Syntha Lab RU

Price and purchase psychology: what to test and how to measure the effect

The document covers techniques of presenting price and purchase (anchor, price endings, discount framing, price ladder, the Veblen effect, drops, post-purchase dissonance, the fitting room) and shows which of them can be tested with an A/B test. It decides which technique to switch on and whether it is worth the effort.

Brand, customer, organisation Brand and retail Verified: 2026-11-14

Open interactively →

Why this document

Techniques such as an anchor or a price ending in 99 do not work everywhere: in the premium segment they can cheapen the brand, in the mass market they can lift conversion. The marketing director, the pricing manager and the analyst read this document when they choose what to implement and measure the effect in roubles. It separates what can be checked with an A/B test from what remains a hypothesis. Without it prices are dressed up by fashion, and nobody knows whether this added revenue or ate margin.

At a glance

What is this document and why is it needed?
The document covers techniques of presenting price and purchase (anchor, price endings, discount framing, price ladder, the Veblen effect, drops, post-purchase dissonance, the fitting room) and shows which of them can be tested with an A/B test. It decides which technique to switch on and whether it is worth the effort.
Who owns the document and how often is it rebuilt?
Brand: pricing director or brand manager with an analyst, tests by the season plan and a quarterly review; the commercial director decides on price. Retailer: category manager and marketing, tests in selected stores and online.
What mistakes are the most common?
The test is stopped at the first “winner” → false conclusions → fix the sample size in advance and do not peek. Conversion is watched instead of revenue per visitor and returns → orders grow but not profit → the main metric RPV and a returns guardrail. Several factors are changed at once → unclear what worked → one factor per test.

For whom

Brand
A brand manages position: too frequent promotions destroy value, and an inflated price without meaning stops the purchase. Tests are run in the own web shop and retail outlets, while wholesale sets a recommended retail price (within legal limits). The brand must check that a psychological technique does not contradict the brand platform.
Retail
A retailer sees price next to competitors and manages presentation: price tags, zones, bundles, promo. A retailer tests faster (many stores and sessions) but is limited by supplier price policies; some techniques work through assortment (good-better-best) and zoning.

What it consists of

Techniques and what practice says

  • Brand: the anchor is a more expensive line or an image model of the brand next to which the core item looks more accessible; line roles are set by the brand platform.
  • Retailer: the anchor is a more expensive brand on the shelf or a ladder step between brands in the category; shelf layout and the order in the site listing create the anchor by themselves.
  • Anchor: a more expensive item or the old price is shown next to the target item; test: the share choosing the target and the average basket.
  • Price endings (990, 999, 1000): perception differs by segment; one cannot assume in advance that nines “work”; test: conversion and revenue per visitor.
  • Brand: the discount presentation in D2C is aligned with partner prices: a brand discount deeper than the partner one breaks parity; the was/now is built on the brand own previous price.
  • Retailer: the discount presentation depends on the chain mechanics (loyalty card, coupon, category cross-discount); part of the discount may be financed by the brand under the contract, which is recorded before the promo starts.
  • Discount framing: in roubles or percent, was/now, free delivery instead of a discount; test: conversion, basket and return share.
  • The good-better-best price ladder: three steps, the middle is often chosen more; test: the sales distribution by step and margin.

Status, scarcity and emotion

  • Brand: The Veblen effect: for status goods demand does not fall as price rises; it is not universal and is checked only on several prices for a customer group.
  • Brand: FOMO and drops: a limited offer and term; the risk is false timers and pressure, so the limit must be real.
  • Post-purchase dissonance: doubt after payment; reduced by confirming the choice, a clear return and care tips.
  • Retailer: The fitting room: the decision moment; lighting, mirror, staff help and the number of items matter; measure the fitting-to-purchase conversion.

Test plan and effect calculation

  • Hypothesis and metric: what changes, by how much, the main metric (often revenue per visitor, not conversion alone).
  • Brand: a test on the brand own site is simpler because traffic is split randomly by user; the result is passed to partners as a recommendation for price and presentation, not as a mandatory rule.
  • Retailer: in a chain the test runs by store pairs, regions and periods (pairs matched by traffic and basket), online by user; days with different promo activity must be excluded.
  • Sample size: computed before launch from the minimum detectable effect; the test is not stopped at the first “winner”.
  • Brand: test metrics include the full-price share and the effect on wholesale partners: a brand promo in D2C must not run ahead of partner prices, otherwise parity and wholesale relations break.
  • Retailer: test metrics include margin after returns and cannibalization of neighboring brands in the category; a discount financed by the brand under the contract is counted separately from the own one.
  • Result: difference in the metric, a confidence interval or p-value, effect on returns and margin; the decision is recorded.

How to read the result and which decisions follow

  • For women z = 1.54, p = 0.12: on one category the difference is not significant, although conversion A is higher. The total over all categories has z = 2.03, p = 0.042: significant, but the effect is small (0.24 pp). Decision: pricing ending decisions are taken on the total and on money, not on a single category.
  • Brand: the sensitivity table shows that with a 20% smaller difference the total loses significance (z = 1.63), and with a 20% smaller sample z = 1.82. Decision: a test is not stopped at the first significant week; fix the sample size in advance.
  • Retailer: to detect a difference of 0.3 pp 58050 visitors per variant are needed, for 0.1 pp already 522449. Decision: in a chain with low traffic per category test on a pooled sample or on larger effects (discount, price format), not on pennies.
  • Brand: price psychology (endings, anchor, discount frame) affects conversion and trust; check not only orders but the full-price share and returns so that the effect does not come back through markdowns.
  • Retailer: the shelf price and the app price must match; a test with different prices in one channel can be run only under platform rules and the advertising law, and a price difference for one client may be inadmissible (unverified).

Formula derivation and sensitivity tables

  • Derivation of the z statistic. Variant conversions p_A = o_A / n and p_B = o_B / n. Under the null hypothesis both equal the pooled share p = (o_A + o_B) / (2n), the variance of the difference is p(1 - p) x (1/n + 1/n). Then z = (p_A - p_B) / √(p(1 - p) x 2/n).
  • Derivation of the sample size. We require the difference d to exceed the significance and power thresholds: d = (z_a + z_b) x √(2p(1 - p)/n). Solving for n gives n = 2 x (z_a + z_b)² x p(1 - p) / d², with z_a = 1.96, z_b = 0.84.
  • Numbers: at p = 3.45%: for d = 0.1 pp 522449 are needed, 0.2 pp 130613, 0.3 pp 58050, 0.5 pp 20898 visitors per variant (code).
  • Sensitivity of the total z (rows visitor count -20..+20%, columns conversion difference -20..+20%): -20%: 1.45 / 1.64 / 1.82 / 2 / 2.18; -10%: 1.54 / 1.74 / 1.93 / 2.12 / 2.31; 0%: 1.63 / 1.83 / 2.03 / 2.24 / 2.44; +10%: 1.71 / 1.92 / 2.13 / 2.35 / 2.56; +20%: 1.78 / 2 / 2.23 / 2.45 / 2.67.
  • Retailer: the significance threshold sets the probability of a false finding in a series of tests: with 5 categories and p < 0.05 in each the probability of at least one random finding is about 23% (1 - 0.95^5); apply a correction or test the total.

Control checks

  • Check: the total conversion of A 3.57% and B 3.33% is computed from sums of orders and visitors (1714 and 1599 of 48000), not as an average of categories.
  • Brand: show test variants to randomly chosen visitors of one period; seasonal days and promotions outside the test distort the comparison.
  • Retailer: before the test set the minimum detectable difference and duration; with significance across several categories remember multiple comparisons (the more categories, the higher the chance of a random success).
  • Brand: compute RPV at the variant price: conversion A is higher but the price is RUB 10 lower, and the difference in money may be smaller than expected.
  • Retailer: use buyer data in anonymized form; under the personal data law a test must not identify the client.

Where the data comes from

Data sourceFields
Web analytics and A/B platformVisitors by variant, orders, revenue, basket, add-to-cart, bounces.
Store POS dataReceipts of test and control stores before and after, average basket, UPT, discounts.
Returns recordsReturn share by variants; without it the real revenue cannot be assessed.

How to build it step by step

  1. Step 1. Choose one technique and state the hypothesis in numbers (“ending 990 instead of 1000 will raise revenue per visitor by at least 2%”).
  2. Step 2. Define the main metric and guardrails: revenue per visitor, margin, returns; the technique must not worsen returns.
  3. Step 3. Compute the sample size from the minimum effect and base conversion; estimate how many days are needed.
  4. Retailer: Step 4. Split traffic randomly and evenly; in stores choose paired outlets and a period; exclude special event days.
  5. Step 5. Run the test for full weeks (at least one to two weekday cycles) and do not change conditions during it.
  6. Step 6. Compute the difference in the metric and the p-value (or confidence interval); a result that does not reach significance means “not proven”, not “no effect”.
  7. Retailer: Step 7. Check the guardrails: returns, margin, effect on neighboring items (cannibalization), customer segments.
  8. Step 8. Estimate the effect in roubles per period with buyout and margin; decide: roll out, repeat, drop.
  9. Step 9. Record the result in a test register (hypothesis, date, size, outcome) so tests are not repeated and the company does not contradict itself.

Formulas

MeasureFormulasNote
Revenue per visitorRPV = Conversion × Price (or average basket)Compare variants by RPV: a higher price may lower conversion but raise revenue.
Z statistic for proportionsz = (pA − pB) ÷ √(p(1−p) × (1/nA + 1/nB)), where p is the pooled share|z| > 1.96 gives p < 0.05 for a two-sided test; multiple checks require a correction.
Sample size per variantn ≈ 2 × (1.96 + 0.84)² × p(1−p) ÷ d²d is the smallest difference of shares one wants to detect; 5% significance, 80% power.
Effect in roublesEffect = ΔRPV × Visitors per period × Buyout % × Margin %Applied only to statistically confirmed effects.

Formulas in notation

z statistic
z = (p_A − p_B) / √(p(1 − p) × (1/n_A + 1/n_B))
p is the pooled conversion share, n_A and n_B the visitors of the variants.
Effect over the period
Effect = ΔRPV × Visitors × Buyout%
Only for a significant difference and with buyout taken into account.
p-value
p = 2 × (1 − Φ(|z|))
Φ is the standard normal distribution function.

Worked example

Exercise: price ending test across five categories, variant A (990) versus B (round price), conditional numbers

CategoryPrices A / B, RUBVisitors per variantOrders AOrders BConversion A, %Conversion B, %Difference, ppzpRevenue per visitor A vs B, %
Women4,990 / 5,00018,0006125603.43.110.291.540.129.1
Men5,490 / 5,5009,0002882793.23.10.10.380.73
Kids2,490 / 2,5007,0003012874.34.10.20.590.564.5
Footwear6,990 / 7,0008,00024023232.90.10.370.713.3
Accessories1,990 / 2,0006,0002732414.554.020.531.440.1512.7
Total (conversion)48,0001,7141,5993.573.330.242.030.04

In none of the 5 categories is the difference significant at the 5% level (significant: 0), but the pooled sample gives p = 0.042: on small per-category samples the test is underpowered rather than showing no effect. Accessories gave p = 0.149, women 0.12. The total row tests conversion only and is not a revenue total because category prices differ. Revenue per visitor is computed within a category. One ending test is not carried over to other categories. All numbers are conditional.

Figures

Conversion of variants A and B by category
Bar chart: Women 3.4, Men 3.2, Kids 4.3, Footwear 3, Accessories 4.55012345WomenWomenA: ending 990: 3.43.4B: round price: 3.113.11MenMenA: ending 990: 3.23.2B: round price: 3.13.1KidsKidsA: ending 990: 4.34.3B: round price: 4.14.1FootwearFootwearA: ending 990: 33B: round price: 2.92.9AccessoriesAccessoriesA: ending 990: 4.554.55B: round price: 4.024.02%
  • A: ending 990
  • B: round price
Data as a table
MeasureA: ending 990B: round price
Women3.43.11
Men3.23.1
Kids4.34.1
Footwear32.9
Accessories4.554.02

Conversion of variants A and B by category. A price ending in 990 (A) against a round one (B): conversion A is higher in all categories, but significance differs.

Sample size and minimum detectable difference
Line chart: Visitors per variant: 522k → 14.5k0100k200k300k400k500k600k0.10.20.30.40.50.6our MDEVisitors per variant, 0.1: 522kVisitors per variant, 0.2: 131kVisitors per variant, 0.3: 58.1kVisitors per variant, 0.4: 32.7kVisitors per variant, 0.5: 20.9kVisitors per variant, 0.6: 14.5kvisitors
Data as a table
MeasureVisitors per variant
0.1522k
0.2131k
0.358.1k
0.432.7k
0.520.9k
0.614.5k

Sample size and minimum detectable difference. How many visitors per variant are needed to see a conversion difference (horizontal axis: difference in pp): the smaller the difference, the faster the sample grows. Power 80%, significance 5%.

Total test significance as the difference and sample shift
Heat map: 5×5; min 1.45, max 2.67Difference -20%Differe…-20%Difference -10%Differe…-10%Difference 0%Differe…0%Difference +10%Differe…+10%Difference +20%Differe…+20%Visitors -20%Visitors -20%1.45Visitors -20% / Difference -20%: 1.451.64Visitors -20% / Difference -10%: 1.641.82Visitors -20% / Difference 0%: 1.822Visitors -20% / Difference +10%: 22.18Visitors -20% / Difference +20%: 2.18Visitors -10%Visitors -10%1.54Visitors -10% / Difference -20%: 1.541.74Visitors -10% / Difference -10%: 1.741.93Visitors -10% / Difference 0%: 1.932.12Visitors -10% / Difference +10%: 2.122.31Visitors -10% / Difference +20%: 2.31Visitors 0%Visitors 0%1.63Visitors 0% / Difference -20%: 1.631.83Visitors 0% / Difference -10%: 1.832.03Visitors 0% / Difference 0%: 2.032.24Visitors 0% / Difference +10%: 2.242.44Visitors 0% / Difference +20%: 2.44Visitors +10%Visitors +10%1.71Visitors +10% / Difference -20%: 1.711.92Visitors +10% / Difference -10%: 1.922.13Visitors +10% / Difference 0%: 2.132.35Visitors +10% / Difference +10%: 2.352.56Visitors +10% / Difference +20%: 2.56Visitors +20%Visitors +20%1.78Visitors +20% / Difference -20%: 1.782Visitors +20% / Difference -10%: 22.23Visitors +20% / Difference 0%: 2.232.45Visitors +20% / Difference +10%: 2.452.67Visitors +20% / Difference +20%: 2.67
  • 1.45 … 0 … 2.67 z
Data as a table
Difference -20%Difference -10%Difference 0%Difference +10%Difference +20%
Visitors -20%1.451.641.8222.18
Visitors -10%1.541.741.932.122.31
Visitors 0%1.631.832.032.242.44
Visitors +10%1.711.922.132.352.56
Visitors +20%1.7822.232.452.67

Total test significance as the difference and sample shift. The z statistic of the total when the conversion difference and visitor count change by ±10% and ±20%: values below 1.96 are not significant.

Specifics by category

Apparel
Endings and anchor are tested by product type: in core the customer knows the price from analogs (T-shirt, jeans), so price sensitivity is higher, while in fashion price is judged by novelty and look. The good-better-best ladder works for types with a clear quality difference (coat, knitwear); the step difference is checked by sales distribution and margin. The was/now presentation rests on a real previous price, otherwise the law is breached.
Footwear
The price per pair is high, so thresholds (round sums such as 5,000 or 7,000) are tested as a hypothesis, without a predetermined “magic of digits”. The main conversion factor is confidence in size, not price: measure try-on-to-purchase conversion in store and the effect of the size hint on the card. Free return raises conversion but worsens margin, so the metric is revenue per visitor net of returns. A discount on one size is compared with a discount on the whole model.
Accessories
There are many impulse and gift purchases: checkout anchor, a “bag plus belt” set and the free delivery threshold are tested well. Gift thresholds (for example RUB 1,000, 3,000, 5,000) are set as a hypothesis and tested, not taken as a norm. For status bags a Veblen effect is possible: a price cut can harm perception, so it is tested on several prices and a limited group rather than rolled out to the whole catalog. Size does not interfere, so the test is cleaner than in apparel and footwear.
Kids
The buyer is the parent and the logic is price per wearing period and safety, so the “price per month” presentation and sets are tested separately. False timers, “only 2 left” pressure and artificial scarcity are especially risky in kids: the limit must be real, and school uniform is not suited to drops at all. 10% VAT applies only to goods on the list (decree no. 908, OKPD2 code): the shelf price is one, and revenue and margin are compared excluding VAT. Seasonal peaks distort a test, so compare periods like for like.
Men
Discount sensitivity in men is lower (a working hypothesis to check on own data), and core repeats, so price stability is worth more than frequent promos: frequent promos train customers to wait and destroy the full-price share. Sets and the free delivery threshold test well, rather than discount depth. There are fewer fittings and decisions are faster, so offline the availability of the needed size matters more than assistant help. Main metric: revenue per visitor and the repeat share over 6 months.
Women
Discount sensitivity is higher, the share of purchases on discount is higher and the was/now presentation works stronger, but frequent promos train customers to wait for markdowns: test the discount depth (10, 20, 30%) and measure the full-price share, not only week revenue. There is more post-purchase dissonance because of size and color: confirming the choice and a clear return reduce it. The fitting room works through the number of items and assistant help; the metric is margin after returns, not conversion alone.

Common mistakes

  • The test is stopped at the first “winner” → false conclusions → fix the sample size in advance and do not peek.
  • Conversion is watched instead of revenue per visitor and returns → orders grow but not profit → the main metric RPV and a returns guardrail.
  • Several factors are changed at once → unclear what worked → one factor per test.
  • False timers and “only 2 left” without real stock → lost trust and complaints → the limit must match warehouse data.
  • A result from one category is carried over to all → a wrong policy → repeat the test by segment and category.

Russian specifics

  • Retailer: Price and discount information must be accurate: the price on the tag and on the receipt is the same, “was/now” rests on a real previous price; inaccurate information violates consumer rights and the advertising law. Rule: Law 2300-1 of 07.02.1992: Law of 07.02.1992 No. 2300-1 on consumer rights protection.
  • Advertising of discounts and “promotions” falls under the advertising law: terms are stated fully, deadlines are real. Rule: Law 38-FZ of 13.03.2006: Federal Law of 13.03.2006 No. 38-FZ on advertising.
  • Brand: A recommended retail price in a contract with retail is allowed, a mandatory fixed resale price is restricted by the competition law. Rule: Law 135-FZ of 26.07.2006: Federal Law of 26.07.2006 No. 135-FZ on protection of competition.

Open interactively →