The $39 Dress That Outsold the $34 One

Pricing-page psychology: what actually survives the research.

A clothing catalog once ran a real experiment. Same dress, three price tests: $34, $39, $44. The $39 dress outsold the $34 dress — charging five dollars more sold more units. In a small pilot, sixteen units at $34, twenty-one at $39. And when Anderson and Simester scaled the test across hundreds of thousands of real catalogs, nine-endings kept winning — up to 35% in one 60,000-catalog study, with the effect strongest for new items and muted when a "Sale" sign was already present.

That result makes no sense if customers compute prices. It makes perfect sense once you know they compare them.

Buyers don't compute prices — they compare them. You don't set a number; you set the comparison. That's what your pricing page is really doing. And most of what you've been taught about it — including the most famous pricing story on the internet — doesn't survive the research.

Four parts. First, anchoring: the effect so robust it moves professional experts while they deny it in writing. Second, the famous story, told correctly for once — and the unglamorous effect that actually deserves its reputation. Third, the psychology of the price digits themselves, including a study of 78 million supermarket prices with a twist ending. Fourth, what 60 real SaaS pricing pages actually do — and then we rebuild a bad one.

Anchors move experts — and they deny it

In 1987, researchers took real-estate agents — professionals with an average of seven years' experience — to a real house in Tucson. Same house, same ten-page information packet, real walkthrough. One thing varied: the listing price printed on the sheet.

Agents who saw a low listing price appraised the house at about $67,800. Agents who saw a high one: about $75,200. Same walls, same roof — a $7,400 swing in professional judgment from one printed number.

Here's the best part. Asked how they reached their appraisals, the experts' written reports — in the paper's words — "flatly denied" using the listing price. Only 24% of the professionals even mentioned it, versus 56% of amateurs. The more expert you are, the more invisible the anchor is to you.

Before you file this under "psychology studies that don't replicate": anchoring is arguably the best-replicating effect in the field. In the Many Labs replication project — 36 labs, more than 6,000 participants — anchoring effects took four of the top five effect sizes tested, and the replications came out larger than the originals.

You will not feel it working. That's the design constraint. In a related study of executives whose bids had just tracked an arbitrary anchor, 72% claimed it had no influence on them.

On a pricing page, anchoring comes down to two humble tools. Order: one eight-week field experiment with a bar menu found that listing expensive items first lifted average spend by about 4% — $6.02 versus $5.78 per beer (Suk, Lee & Lichtenstein 2012, Journal of Marketing Research). Small, real, honest number. And the reference price — "was $79, now $49" — which peer-reviewed work found raises perceived value even when the reference price is implausible. Which is precisely why regulators watch fake anchors: it works, so people fake it, so it's policed. Anchor with numbers that are true.

The famous effect is the fragile one

Now the famous story. The Economist once offered three subscriptions: web-only for $59, print-only for $125, and print-plus-web — also $125. The middle option looks insane. Nobody should pick it. In the standard telling, that's the genius: a decoy that makes the expensive bundle look brilliant.

Here's what actually happened. The Economist never ran an experiment. Dan Ariely saw the ad and surveyed a hundred MBA students — a classroom demo, described in Predictably Irrational (2008), chapter one. With the decoy present, 84 of 100 chose the bundle. Delete the decoy and ask a different hundred students, and the bundle collapsed to 32. It's a great demo. But The Economist measured nothing — zero real purchases. Nobody on earth knows what that pricing did to actual subscriptions.

And when researchers took the decoy effect to realistic products, it fell apart. A 2014 paper — "The Limits of Attraction" — ran the test across 38 studies. With abstract numbers in a table, the effect appears. With products you can see, taste, or hold: zero significant decoy effects in 27 attempts — and occasionally a reversal, where the ugly option next to the target made people flee the whole category.

Even the researchers who discovered the effect in 1982 published a response conceding — their words — that they suspect it "occurs rarely in the marketplace today," and that their original stimuli were tuned until the demonstration worked.

So is tiered-pricing psychology all fake? No — the wrong effect got famous. The famous effect is the fragile one. The one that survives is its unglamorous cousin: the compromise effect. People avoid extremes and gravitate to the middle.

The cleanest demo: two cameras, $170 and $240 — buyers split fifty-fifty. Add a $470 camera on top, and the middle camera jumps to 57% while the cheap one crashes to 22%. Adding a premium option didn't sell the premium option — it sold the middle.

And unlike the decoy, this replicates. A meta-analysis across 142 observations finds extremeness aversion robust: the same product becomes dramatically more attractive when it's positioned as the intermediate choice.

That's why good-better-best exists. Not the decoy. Your third tier isn't there to be chosen — it's there to make the middle safe.

The left digit — and who's really irrational

Now the digits themselves. In the lab, $2.99 feels meaningfully smaller than $3.00 — but $3.59 versus $3.60 does nothing. Same one cent. The difference is the left digit: your brain anchors on it before it finishes reading the number.

How big is this in the real world? A 2023 study analyzed 78 million supermarket price observations across 25 US chains. Crossing a dollar boundary — $4.99 to $5.00 — hits demand like a twenty-cent increase. A one-cent move, priced by shoppers as twenty.

But here's the twist, and it's the best finding in this whole piece: the irrational actor is the firm. Given a bias that size, essentially every price should end in 99. Retailers use it on roughly a third of prices, pricing as if the bias were a tenth of its real size — and the paper estimates they forgo 1–4% of gross profits. The customers are predictably biased; the companies are the ones leaving money on the table.

One boundary before you 99-everything: round prices carry their own signal — quality. Research on hedonic purchases finds round numbers can win for premium, feel-good products — champagne at $40 outperformed $39.72. Charm endings whisper deal. Round numbers whisper class. Choose the whisper that matches your product — because the best software companies already have.

What 60 real pricing pages do

We tallied 60 pricing pages from top software companies ourselves. Four findings.

What 60 pricing pages from top SaaS companies do — our sample, Aug 2026

First: round numbers beat nine-endings. Premium B2B has quietly chosen the quality whisper. Several products split the difference — charm-priced monthly, round annual: Copy.ai shows $49 a month, or a clean $36 when billed yearly. The deal signal on the price they anchor with; the quality signal on the plan they want you on.

Second: the compromise structure is the standard. Over half show exactly three tiers, two-thirds park an unpriced "Contact sales" tier on top as the extreme, and when a tier is highlighted, 83% of the time it's an interior one. The market converged on exactly what the replicating research predicts.

Third: annual is the default two-thirds of the time, framed as percent-off.

Fourth — the surprise: pressure is rare at the top. Countdown urgency, 5%. Clear decoys, 5%. Strikethroughs, 12%. Elite pricing pages sell with structure and defaults, not pressure.

Teardown: rebuilding Quill's pricing page

Let's rebuild one. Quill is a fictional docs-collaboration SaaS whose "before" page commits only sins we actually found in the sample. Explore the interactive before/after: demo.html.

Structure. Before: four cards, identical visual weight, listed backwards — most expensive first, free plan last — and nothing highlighted. Every tier screams equally, so nothing is safe to choose. After: three tiers plus "Contact sales." The premium tier stays — it's not there to sell; it's there to make the middle safe. The middle tier gets the highlight and the "Most popular" badge, where 83% of top pages in our sample put it.

The anchor. Before: the page hides its expensive option in a footnote. After: "Contact sales" sits visibly on the right, and the strikethrough on the annual price — $180 crossed out, $144 — is real math from real monthly billing, not a fake was-price. True anchors only: reference prices work even when implausible, which is exactly why faking them is both tempting and policed.

The default. Before: monthly-first, annual hidden in fine print. After: the toggle defaults to annual — like two-thirds of top pages — framed as "Save 20%," with the monthly equivalent shown and the true billed total. Legal, honest, and the norm.

The digits. Before: $23.99 / $47.99 / $95.99 — deal-whisper pricing on a product pitching itself to design teams. After: $12 / $24 / $48 per seat. Round, confident, computable — you can do the seat math in your head, and the doubling pattern makes the ladder legible. If your brand sells premium, price like it.

The pressure. Before: a countdown timer on a SaaS subscription — "offer ends in 4:59" — and a vague "5x more usage" as the only differentiation. After: the timer is deleted, and each tier gets one line of who it's for — "For personal projects," "For growing teams," "For organizations." Audience labels do what fake urgency can't: they let the buyer self-select into the right comparison.

Final tally: anchors set honestly, 3; pressure tactics, 0. Nothing on the after page is a trick — every element just chooses the comparison the buyer was going to make anyway.

Set the comparison

The $39 dress outsold the $34 dress because nobody computes prices — we compare them. Your pricing page can't opt out of that. The only choice is whether you set the comparisons deliberately or by accident.

So: anchor high and honestly. Make the middle safe — that's the effect that replicates. Match your digits to your signal: charm for deals, round for premium. And skip the countdown timer — the best companies in software already have.

Tonight, open your pricing page and ask one question: what is my most expensive option doing for my middle one? If the answer is "nothing" — now you know what to build.

Sources

• Anderson & Simester (2003), "Effects of $9 Price Endings on Retail Sales," Quantitative Marketing and Economics 1(1) — PDF
• Northcraft & Neale (1987), "Experts, Amateurs, and Real Estate," OBHDP 39 — PDF
• Klein et al. (2014), Many Labs 1, Social Psychology 45 — PDF
• Ariely, Loewenstein & Prelec (2003), "Coherent Arbitrariness," QJE 118(1) — PDF
• Suk, Lee & Lichtenstein (2012), Journal of Marketing Research 49(5), 708–717
• Urbany, Bearden & Weilbaker (1988), Journal of Consumer Research 15(1) — abstract
• Ariely (2008), Predictably Irrational, ch. 1 (the Economist classroom demo)
• Frederick, Lee & Baskin (2014), "The Limits of Attraction," JMR 51(4) — PDF
• Huber, Payne & Puto (2014), "Let's Be Honest About the Attraction Effect," JMR 51(4) — PDF
• Simonson & Tversky (1992), JMR 29(3) — PDF
• Neumann, Böckenholt & Sinha (2016), meta-analysis, Journal of Consumer Psychology 26(2) — DOI
• Thomas & Morwitz (2005), "Penny Wise and Pound Foolish," JCR 32(1) — archived PDF
• Strulov-Shlain (2023), "More Than a Penny's Worth," Review of Economic Studies 90(5) — abstract
• Wadhwa & Zhang (2015), JCR 41(5) — PDF
• Pricing-page audit: our sample of 60 web pricing pages from top SaaS companies, collected via Mobbin, August 2026

Evidence types: peer-reviewed lab and field studies throughout, except the Anderson & Simester and Suk et al. results (randomized field experiments), the Ariely Economist numbers (a classroom survey, not a field test), and the 60-page audit (our own curated sample, not a random draw of the industry).