Syntha Lab RU

Reader question: what price to set on a novelty when there is no elasticity

An answer to the question of what to do with the price of a model that has no sales history and so no elasticity can be estimated: how to find the lower bound from costs and margin, the upper bound from the market and willingness to pay, how to pick several price points for a test and how much traffic is needed to tell a real difference from noise. The result: a price corridor, two or three test points and a rule by which the winner is chosen.

Commerce and buying Brand and retail Verified: 2026-11-14

Open interactively →

Why this document

A novelty has no history and the price must be set before it goes on sale. Then either the price “as for the neighboring model” is taken, or a markup on costs is set and nobody checks whether the market is ready to pay. Both ways are blind. The category manager, the head of e-commerce and the CFO read this document when a model launch is prepared. It gives an order of reasoning: the lower bound, the upper bound, the corridor, a test and a rule of choice. The calculation shows by how much conversion may fall when the price rises so that profit does not fall, and how many observations are needed to see this difference.

At a glance

What is this document and why is it needed?
An answer to the question of what to do with the price of a model that has no sales history and so no elasticity can be estimated: how to find the lower bound from costs and margin, the upper bound from the market and willingness to pay, how to pick several price points for a test and how much traffic is needed to tell a real difference from noise. The result: a price corridor, two or three test points and a rule by which the winner is chosen.
Who owns the document and how often is it rebuilt?
The category manager and the head of e-commerce prepare, the CFO approves the lower bound; the decision is made before launch and reviewed after the test results.
What mistakes are the most common?
Brand: the price is set as for the neighboring model → willingness to pay is not checked → set a corridor and test two or three points. Retail: the winner is chosen by conversion → the price is understated and profit falls → choose by profit per visitor. Brand: the test goes in sequence (a week of one price, then another) → the season and promotions distort the comparison → split the traffic simultaneously.

For whom

Brand
A brand sets the recommended price of a model and the price for partners. The main things: the lower bound by margin including commissions, the corridor against competitors and an agreed procedure for changing the price after the test.
Retail
A retailer sets the price of a private label or an exclusive and tests it on part of the traffic or stores. The main things: a test on comparable stores and periods and not “before and after”, and a stopping rule.

What it consists of

Three anchors of the price

  • The lower bound: unit costs plus the minimum margin and channel commissions; below this price the model is not sold, even with record conversion.
  • The upper bound: the price at which the model drops out of its price segment; guides: the prices of similar models of competitors and of your own line (price steps), a quick willingness-to-pay survey or reviews of an analog model.
  • The corridor: between the lower and the upper bound; two or three points are chosen inside it for the test; one point almost always removes the possibility of comparison.

The price test

  • Brand: two or three price options on equal traffic groups (a random split of site visitors or stores equal in traffic), simultaneously and not in sequence; record the conversion into an order and the buyout.
  • Retail: count not conversion but profit per visitor: conversion × (price without VAT − costs); a choice by conversion pushes the price down.
  • The size of the test: to notice a fall of conversion by a given share the number of observations of the order of 16 × p × (1 − p) / δ² per arm is needed (δ: the noticeable difference); a small price step needs a great deal of traffic.

The choice rule

  • The winner is chosen by a rule set in advance: the maximum profit per visitor at an acceptable buyout and return rate; no decision is made on the first days, a full weekly cycle is needed.
  • Brand: if a test is impossible (small traffic), take the middle of the corridor, record the hypothesis and change the price on the results of the first weeks taking partners’ rules into account.

Where the data comes from

Data sourceFields
Costs and channelcost, logistics, commission, minimum margin, VAT rate
Market and lineprices of similar models, price steps of your own line, survey results
Trafficthe number of visitors per card, the base conversion of an analog model, buyout and return

How to build it step by step

  1. Step 1: Brand: count the lower bound: unit costs plus the minimum margin and commissions, then the shelf price with VAT.
  2. Step 2: Find the upper bound from similar competitor models, your own line and a survey.
  3. Step 3: Choose two or three prices in the corridor with a step of at least 8–10%; write them into the test.
  4. Step 4: Count by how much conversion may fall at each price so that profit per visitor does not fall (see the example).
  5. Step 5: Retail: estimate the number of observations per arm and check it against traffic; if there is little, enlarge the price step or the time.
  6. Step 6: Set the choice rule and the test time in advance: profit per visitor, minimum weeks, acceptable buyout.
  7. Step 7: Retail: run the test simultaneously on comparable stores and traffic groups, do not change other conditions.
  8. Step 8: Brand: apply the winner, agree it with partners and record the result as experience for the next models.

Formulas

MeasureFormulasNote
Profit per visitorP = c × (N − C)c: conversion into an order (net of buyout), N: price without VAT, C: unit costs.
Permissible fall of conversionc₂ / c₁ ≥ (N₁ − C) / (N₂ − C)At a higher price conversion may fall no lower than this ratio, otherwise profit per visitor is lower.
Observations per armn ≈ 16 × p × (1 − p) / δ²p: the base conversion, δ: the absolute difference one wants to notice; an approximate formula (power about 80%, significance level 5%).

Formulas in notation

Permissible fall of conversion
c₂/c₁ ≥ (N₁ − C)/(N₂ − C), N = P/(1 + t)
P: the shelf price; t: VAT; C: unit costs; c₁, c₂: conversions at prices 1 and 2.
Sample size
n ≈ 16·p·(1 − p) / δ², δ = p·(1 − c₂/c₁)
p: the base conversion; δ: the absolute difference. An approximate rule.

Worked example

Four shelf prices with 22% VAT; unit costs RUB 1,400 without VAT; the base conversion 4%; all values conditional

Shelf price, RUBProfit per unit, RUBConversion may fall to, % of the baseObservations per arm to check
3,5001,469100.0—
3,9001,79781.811,532
4,3002,12569.14,031
4,7002,45259.92,387

At a price of 3,900 instead of 3,500 conversion may fall by about 18% and profit per visitor will not decrease; but to notice such a fall at a base of 4% more than ten thousand observations per arm are needed: so for small price steps the test needs a lot of traffic.

Figures

Permissible fall of conversion as the price rises
Line chart: Required share of the base conversion: 100 → 59.950607080901003 5003 9004 3004 700Required share of the base conversion, 3 500: 100Required share of the base conversion, 3 900: 81.8Required share of the base conversion, 4 300: 69.1Required share of the base conversion, 4 700: 59.9% of the base conversionShelf price, RUB
Data as a table
Shelf price, RUBRequired share of the base conversion
3 500100
3 90081.8
4 30069.1
4 70059.9

Permissible fall of conversion as the price rises. The higher the price, the more conversion can fall without losing profit per visitor: the reserve grows, but the circle of buyers narrows.

How many observations are needed to notice the permissible fall of conversion
Bar chart: 3 900 11.5k, 4 300 4,031, 4 700 2,38702,0004,0006,0008,00010k12k3 9003 900Observations per arm: 11.5k11.5k4 3004 300Observations per arm: 4,0314,0314 7004 700Observations per arm: 2,3872,387observations
Data as a table
MeasureObservations per arm
3 90011.5k
4 3004,031
4 7002,387

How many observations are needed to notice the permissible fall of conversion. The smaller the price step, the smaller the expected fall and the more observations are needed.

Specifics by category

Apparel
Apparel: the buyer compares by brand and style, price steps are wide; a price test is convenient on twin models, and size returns are included in profit per visitor.
Footwear
Footwear: the unit price is higher, traffic per model is lower, size returns are costlier; the test needs more time, so the middle of the corridor is more often chosen and adjusted by the first weeks.
Accessories
Accessories: a low unit price and impulse demand; the price step in percent can be smaller, but the guide is a psychological threshold (a round price) and accompanying purchases.
Kids
Kids: parents buy, the price is compared with the expected “short” life of the item; a reduced VAT rate is possible by a list and changes the calculation, and discounts in the school season are expected.
Men
Men: demand is less sensitive to small changes in price and more to brand and quality; steps are wider, the test can be run less often.
Women
Women: a wide choice and high sensitivity to promotions; the test is run outside sale periods and profit per visitor is counted with returns.

Common mistakes

  • Brand: the price is set as for the neighboring model → willingness to pay is not checked → set a corridor and test two or three points.
  • Retail: the winner is chosen by conversion → the price is understated and profit falls → choose by profit per visitor.
  • Brand: the test goes in sequence (a week of one price, then another) → the season and promotions distort the comparison → split the traffic simultaneously.
  • Retail: the test is stopped on the first days → a random spike is taken for an effect → set the term and the number of observations in advance.
  • Retail: a price step of 2–3% → the difference cannot be seen without tens of thousands of observations → make the step at least 8–10% or extend the test.

Russian specifics

  • Brand: VAT 22% from 01.01.2026 (425-FZ), for children’s goods 10% under the conditions of the law and confirmation in the certificate data (218-FZ from 01.10.2026): check the code and conditions when counting the price. Rule: Law 425-FZ of 28.11.2025: Federal Law of 28.11.2025 No. 425-FZ (amendments to parts one and two of the Tax Code; VAT 22%).
  • Retail: promotions and discounts in the test are subject to advertising and consumer rules; the price on the tag must be the one at which the goods are sold: check the consumer protection law. Rule: Law 2300-1 of 07.02.1992: Law of 07.02.1992 No. 2300-1 on consumer rights protection.
  • Brand: different prices for different groups of visitors for the same goods on a site must be compatible with the rules on competition and consumer protection: check with a lawyer before launching the test.
  • Retail: the sample size formula is approximate; for decisions on large batches use an exact test power calculation.

Open interactively →