statistic. The eBay 37% sniping figure and the Amazon sub-1% comparison come from the Roth and Ockenfels work on early-2000s auction data; the format has since declined sharply on eBay, from about 96% of active listings in 2003 to 7.2% by 2015, so the behavioral finding is well evidenced and the market it describes is much smaller now.

What this study does not claim

It does not establish that any of the seven conditions is individually necessary. It establishes that all seven are present in the dominant cases and that at least two are violated in each failure, which is a pattern rather than a proof. It does not measure consumer willingness to bid in ticketing, which is untested. It does not claim that agent-mediated buying will or will not change the picture; it identifies which of the seven conditions agents plausibly affect and which they do not.

Numbering note

Filed as 037. The library currently holds 001–031 and 033–036; no study numbered 032 was present in the reviewed set.

PastBehavior Research. Compiled August 2026 from public sources.

Download PDF