---
title: "Personalised pricing software and the return of haggling, run from the seller's side"
kind: "Article"
topic: "Prices & markets"
date: "2026-10-03"
url: "https://canopticon.com/articles/personalised-pricing-and-the-return-of-haggling"
author: "Chris Telford"
license: "CC BY 4.0"
---
This came from a question about Canadian living costs: are the prices of necessities still set by markets at all? The answer starts with something everyone takes for granted. The price tag, one number for every buyer, is a 19th-century invention, and software is replacing it with prices worked out for each buyer from what the seller knows about them. That brings back the haggle, except that now only the seller gets to bargain.

## The price tag is a 19th-century invention

The price on the tag feels like the natural state of things. It is not. <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">Before the mid-19th century, most retail buying in America was done by haggling; merchants intentionally withheld a fixed price (if a merchant thought he could get $10 for an axe, he would not let you see a $7 price).</q> The one-price system began with conviction. Philadelphia's Quaker merchants <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">decided all should pay the same</q>; a New York merchant made it commercial <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">around 1846</q>; and <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">John Wanamaker, who opened his first store in 1861 and formalized fixed pricing and price tags at his Grand Depot department store (1875–1876)</q>, made it ordinary. Scale is what made it stick: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">The department store and the fixed price were the same event: mass scale made person-by-person bargaining impossible, and the posted price was the technology that resolved it.</q>

Treat the tag as a technology and the question changes. Technologies get replaced when something cheaper does the job, and bargaining with each customer one at a time is now cheap. Online prices already move faster than any shelf could follow: in 2013 Amazon was found to <q doc="the-price-frontier-from-discovery-to-administration-a-canopticon-resea" section="Details">make more than 2.5 million price changes per day</q>, a figure the research calls <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">real but a decade old</q>.

## A price worked out for you is not a faster price tag

Two different things get called dynamic pricing, and the distinction carries the argument. A price that moves with supply and demand is still one public number. The research's rule: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Details">Concede dynamic pricing is faster discovery; insist personalized pricing is something else.</q> In the personalised case <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Details">there is no longer a single public price to discover; the seller computes a private number from your data</q>. New York's law draws the same line. Its required notice reads <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">"THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA."</q> and <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">pricing varying solely on supply/demand or time of day is exempt</q>.

The data behind it is detailed. A US consumer-protection study of pricing intermediaries found they <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">"worked with at least 250 clients that sell goods or services ranging from grocery stores to apparel retailers,"</q> and that <q doc="the-price-frontier-from-discovery-to-administration-a-canopticon-resea" section="Details">details like a person's precise location or browser history can be frequently used to target individual consumers with different prices for the same goods and services</q>. The agency's chair at the time described prices set <q doc="the-price-frontier-from-discovery-to-administration-a-canopticon-resea" section="Details">"from a person's location and demographics, down to their mouse movements on a webpage."</q> The research notes the study's limits: its findings are <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">preliminary staff observations, not a finding of illegality</q>.

> **Objection: Sellers have always charged different people different prices**
>
> The research concedes it: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Details">True — first-, second-, and third-degree discrimination (coupons, student discounts, airline fences) are old.</q> What changed is the reach: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Details">machine learning approximates first-degree (perfect) discrimination at population scale, using surveillance data the buyer cannot see or contest</q>. A student discount is a published rule; a price computed from your data is visible only to the seller.

## The haggle comes back, with the seller holding the information

The research's sharpest description of where this leads is the market the price tag replaced: <q doc="the-price-frontier-from-discovery-to-administration-a-canopticon-resea" section="Details">surveillance pricing is the return of the haggle, but with the asymmetry inverted — the seller now knows everything about the buyer.</q> In a bazaar the buyer could read the seller's face and walk away; here the seller reads the buyer's data and the buyer sees nothing.

The incentive is not subtle. In one field experiment, personalized pricing <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="The pricing algorithm">"improves the firm's expected posterior profits by an additional 19%, relative to the optimized uniform price, and by 86%, relative to the firm's unoptimized status quo price"</q>. A merchant who can earn a fifth more by reading each customer will not keep posting one number for long. The research names the end point too: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="The pricing algorithm">The limit case is perfect (first-degree) price discrimination, where consumer surplus is driven to zero.</q>

<figure data-kind="chart" doc="price-without-discovery-from-the-signal-to-the-instruction" section="The pricing algorithm">
caption: Extra profit from pricing each customer separately
- vs. the best single price | +19% | 21 | #6b46a8
- vs. the firm’s usual price | +86% | 96 | #a892d4
note: One field experiment
</figure>

Who gains is contested, and the research does not pretend otherwise. One study of personalised pricing found that <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Details">"over 60% of consumers benefit from lower prices under personalization,"</q> but also that <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Details">"total consumer surplus declines 23% under personalized pricing relative to uniform pricing"</q>, while firm profits rise. The research places the objection accordingly: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Details">The honest position concedes ambiguous welfare and locates the objection in distribution, asymmetry, and autonomy — not in pure efficiency.</q>

## When every landlord uses the same software

Rent shows what happens when the same software serves every seller in a market. In the US litigation, <q doc="realpage-north-algorithmic-rent-setting-in-canada-a-forensic-investiga" section="Key findings">RealPage's clients comprised nearly 90% of the market for multifamily rental housing units of 50 units or more, and court documents cited landlords accepting the software's pricing recommendations 80–90% of the time.</q> The company's own words for the goal were <q doc="realpage-north-algorithmic-rent-setting-in-canada-a-forensic-investiga" section="Key findings">"driving every possible opportunity to increase price"</q>. The US Justice Department's antitrust chief put it in one sentence: <q doc="realpage-north-algorithmic-rent-setting-in-canada-a-forensic-investiga" section="Key findings">"RealPage was replacing competition with coordination, and renters paid the price."</q>

<figure data-kind="chart" doc="realpage-north-algorithmic-rent-setting-in-canada-a-forensic-investiga" section="Key findings">
caption: RealPage in the US rental market
- Share of large apartment units | ~90% | 90 | #c2603f
- Recommendations landlords accepted | 80–90% | 85 | #e0a58f
note: Buildings of 50 units or more
</figure>

The US government's framing is that the meeting has moved into the tool: where collusion once required <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">"a formal handshake in a clandestine meeting," now "algorithms are the new frontier."</q> Prosecutors <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">allege one landlord raised rents more than 25% within 11 months of adopting the software</q>. The settlement filed in November 2025 <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">imposes no fine and no admission of wrongdoing but bars RealPage from using nonpublic, competitively sensitive data in model training and runtime</q>.

> **Objection: Not every pricing program pools competitors' data**
>
> One rival vendor says its product uses <q doc="realpage-north-algorithmic-rent-setting-in-canada-a-forensic-investiga" section="Key findings">only the client's own property data plus publicly available asking rents and "never confidential competitor pricing data"</q>, and a California court accepted that defence. That narrows the claim: the problem is software that pools sellers' private data, not pricing software as such.

## Software can learn to coordinate without being told

The harder question is whether pricing programs drift toward high prices on their own. In simulation, learning algorithms <q doc="the-price-frontier-from-discovery-to-administration-a-canopticon-resea" section="Details">"consistently learn to charge supracompetitive prices, without communicating with one another,"</q> though later work showed <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">the outcome flips with the learning protocol</q>. The strongest real-market study, of German petrol stations, found adoption <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">"increases margins by 9%, but only in non-monopoly markets,"</q> and that margins <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">"increase by 28% in markets where both [duopolists]"</q> adopt. The timing points to learning rather than agreement: the effect appeared <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">"about a year after market-wide adoption, suggesting that algorithms in this market learn tacitly-collusive strategies."</q> The research's own instruction is plain: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">Present this as contested, not proven</q>

> **Objection: This is just central planning by another name**
>
> No, and the research rules that version out: <q doc="the-price-frontier-from-discovery-to-administration-a-canopticon-resea" section="Section 4">seductive but overstated: what's demonstrably happening is private firms administering prices within a market system, not a central computer replacing the market's allocative function.</q> The claim here is narrower: individual firms are setting prices by computation inside an ordinary market. Even the term is not new — <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="The fixed price">"administered prices" is not a neologism — economist Gardiner Means coined the term in the 1930s</q>; what is new is that the setting now reaches each buyer.

## Canada is still looking for the handshake

Canadian use was real, and small. A CBC investigation <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">found YieldStar has been used in Canada since at least 2017</q>; several large landlords said they had used it and stopped, and the vendor said its software <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">has never served more than 1% of the rental market in Canada</q>. Concentration is what makes it matter: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">Algorithmic rent-setting is dangerous in proportion to concentration, and Canadian concentration is high and rising even as YieldStar specifically retreats.</q>

The software has been here for years. The earliest documentary trace is a 2017 RealPage press release naming a Canadian client reporting <q doc="realpage-north-algorithmic-rent-setting-in-canada-a-forensic-investiga" section="Key findings">"2%–4% revenue outperformance"</q>. Both attempts to hold anyone to account have now ended. <q doc="realpage-north-algorithmic-rent-setting-in-canada-a-forensic-investiga" section="Key findings">The Competition Bureau discontinued its civil inquiry into RealPage Canada and Yardi Canada on November 10, 2025, finding adoption "not sufficiently widespread"</q>, and the class action was discontinued before certification. Meanwhile, the Bureau's own discussion paper <q doc="realpage-north-algorithmic-rent-setting-in-canada-a-forensic-investiga" section="Key findings">noted Canada has **over 60 companies offering algorithmic pricing services** across sectors.</q>

The law explains the outcome. <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">Section 45 (criminal conspiracy) requires proof of an agreement beyond a reasonable doubt; section 90.1 (civil) reaches anticompetitive agreements or arrangements but still demands an "agreement" — a "meeting of the minds."</q> <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">Neither reaches tacit algorithmic coordination, and conscious parallelism is not illegal in Canada.</q> The research's summary: <q doc="price-without-discovery-from-the-signal-to-the-instruction" section="Key findings">the law is built for human agreements, and the technology routes around it.</q> A law written to catch competitors agreeing will not catch competitors who never need to.

## Price the same basket on different phones

Rent is the case with court records. Everyday shopping is the case nobody in Canada has measured. The research proposes the test: a tool that <q doc="the-price-frontier-from-discovery-to-administration-a-canopticon-resea" section="Candidate A">detects/documents price variation by user signal (location, device, history) across Canadian e-commerce</q>. If the same basket costs different amounts on different phones, the price tag is already gone, and only the tag on the shelf is left to pretend otherwise.