A one-cent change looks like a small decision. Across a large catalogue, it changes both the price paid and the signal sent. A $39 field result, a $2.99 perception experiment and a proposed €19.99 product-page test are different comparisons.
Use price endings as a testable part of the offer. Decide whether the relevant change is a digit threshold, product familiarity, a Sale cue or actual buying behavior.
🔎 What research says
The evidence supports conditional effects. A 2024 meta-analysis combined 69 studies with 40,541 participants. Its average purchase-decision effect was 0.13 standard deviations, a small difference relative to the variation within the studies. The 95% confidence interval ran from 0.01 to 0.25 standard deviations. Results varied widely, and publication bias may make the average look larger than it is. This standardized comparison combines different purchase measures; it cannot be read as 13% more sales. Quality perception showed no clear average difference. Historical field gains coexist with an incentivized online null.
🗺️ In this guide
1. Test the digit change, not just the final 9
📈 Recommendation
Define which threshold the proposed price actually crosses. A final 9 can change the leading digit, the amount saved and the message of a price at the same time.
For a €20 product, compare €19.99 with €20.00 while keeping the page and terms fixed. Treat this as a proposed store test, not a replication of the pen task.
Record paid orders and contribution, as well as perceived price. A lower price rating cannot pay the advertising bill.
🎓 Findings
Thomas and Morwitz (2005) asked 52 students to judge pen prices against a $4 comparison. On a 1–5 “price is high” scale, a $2.99 ballpoint scored 2.07 versus 2.76 at $3.00: 0.69 points lower. For a fountain pen at $3.59 versus $3.60, the corresponding scores were 2.61 versus 2.65, with no clear difference.
The difference between these two price comparisons was statistically clear. The paper reports p < .05, its test threshold; this is not a measure of the sales benefit. These were price perceptions, not purchases. The first study also confounded product and distance from the comparison price; the paper’s subsequent studies addressed those issues.
🧠 Why it works
Reading a price from left to right may make a changed leading digit especially noticeable. The experiments support a conditional perception mechanism.
✋ Limitations
The small student task used cheap writing instruments and no purchase requirement. It provides no percentage sales lift or universal threshold rule.
2. Separate new products from established products
📈 Recommendation
Test price endings separately for new and familiar ranges. Do not let an overall average hide where customers already know the usual price.
Define newness before the test, using product launch or prior availability rather than the eventual result.
For a new apparel range, compare kept units and contribution with an established range. Keep the price change visible in the margin calculation.
🎓 Findings
Anderson and Simester (2003) randomly mailed two women’s clothing catalogue versions to 31,250 customers each in Study 2. Their model adjusted for item and catalogue version and used 422 item-by-version sales observations. The authors estimated that a whole-dollar $9 ending added about 1.3 units per item, against an average of 8.7, or roughly 15%.
The model estimate was about 22% for new items versus 10% for established items. The new-item coefficient was clear; the established-item and footwear coefficients were not. The paper does not establish a statistically clear difference between the new- and established-item estimates in that table.
🧠 Why it works
When shoppers have less price history for an item, an ending may serve as a value cue. This is the authors’ explanation, not a directly measured belief in the field trial.
✋ Limitations
These were historical mail-order customers and whole-dollar prices such as $39, not a test of €39.99 versus €40. The estimates concern catalogue demand per item, net of returns and cancellations, not website visitor conversion or a forecast for another store.
3. Test sale labels and endings together
📈 Recommendation
Treat the sale label and the price ending as a joint merchandising decision. Do not assume that two individually useful cues produce the sum of their effects.
Where a genuine promotion exists, test ending and Sale label in separate randomized combinations. Keep the actual offer and comparison-price basis accurate.
Use incremental kept-order contribution to choose the combination. Include the cost of any reduction in the selling price.
🎓 Findings
Study 3 in Anderson and Simester (2003) mailed three catalogue versions to 90,000 customers each and modeled 924 item-by-version observations. For new items, the authors’ model implied an additional 3.9% demand from adding a $9 ending when a Sale cue was already present, versus 8.5% without that cue.
The authors explicitly call this comparison tentative because relatively few items combined the Sale cue and $9 ending. It is not a secure estimate of a 4.6-percentage-point loss, and it does not show that stacking the cues harms demand.
🧠 Why it works
A Sale label can already communicate a bargain, leaving less additional information for the ending to supply. That interpretation remains tentative here.
✋ Limitations
The reported percentages are the authors’ differences between model estimates, not raw order rates or a general interaction effect. The small combination cell and historical catalogue setting limit transfer.
4. Require buying evidence before rolling an ending out
📈 Recommendation
Choose a price ending from a planned commercial test, with a decision rule fixed before looking at results. Measure behavior close to the decision you need to make.
For a product-page test, use a stable customer assignment and a fixed test period. Track paid and kept orders, returns, revenue and contribution per assigned visitor.
An illustrative margin check: if contribution is €8 at €20.00 and all other costs stay fixed, €19.99 leaves €7.99. It would require about 0.13% more kept units just to preserve contribution. This is arithmetic under stated assumptions, not an estimated research effect.
🎓 Findings
Fenneman and colleagues (2022) studied 266 valid online participants, each making 18 lottery-ticket decisions: 4,788 repeated decisions, not 4,788 independent shoppers. One randomly selected decision determined each person’s bonus payment. Ticket payoffs were adjusted with the price so the economic attractiveness stayed comparable.
Purchase rates were 47.1% for just-below, 45.9% for round, and 45.3% for other precise prices. The 1.2-percentage-point gap between the first two is descriptive. Neither just-below nor round prices had a statistically clear advantage over the precise-price control in the main model. The predicted stronger just-below effect at the lower price level was also not supported.
🧠 Why it works
Attention to a digit need not become a purchase. An ending earns a rollout only when any behavioral gain covers its actual commercial cost.
✋ Limitations
Lottery decisions are not ordinary retail baskets. Of 390 completers, 124 were excluded for buying dominated attention-check tickets; including them did not change the main conclusion. Null results do not prove that every retail ending has zero effect.
💡 The takeaway
A price ending is a commercial hypothesis, not a growth constant.
Choose the exact comparison, define the segment and measure kept-order contribution. Keep a round price when the tested alternative has no worthwhile advantage.
📚 Sources and study types
Thomas, M., & Morwitz, V. (2005). Penny Wise and Pound Foolish: The Left-Digit Effect in Price Cognition. Journal of Consumer Research, 32(1), 54–64. Five experiments; final author-upload text reviewed. Price perception is not buying.
Anderson, E. T., & Simester, D. I. (2003). Effects of $9 Price Endings on Retail Sales: Evidence from Field Experiments. Quantitative Marketing and Economics, 1, 93–110. Randomized catalogue field experiments; final PDF tables read visually. Whole-dollar endings and model-adjusted units.
Fenneman, A., Sickmann, J., Füllbrunn, S., Goldbach, C., & Pitz, T. (2022). Psychological price perception may exert a weaker effect on purchasing decisions than previously suggested: Results from a large online experiment fail to reproduce either a left-digit or perceptual-fluency effect. PLOS ONE, 17(8), e0270850. Preregistered incentivized lottery task; primary PDF reviewed.
Troll, E. S., Frankenbach, J., Friese, M., & Loschelder, D. D. (2024; online 2023). A meta-analysis on the effects of just-below versus round prices. Journal of Consumer Psychology, 34(2), 299–325. Relevant primary publisher full-text results reviewed; original PDF not recovered. Synthesis is not independent replication.
