Shrinkflation is the practice of reducing the size or quantity of a product while holding the price roughly constant, so the price per unit rises without the shelf price moving. It is the most quietly used lever in consumer goods pricing, and the evidence on whether it works splits into two literatures that appear to flatly contradict each other.
They do not contradict each other. They measure different people. Once you see that, shrinkflation stops being a pricing tactic and becomes something more uncomfortable: a liability that sits latent on your balance sheet until somebody tells the shopper.
What the transaction data says: it works, and almost nobody reacts
The best evidence available is Janssen and Kasinger, "Shrinkflation and Consumer Demand", published in Marketing Science (45(1), 142-158, 2026; published online 28 October 2025). They analysed a decade of NielsenIQ Retail Scanner data covering roughly four million products across about 50,000 U.S. retail establishments, spanning 1,100 product modules.
The findings are specific:
- 1.92% of products were downsized, against 1.11% upsized
- Downsized products accounted for 1.89% of total sales, or $38.57 billion; upsized products just 0.35%, or $7.10 billion. By sales, downsizing is more than five times as prevalent as upsizing
- Downsizing typically arrived without a corresponding price cut, producing an average 12% increase in price per unit volume. Upsized products showed roughly a 2% decrease
- One year after the change, downsized products showed an average 6% sales increase; upsized products 15%
Then the number that explains why the practice persists. In their benchmark demand model, the weighted average price elasticity was -1.19 while the size elasticity was 0.56. Consumers were about 2.1 times more responsive to price changes than to size changes. Split by direction, it gets starker: for downsized products the size elasticity was approximately zero, while for upsized products it was 0.77. Price elasticities were similar across both groups at roughly -1.16.
Read that last pair again. Shoppers respond when a pack gets bigger. They do not respond when it gets smaller. The authors' conclusion is blunt and commercially accurate: reducing product size is an effective way to increase margin or absorb cost pressure.
This is the empirical foundation under every shrinkflation decision ever taken, and on its own terms it is correct.
What the fairness research says: it is worse than a price rise
Now the other literature. Evangelidis, "Frontiers: Shrinkflation Aversion: When and Why Product Size Decreases Are Seen as More Unfair than Equivalent Price Increases", Marketing Science (43(2), 280-288, 2024), ran five preregistered experiments.
The finding: while the vast majority of people judge cost-driven price increases to be fair, that pattern is attenuated or even reversed for downsizing. A larger share of consumers view product downsizing as unfair than view an economically equivalent price increase as unfair.
The mechanism is the important part. Shrinkflation aversion is driven predominantly by the belief that downsizing is deceptive. It is not the money. Two moderators do the work: whether the change was disclosed transparently, and whether the firm's costs genuinely rose.
So a manufacturer facing real input-cost inflation has two options that cost the shopper the same amount. One is widely accepted as fair. The other is read as an attempt to trick them.
| Janssen and Kasinger (scanner data) | Evangelidis (preregistered experiments) | |
|---|---|---|
| What it measures | Revealed behaviour at the shelf | Stated fairness once informed |
| Who is in the sample | Shoppers who did not detect the change | Shoppers who have been told about the change |
| Headline result | Size elasticity approximately zero for downsized products | Downsizing judged more unfair than an equivalent price rise |
| Driver | Detection is hard and happens after purchase | Perceived deception, not the money |
| What it implies | Downsizing reliably protects margin | Downsizing carries a reputational liability |
The two findings are about two different groups of shoppers
Here is the resolution, and it is the point of this article.
The scanner data measures the shoppers who did not notice. The fairness experiments measure shoppers who have been told. Between those two populations sits a single event: disclosure.
A size elasticity of approximately zero is not evidence that shoppers are relaxed about downsizing. It is evidence that they did not detect it. Detection is structurally hard at the shelf, and it is decoupled from the moment of purchase in a way price changes never are:
- A price change is displayed, in currency, at the exact point of decision
- A size change is displayed in grams or millilitres, in small type, and requires the shopper to remember the previous value
- Most shoppers only encounter the consequence at consumption, when the packet runs out sooner, which is days after the money changed hands and far from any decision they can reverse
That is why both results are true simultaneously, and why the tactic looks so attractive in a P&L review. The cost does not appear in the same reporting period as the benefit, and it does not appear in the dataset the decision was made from.
Shrinkflation is not a price decision. It is an unbooked liability that converts into a cost the moment somebody tells the shopper, and you do not control who does the telling.
Regulators are now industrialising the telling
This is the part that changes the calculation for 2026. The disclosure event that converts the liability used to be random: a journalist, a viral post, a sharp-eyed customer. Increasingly it is statutory.
France now requires it directly. Under the Order of 16 April 2024, as amended by the Order of 28 June 2024, effective 1 July 2024, retailers with a sales area greater than 400 square metres selling predominantly food products must display a notice on or beside affected products stating that the quantity sold has decreased from X to Y and the price per unit of measurement has risen by a stated percentage or amount. The notice must run for the first two months of marketing. Bulk products and pre-packaged foodstuffs of variable quantity are exempt. Penalties reach EUR 3,000 for an individual and EUR 15,000 for a company.
Note what that regulation does in research terms. It does not ban the practice. It forcibly moves shoppers from the population with zero size elasticity into the population that considers the practice deceptive, at the shelf, at the point of decision, for two months.
In the United States, the Shrinkflation Prevention Act was introduced by Senator Bob Casey on 28 February 2024, cosponsored by Senators Baldwin, Warren, Rosen, Booker, Whitehouse, Brown, Murray and Sanders. It would direct the FTC to establish shrinkflation as an unfair or deceptive act or practice, authorise FTC civil actions, and let state attorneys general sue. It has not been enacted, and it should be described as a proposal rather than a rule, but it establishes the framing regulators are working from: shrinkflation as deception, which is precisely the mechanism Evangelidis identified.
How to research shrinkflation properly
Most shrinkflation research is worthless because it asks a single question that cannot be answered honestly: "would you notice if this product got smaller?" Everyone says yes. The scanner data says otherwise.
Run it as a two-stage design instead, because you are studying two different populations.
Stage 1: detection, under realistic conditions. Show a shelf, not a spec sheet. Do not mention size. Ask what they would buy and why, then ask whether anything about the pack seemed different. Use a yes_no question for detection and a scale question for confidence. What you are measuring is the rate at which your change crosses into awareness at all, which is the true input to the size elasticity you will observe.
Stage 2: reaction, after disclosure. Now tell them, in the same words a French shelf label or a news story would use. Measure perceived fairness on a scale, ask a single_choice on what they would do next, and use ranking to order the responses they say they would consider: switch brand, switch to private label, buy less often, buy the larger pack, do nothing. Then let the AI moderator probe the reason in open_ended conversation, because "unfair" is a conclusion, not a reason, and the reason is where the recoverable ground is.
Two design points that make or break it:
- Include the honest alternative. Test downsizing against an explicit, transparently communicated price increase of equivalent value. Evangelidis found transparency and genuine cost pressure both moderate the aversion, so the choice is not shrink-or-nothing. Many brands have never quantified what an openly explained price rise would actually cost them, and assume it is worse. The evidence suggests it often is not.
- Ask multiple_choice on where they would expect to hear about it. The disclosure channel determines the damage. A regulator-mandated shelf label lands differently from a stranger on social media claiming your brand is cheating.
This work pairs naturally with a price increase study, with packaging concept testing if the new pack changes shelf appearance, and with the display comprehension methods in reference prices and drip pricing. If the goal is defending a claim about the change, the standards in advertising claim substantiation apply.
Why AI-moderated interviews fit this problem
The traditional options are poor. A Typeform or SurveyMonkey survey can ask whether people would notice, which is the question that produces a false answer. Qualtrics can field it at scale but still cannot follow up on a surprising response. A focus group can probe properly but delivers eight people, and by the time you have run four groups across two markets you have spent six weeks and a large budget establishing something a regulator may make moot.
Koji is the AI-native customer research platform built for this shape of problem. It runs AI-moderated voice or text interviews with several hundred category shoppers in days, probes every unexpected answer the way a skilled moderator would, and returns automatic thematic analysis rather than a folder of transcripts. Because the moderator is an AI consultant you configure rather than a person with a view, nobody nudges a respondent toward the reassuring answer, which matters enormously on a topic where the client is hoping to hear that nobody minds.
And it holds both stages in one study: structured questions for the detection and fairness measurements, open conversation for the reasons. Six structured question types are available - open_ended, scale, single_choice, multiple_choice, ranking and yes_no - so the quantitative and qualitative halves come from the same respondents rather than two separate projects. From question to insight in hours, not weeks, with no research expertise required.
Frequently Asked Questions
What is shrinkflation?
Shrinkflation is reducing the size or quantity of a product while keeping the price the same or nearly the same, which raises the price per unit without changing the shelf price. Research on a decade of U.S. scanner data found approximately 1.92% of products were downsized, typically producing an average 12% increase in price per unit volume.
Does shrinkflation actually work commercially?
On the evidence, yes, in the short term. Janssen and Kasinger found a weighted average price elasticity of -1.19 against a size elasticity of 0.56, meaning consumers were roughly 2.1 times more responsive to price changes than size changes. For downsized products specifically, size elasticity was approximately zero. Downsized products still showed an average 6% sales increase a year later.
If shoppers do not react, why is shrinkflation risky?
Because the near-zero size elasticity measures shoppers who did not notice, not shoppers who approved. Five preregistered experiments found that consumers judge downsizing as more unfair than an equivalent price increase, driven predominantly by the belief that it is deceptive. The risk crystallises when disclosure occurs, and regulators are increasingly mandating that disclosure.
Which countries require shrinkflation to be labelled?
France requires it. Since 1 July 2024, food retailers with a sales area over 400 square metres must display a notice for the first two months stating that the quantity has fallen and the price per unit of measurement has risen, with fines up to EUR 3,000 for individuals and EUR 15,000 for companies. In the United States, the Shrinkflation Prevention Act was introduced in February 2024 but has not been enacted.
Is downsizing better or worse than raising the price?
It depends on whether shoppers find out. Undetected, downsizing outperforms a price rise because size elasticity is far lower than price elasticity. Once disclosed, it performs worse, because it is read as deception rather than as a response to costs. Transparency and genuine cost increases both reduce the fairness penalty, so a clearly explained price rise is often the lower-risk option.
How do I test a pack size change before launching it?
Use a two-stage design. First measure detection under realistic shelf conditions without mentioning size, using a yes_no and a confidence scale. Then disclose the change in the wording a label or news story would use, and measure fairness, intended switching and the reasons behind them. Always include a transparently communicated equivalent price increase as a comparison arm.
Find out what happens when they are told
The size elasticity in your category is a measure of who has not noticed yet. It is not a measure of your risk. The number you actually need is what happens to purchase intent and brand trust in the two months after somebody puts a label on the shelf.
Koji can tell you that this week. AI-moderated interviews with several hundred of your category's shoppers, both stages in one study, automatic thematic analysis, and a report you can put in front of a pricing committee before the decision is taken rather than after the headline.
Test the disclosure, not just the pack.
