Charm pricing and price presentation: .99 endings, round prices and currency symbols

Note: This text is not legal advice. Have your specific implementation reviewed by a lawyer.

€19.99 or €20?

Hardly any pricing decision is made as often as this one, and hardly any is taken for granted as readily: .99 sells better, end of story.

The research says something more careful. The effect exists, but it is small, erratic, and tied to conditions a pricing page often fails to meet. In some situations the round price sells better.

Why 2.99 looks smaller than 3.00

In the left-digit effect we give the frontmost digit of a price too much weight. 2.99 looks cheaper than 3.00 because at the front a 2 becomes a 3 — the single cent in between barely registers.

Thomas and Morwitz (2005) tested this in five experiments. Their result is narrower than the short versions suggest: two conditions must hold at the same time.

First, the left digit has to change. Participants rated on a scale how large a price seemed to them — the lower the value, the cheaper the impression.

For 2.80 against 2.79 and for 3.20 against 3.19, nothing happened either. So a cent less helps only when it puts a different digit at the front.

The “p < 0.01” means: a difference of this size would arise from chance alone in fewer than one case in a hundred. A finding is conventionally called statistically meaningful from p < 0.05.

Second, a comparison price has to be nearby. This condition is missing from almost every short version, and it is the more consequential one. For 3.99 versus 4.00 the effect appeared only when the comparison price was one dollar away, at $3 or $5: 2.50 versus 3.39 (p < 0.01). When it was two dollars away, at $2 or $6, it had vanished entirely: 2.96 versus 3.09.

The digit change is therefore necessary but not sufficient. Without a nearby comparison, nothing happens.

Mere habituation to shop prices does not explain this: the same patterns appeared with quality ratings and with bare numbers carrying no product context at all.

What does explain it is open — the two papers that test the mechanism directly disagree.

One says: we cut off the trailing digits. Bizer and Schindler (2005) had 122 students estimate how many items at a given price they could buy for $73. For 99 endings the estimates came out 4.5 percent higher although the prices differed by only 0.2 percent (p < .001). Arithmetic errors of the kind produced by dropping the right-hand digits were more frequent: 12.1 against 7.6 percent.

What matters is for whom. The effect was meaningful only among participants who were neither instructed nor inclined to calculate carefully. Those who thought it through showed it only faintly (p = .11, above the conventional threshold). A condition that decides whether an effect appears at all is called a moderator.

The other says: the 9 is a signal. Anderson and Simester (2003) tested the same cutting-off account against real sales data and rejected it: left and right digits carried equal weight (−0.0631 versus −0.0651, the difference is not meaningful). They find it more plausible that the 9 simply tells customers this price is low. Why the 9 of all digits took on that role, they write, remains unexplained.

How strong is the effect really?

The best overall picture available is a preregistered meta-analysis by Troll, Frankenbach, Friese and Loschelder (2024): 69 studies, 362 effect sizes, 40,541 participants.

Two terms, because they are what makes this work valuable. Preregistered means the question and the analysis plan were fixed publicly before the data were in — which stops the calculation that looks best from being chosen at the end. The g is a measure of how large an effect is: 0.2 counts as small, 0.5 as medium, 0.8 as large.

The last line clears away a common worry: on this collection there is no quality penalty for just-below prices. The line above it is the actual purchase effect — and at 0.13 it sits below what already counts as a small effect. The interval gives the range in which the true value lies with 95 percent confidence; its lower end nearly touches zero.

What matters is when the purchase effect appeared at all.

The authors raise the obvious objection themselves: the effect might then be an artefact of the study design — the ending is noticed only because both variants sit side by side. Their counter-argument is that in a real shop just-below and round prices do sit side by side. For a pricing page where each plan carries exactly one price, that is the weaker case.

Then there is the question of how stable the figures are. Sixty percent of the 362 effect sizes were not meaningful to begin with. And after correcting for publication bias the purchase effect shrinks to g = −0.01 and is no longer detectable; an alternative multilevel model pushes it below the threshold too.

Publication bias means: studies with a striking result are published more readily than studies with no finding. Once you estimate how many null results probably stayed unpublished, nothing of this effect is left. The authors write that their estimates should be read as an upper bound on the true effects.

Two conditions are usable in practice. Price image benefits markedly more when the digit before the decimal changes (€79 instead of €80, g = 0.62) than when only the decimals are uneven (€79.95 instead of €80.00, g = 0.17). And this effect too disappeared entirely when no round comparison price was visible: g = 0.00 instead of g = 0.47.

In short: there are indications of an effect. It is small, heavily dependent on context — and it does not survive the correction for publication bias.

What field studies show

• Catalog field experiments: Anderson and Simester (2003) manipulated the price endings of two mail-order catalogs selling women’s clothing (private label, average price around $50). A .9 ending increased demand in all three studies, but by very different amounts: 35 percent with 20,000 recipients per catalog version, 15 percent with 31,250 and 7 percent with 90,000. In the largest study, significance also depended on whether price entered the model as its own variable. The ending worked more strongly for new items than for established ones (22 versus 10 percent) and not at all for footwear. A .50 ending did not work; if anything it reduced demand.
• Supermarket data: From scanner data covering ten categories across 83 stores over 399 weeks, Macé (2012) estimated 43,144 separate elasticities — one per item per store. An elasticity here is simply the answer to one question: how much do sales of this particular item in this particular store change when the price ends in 9 rather than not? The median effect is +10.5% in units sold. Nine-endings work mainly for small brands with low prices and for new items, provided the category is low-priced too.

The prior findings Macé assembles openly contradict each other.

The median hides the fact that usually nothing happens at all.

In nearly three cases out of four no effect is measurable. The +10.5% arises because the top quarter swings hard. And in roughly one estimate in twelve the effect goes the other way, by 11 percent on average.

Who loses can be described precisely. The items with a negative effect are premium brands long established in the range, with high market share and a high price. Macé explicitly advises their manufacturers to negotiate the ending with the retailer — it is usually a store decision, not a brand one.

The categories are worlds apart. For paper towels the median effect is 23.7 percent, for oatmeal 5.0 — and for crackers, at 0.02 percent, it is indistinguishable from zero. There, twice as many estimates are negative as positive (12.6 against 6.3 percent).

And the ending wears out. In stores where more than 70 percent of prices end in 9, sales of the affected items are 6 percent lower than in stores below 40 percent. Macé’s simulation shows where the optimum lies:

In practice an average of 53.6 percent of all prices end in 9. Three of the four simulated categories are therefore already at or past their optimum.

• Overuse devalues the ending: Both field sources converge on the same point. Anderson and Simester argue that a price signal loses its force when it is everywhere, much like a sale sign on every item, so retailers should use .9 endings selectively. Macé reports that the effect erodes the more intensively a store uses nine-endings, and that a simulation shows a threshold beyond which the practice becomes counterproductive. For a pricing page that means: .99 on every tariff is the weaker option, not the stronger one.
• Round prices and convenience: Wieseke, Kolberg and Schons (2016) present four studies, but only the first measures sales: 306 drink purchases in three university cafeterias, cash payment, impulse buying. There the round price (€1.00) reached the highest share of total sales, ahead of the just-below (€0.99) and the precise variant (€1.01); χ² = 6.78, p = .03. The other three studies are surveys of 3,013, 799 and 10,207 people and measure perception, not behaviour. In those, round prices count as more convenient throughout, across every price level tested from €1 to €100.
• More convenient is not more attractive: The fourth study worked out which intermediate steps the price acts through. Directly, the round price markedly lowers perceived price attractiveness (b = −0.52; the minus sign means it pushes it down). It turns positive only by way of convenience (indirect b = +0.26) — and only for people who value convenience. Those who care little about it find the round price less attractive. Precise prices such as €1.01 or €46.52 came last throughout.

The overall picture: there is no universal “.99 wins”. The more information customers have and the more important credibility or convenience is, the more likely the advantage flips.

Practical cases

Price presentation: what about the currency symbol?

You often read online that a missing € or $ sign lowers the “pain of paying”. The evidence is thin.

The best-known study is a field experiment in a single restaurant. Yang, Kimes and Sessarego (2009) randomly assigned one of three menus to 201 analysed parties: prices with a dollar sign, as a bare numeral, or spelled out. The overall effect of format was not significant (p > .144). One single contrast was: tables with the bare numeral spent $5.55 more than the two other formats combined (p < .05), which is 8.15 percent per person ($23.00 versus $24.87).

What stayed the same matters more. Between the dollar sign and the spelled-out currency there was no difference at all (p > .99). Both of the authors’ hypotheses failed: they had expected numerals to lower spending relative to spelled-out amounts, and the symbol to be decisive. Neither held. What worked was apparently not the sign but any recognisable reference to money. One restaurant at a culinary school, one lunch service, three months, assignment by table.

The “pain of paying” explanation comes from Prelec and Loewenstein (1998). That is a modelling paper: its own data are ratings of hypothetical payment schedules by 86 to 91 respondents at a conservatory, an airport and a university. Price presentation does not appear in it at all.

The brain findings also carry less than they are asked to. Knutson et al. (2007) are cited as showing that excessive prices activate the insula — that is what their summary says. In the discussion the authors themselves write that insula activation did not correlate significantly with the price differential (Z = −1.36), and that the specificity of this response remains to be clarified. Twenty-six analysed participants, prices discounted 75 percent off retail, 60 percent prediction accuracy against 50 percent chance. That a currency symbol triggers “pain” appears nowhere in it.

For online shops and pricing pages there is no verified study. If you try it, do it as an A/B test and without making prices harder to read.

Ethics check

Limits and criticism