BRENNEN

Industries / Technology / Marketplaces and gig platforms

Setting prices and fees

Prices on a marketplace are set by sellers, by their software or by the platform's rules, and the platform's own fees are the price sellers pay to reach its buyers.

7

Milestones achieved, from 2009 to 2026.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2009

    Set prices replace auctions

    eBay started in 1995 as an online auction site, but by 2009 more than half the value of its sales was at fixed, posted prices. In eBay's data, fixed-price listings made up 20.9% of the dollar value of sales in 2005 and 53.9% in 2009, and the "auction discount", the gap between auction prices and fixed-price sales of the same item, grew from within 5% in 2003 to more than 16% in 2009. The authors found that the shift came mostly from buyers turning away from auctions, not from changes in what was sold or who sold it.

    LessonThe sales format that made the first big online marketplace famous lost out once buyers preferred a known price.

    nber.orgapi.crossref.org

    Credit

    Liran Einav, Chiara Farronato and Jonathan Levin of Stanford University and Neel Sundaresan of eBay Research Labs; the paper appeared in the Journal of Political Economy in 2018

  2. 2010

    Services at a set price

    Fiverr, founded in 2010 and based in Tel Aviv, lists freelance services as "Gigs" with a fixed price, scope and delivery time, a model it calls Service-as-a-Product; at first the base price of a Gig was $5. It later let freelancers set base prices above $5 and offer several tiers. Fiverr keeps a share of each order: its take rate, revenue divided by the value of orders, was 25.7% in 2018 and 24.5% in 2017, by its 2019 filing to the SEC.

    sec.gov

    Credit

    Fiverr

  3. 2011

    Two pricing bots run away

    In April 2011 the biologist Michael Eisen found new copies of Peter Lawrence's out-of-print textbook The Making of a Fly listed on Amazon for more than $1 million. Two sellers were setting their prices from each other's by formula: profnath at 99.83% of bordeebook's price, and bordeebook at about 127% of profnath's, so each round raised both. The price peaked at $23,698,655.93 on 18 April, when profnath cut it to $106.23.

    LessonA rule that prices only against a rival needs a ceiling and a person who looks at the result.

    scholarlykitchen.sspnet.org

    Credit

    Michael Eisen, who worked it out; Kent Anderson of The Scholarly Kitchen, who reported it

  4. 2016

    Repricing software runs a cartel

    The UK Competition and Markets Authority (CMA) found that two sellers of posters and frames on Amazon's UK site, Trod Ltd and GB eye Ltd (trading as GB Posters), had agreed not to undercut each other from 24 March 2011 at the latest to 1 July 2015 at the earliest. They carried out the agreement with automated repricing software that each configured for it. The CMA fined Trod £163,371 on 12 August 2016; GB eye was not fined because it reported the cartel. In April 2015 the US Justice Department had brought what it called its first online marketplace prosecution, over price fixing on Amazon by algorithm.

    LessonSoftware that carries out a price agreement does not make it lawful; the agreement is the offense.

    gov.ukgov.ukmislove.org

    Credit

    Competition and Markets Authority; US Department of Justice, Antitrust Division

  5. 2022

    Amazon's Buy Box opened up

    The Buy Box is the panel on an Amazon product page that shows one seller's offer and lets a buyer purchase it with one click. On 20 December 2022 the European Commission made binding Amazon's commitments to treat all sellers equally when it picks the Buy Box winner, to show a second competing offer when one differs enough on price or delivery, to set the same conditions for all sellers in its Prime program, and not to use sellers' non-public data for its own retail business. The Buy Box and Prime commitments run for seven years and the rest for five, under an independent trustee, and a breach can be fined up to 10% of Amazon's yearly turnover.

    LessonThe rule that decides whose price a buyer sees first is itself a pricing decision that a regulator can review.

    ec.europa.eu

    Credit

    European Commission (Margrethe Vestager, competition commissioner)

  6. 2022

    Sellers strike over a fee

    On 11 April 2022 Etsy raised its fee on each sale from 5% to 6.5%, a few years after raising it from 3.5% to 5%, its first rise since its founding in 2005. More than 14,000 of its 5.3 million sellers put their shops on hold for a week, and a petition by the seller Kristi Cassidy passed 48,000 signatures. The petition also objected to Offsite Ads, compulsory since February 2020 for sellers with more than $10,000 a year in sales, which it said "can take more than 20% of each transaction". Etsy said the money would pay for marketing and a 20% larger support team.

    LessonSellers judge a marketplace by the total it takes from a sale, ad charges included, and not by the headline rate.

    techcrunch.com

    Credit

    Kristi Cassidy and the striking sellers; Etsy, led by Josh Silverman

  7. 2026

    Repricers that push prices up

    Leon Musolff studied repricing tools, software that changes a seller's prices automatically, in a new e-commerce data set. Sellers that started using them cut their prices by 16.93%, and market prices fell by 9.67%. But the tool makers added "resetting" strategies that raise prices at intervals in the hope that rivals follow; where one was used in a market with fewer than six serious competitors, prices eventually rose by 11.4%. Such price cycles were still rare in the data.

    LessonThe same tools that start price wars can be set to end them.

    api.crossref.org

    Credit

    Leon Musolff, in Management Science, published online on 9 July 2026

Watch for

  1. Seller agents that set prices

    Amazon said in September 2026 that Seller Assistant, its AI agent for marketplace sellers, can now run standing workflows: for example, watch a product category and, within limits the seller sets, change prices and update listings without being asked again, with each action kept in an audit trail. Amazon says the assistant reaches more than 90% of its selling partners and has hundreds of thousands of active users, and that sellers accept its recommendations more than 90% of the time.

    Achieved whenAn outside study measures how many sellers let the agent change their prices, and how prices and price swings move on the products where it does.

    distributionstrategy.com

    Claimed by

    Amazon, September 2026

Open

  1. 01

    Crash tests for pricing bots

    An open test bench in which common repricing rules (match the lowest price, undercut by a cent, price at a fixed ratio to a rival, floors and ceilings, periodic resets) and AI-model pricing agents compete for a simulated Buy Box with synthetic buyers. It reports which settings run away, as the fly book did, and which settle above the competitive price, and gives tool makers a pass or fail check to run before release. All sellers and buyers are simulated.

    The brief
    Credit · unclaimed
  2. 02

    What each sale costs sellers

    A public table, updated monthly, of what 25 marketplaces and freelance platforms take from the same two reference sales, a $50 item shipped within the country and a $500 freelance job, in 5 countries. It adds up listing, transaction, payment processing, compulsory advertising, currency conversion and payout fees from published fee schedules, links every figure to its clause, and comes with a calculator for other prices.

    The brief
    Credit · unclaimed

Open 01 · the brief

Crash tests for pricing bots

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open test bench in which common repricing rules (match the lowest price, undercut by a cent, price at a fixed ratio to a rival, floors and ceilings, periodic resets) and AI-model pricing agents compete for a simulated Buy Box with synthetic buyers. It reports which settings run away, as the fly book did, and which settle above the competitive price, and gives tool makers a pass or fail check to run before release. All sellers and buyers are simulated.

Why it is still open

Research simulators test stylized agents, not the settings that sellers buy. Fish, Gonczarowski and Shorrer found that pricing agents built on large language models "quickly and autonomously reach supracompetitive prices" (accepted to the EC 2026 conference), a 2025 Barcelona School of Economics thesis extended their setup to five sellers, and an open simulator, price_simulator, covers learning agents. Musolff found that real repricers' resetting strategies raise prices. None of them offers a test that a repricing product's own rules can be run against, with a pass mark set in advance.

arxiv.orggithub.comgithub.comapi.crossref.org

Built with what exists

Python; price_simulator or PettingZoo, a standard interface for environments with many agents; open-weight models such as Mistral Large 4 (Note: "A trillion open parameters") run locally with llama.cpp on the ggml library (Note: "A first stable number for local AI"); the Buy Box rules in the European Commission's 2022 commitments as one way to pick the winning offer.

  1. First step

    Rebuild the fly-book loop with two simulated sellers at ratios of 0.9983 and 1.27, and show the price ceiling that stops it.

  2. Done looks like

    12 market scenarios and 6 rule families, 1,000 runs each, published price paths, and a pass mark stated in advance, for example no price above 10 times its starting level within 30 rounds and average prices within 5% of the competitive price computed for each scenario.

Open 02 · the brief

What each sale costs sellers

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table, updated monthly, of what 25 marketplaces and freelance platforms take from the same two reference sales, a $50 item shipped within the country and a $500 freelance job, in 5 countries. It adds up listing, transaction, payment processing, compulsory advertising, currency conversion and payout fees from published fee schedules, links every figure to its clause, and comes with a calculator for other prices.

Why it is still open

Companies report their take rate only as a yearly average over all sales, such as Fiverr's 25.7% in 2018, and eBay says only that its take rate rose with managed payments. Fee pages are built differently on each platform, and compulsory ad charges such as Etsy's Offsite Ads are left out of headline rates. The closest public work, FLUF-io's comparison (last updated March 2026), lists headline seller and buyer fees for 12 resale and shop platforms, with no freelance platforms, no per-sale totals and no clause citations.

github.comsec.govsec.gov

Built with what exists

Open Terms Archive's engine to record each version of the fee pages; Python and pandas; a spreadsheet calculator; an AI model to draft each extraction, checked by a person.

  1. First step

    Fix the two reference sales and work them out by hand for eBay, Etsy, Amazon, Upwork and Fiverr in the US.

  2. Done looks like

    25 platforms, 2 reference sales and 5 countries, every figure cited, totals within 1% of each platform's own fee calculator where one exists, and six monthly updates with changes logged.