Syntha Lab RU

Promotions: planning and evaluation

Whether a promotion paid back, how much of the uplift the post-promotion dip ate, what a control group shows and whether to repeat it.

both Price and discounts

Open interactively →

About the topic

A promotion almost always looks successful: weekly revenue is up, the warehouse is unloaded, there is a nice figure in the report. But revenue is not the measure of success. Some shoppers would have bought without the promotion. Some are taken from neighboring goods and from following weeks. And the discount, advertising of the promotion and higher returns go into costs. The topic teaches you to evaluate a promotion by incremental profit: how much money it earned beyond what would have been without it.

The decisions here are these. Whether to repeat a promotion, change it or stop. What depth of discount and what duration to choose. On which goods and in which stores the promotion works and where it does not. Whether promo codes and first-order discounts pay back. What to write into the conclusions so that the next promotion is better.

The reasoning rests on comparison with what would have happened without the promotion (the baseline). Example (prices excluding VAT): price RUB 2,500, cost RUB 1,250, discount 20%. Without the promotion the week gives 1,000 units and RUB 1,250,000 of profit. With the promotion 1,800 units were sold at RUB 2,000, profit is 1,800 × 750 = RUB 1,350,000, revenue grew from 2.5 to RUB 3.6 million (by 44%). Profit gain is RUB 100,000. Subtract the advertising cost of the promotion, RUB 60,000: RUB 40,000 remain. Now the post-promotion dip: customers stocked up, and over the next two weeks 150 units less than usual were sold: minus 150 × 1,250 = RUB 187,500. Result: −RUB 147,500. A promotion with 44% revenue growth is loss-making.

Three rules of honest evaluation follow. The baseline is taken from a control group or a similar period adjusted for season and calendar, not by eye. The post-promotion dip is measured separately, two to four weeks later. And the effect is checked by control: test stores against similar stores without the promotion. A difference-in-differences separates the effect of the promotion from the general growth of the season.

Start with an honest baseline: take sales of the same goods in the same weeks without the promotion or in stores without it. Traps: judging a promotion by revenue instead of profit, choosing “successful” stores for a promotion (self-selection overstates the effect), overlapping promotions and not measuring the dip afterwards.

Key ideas

Incremental profit

Volume growth is not success. Price RUB 3,000, margin RUB 900, baseline 800 units. A 20% promotion leaves RUB 300 per unit, 1,200 sold, 50% more. Promotion profit RUB 360,000, baseline RUB 720,000, costs RUB 40,000: −RUB 400,000. Volume grew and the company lost.

Baseline: what would have been without the promotion

The baseline cannot be taken from last year without adjustments: range, prices and calendar have changed. It is safer to compare with control stores or similar goods. An overstated baseline understates the effect, an understated one inflates it, so it is tested on periods without promotions.

Post-promotion dip

Part of the promotional gain is pulled-forward demand: the customer bought in advance what they would have taken later. It shows as a fall in sales one to four weeks later. If the dip is 150 units on a gain of 800, 19% of the promotional gain was taken from the future.

Control group and difference-in-differences

Test stores grew from 1,000 to 1,600 units (+600), control stores without the promotion from 1,000 to 1,250 (+250) over the same weeks because of the season. The promotion’s effect is 600 − 250 = 350 units, not 600. The method works if the groups were similar before the promotion.

Self-selection and the promo code

The promo code is received and used by those who were going to buy anyway. Of 1,000 orders with a 10% code (RUB 250 at a margin of RUB 1,250) suppose 400 would have happened without it: losses are 400 × 250 = RUB 100,000, profit from 600 new orders is 600 × 1,000 = RUB 600,000. The net effect is RUB 500,000, not 1,000,000.

First-order discount

A first-order discount pays back through repeat orders. A discount of RUB 400 and acquisition cost of RUB 800: the first order gives 1,250 − 400 = RUB 850, and 40% will return for one more order with a margin of RUB 1,250, which is RUB 500. The result is 850 + 500 − 800 = RUB 550 per customer. If only 10% return, the result falls to RUB 175.

A price has an optimum if the elasticity exceeds one in absolute value

At ε = -2.5, a cost of RUB 1,800 and fees of 15% the price excl. VAT is RUB 3,529, on the shelf RUB 4,306, profit RUB 991,878 against RUB 979,918 at the current price. This works only in the range of observed prices.

Formula: p* = c ÷ (1 − fees) × ε ÷ (1 + ε)

What to do: Move to the price in steps measuring the volume; check the VAT rate against the current act.

Think first: What does an elasticity of −0.8 mean for the optimal price?

Answer: There is no interior optimum: the profit grows with the price, only the corridor limits it.

A promotion is assessed against a baseline, not the weeks before it

The baseline 500 pcs is taken from last year and the trend growth 1.25; the actual 760 pcs gives an uplift of 52%, but the promotion profit is −RUB 188,000: break-even needed 917 pcs.

Formula: Promotion profit = actual × margin with discount − baseline × margin without discount

What to do: Compare the same calendar weeks and count the dip after the promotion.

How to work through it

  1. Before the promotion set the goal, depth, duration, goods, stores and the way of measuring.
  2. Prepare control stores or a period similar to the test ones and fix the baseline.
  3. After the promotion count profit, not revenue: price, cost, returns, promotion expenses.
  4. Measure the post-promotion dip two to four weeks later and subtract it from the effect.
  5. Compare with the control and get the effect by difference-in-differences.
  6. For a promo code and a first-order discount account for self-selection and repeat purchases.
  7. Check cannibalization: did neighboring goods and categories fall.
  8. Write a post-mortem: what was measured poorly, repeat, change or stop.

Where to start

If the promotion has run, start with “Promotion result” and enter the baseline honestly. Then “Post-promotion dip” 2–4 weeks later and “Control group” if there was something to compare with. Finish with the post-mortem and its verdict. Calculations of the topic in order: “Promotion result: incremental profit”, “Post-promotion dip and pull-forward demand”, “Control group: promotion effect (difference-in-differences)”, “Promo code: payback with self-selection”, “First-order discount: CAC and LTV”, “Promotion post-mortem: plan, actual and verdict”, “Promo effect without a control group: a baseline from last year”, “Promo item selection under a margin-loss budget”.

Calculations in this topic

  • Promotion result: incremental profit

    Compares the profit of a promotion with the profit you would have had without it and subtracts cannibalization, pull-forward demand and costs. Shows the required and actual sales growth, the baseline cushion and the sensitivity to its estimate.

    Did we earn from the promotion beyond what we would have earned without it.

  • Post-promotion dip and pull-forward demand

    From weeks before, during and after a promotion, estimates the baseline, the uplift, the post-promotion dip and the share of uplift eaten by the dip. Shows how many weeks it took sales to return to baseline and suggests a gap before the next promotion.

    How much of the promotion uplift was pulled-forward demand and when sales returned to normal.

  • Control group: promotion effect (difference-in-differences)

    Compares stores or regions in a promotion with similar ones without it, before and after. Shows the effect of the promotion in pieces and percent, a rough interval from the pre-promotion spread, a “parallel trends” check before the promotion and the result in money.

    What effect the promotion had when compared with similar stores without it.

  • Promo code: payback with self-selection

    Splits orders with a code into “would have bought anyway” and truly additional ones, calculates the code’s profit, the discount lost on “own” buyers and the threshold share of “would have bought anyway” above which the code loses money.

    Does a code pay back if part of the buyers would have bought without it.

  • First-order discount: CAC and LTV

    Calculates the margin of a discounted first order, repeat purchases over 90 and 180 days, the acquisition cost and the customer payback period; compares with spending the same money without a discount if you have a control group.

    Does a customer who came on a first-order discount pay back and is that better than acquisition without a discount.

  • Promotion post-mortem: plan, actual and verdict

    Matches plan and actual on six indicators, takes into account the baseline method and what was measured, and gives a “repeat, change or stop” verdict with a list of measurement errors. Do it 2–4 weeks after the promotion ends, when the dip is visible.

    Repeat the promotion, change it or stop it, and what we measured badly.

  • Promo effect without a control group: a baseline from last year

    Estimates how much would have sold without the promotion when there is no control group. Takes last year’s sales in the same weeks as the promotion and multiplies by the trend growth: this year’s sales outside the promotion against last year’s for the same weeks. From the baseline it gets the promotion uplift, the promotion profit and the break-even volume, and optionally takes the post-promotion dip into account and gives a single result “promotion plus dip”. It complements “Promotion payback” (it needs a known “without the promotion”) and “Difference in differences” (a control group is needed): here is a way to get “without the promotion” from last year.

    How much would have sold without the promotion, what the uplift is and whether the promotion made a profit, including the dip after it.

  • Promo item selection under a margin-loss budget

    Selects which items to put into a promotion so as to free as much cash tied up in stock as possible under a given margin-loss budget. For each item it counts the stock coverage in weeks, the extra pieces from the discount by its elasticity (not above the stock), the capital freed at cost and the margin lost on the volume that would have sold anyway. The priority is the capital freed per rouble of lost margin; items are taken by priority until the budget is exhausted. It complements “Item selection for buying” and “ABC by sales”: those choose the buy and classes, here is a selection for a promotion by stock and elasticity.

    Which items to put into a promotion to free more cash from stock at a given margin budget.

Self-check questions

  • Did we earn on the promotion beyond what would have been without it?
  • How is the baseline defined and has it been tested on periods without promotions?
  • Is profit counted, not revenue, with returns and promotion costs?
  • How much of the gain was pulled-forward demand and when did sales return to normal?
  • Is there a control group similar to the test group before the promotion?
  • Were “successful” stores not picked for the promotion?
  • What share of orders with the promo code would have happened without it?
  • Does the first-order discount pay back counting repeat purchases?
  • Did promotions overlap and can the effect of each be separated?
  • What is written in the conclusions: repeat, change, stop?

Knowledge check

Multiple-choice questions. The correct answer and the explanation follow each question.

  1. Price RUB 1,000, cost RUB 600 (excl. VAT), discount 20%. Without the promotion 100 units would have sold, in the promotion 175 sold: revenue grew by 40%. What happened to profit?

    • A. Grew by 40%, like revenue
    • B. Fell by RUB 5,000: RUB 35,000 against RUB 40,000
    • C. Grew by RUB 35,000
    • D. Did not change: the growth in units offset the discount

    Correct answer: B. Fell by RUB 5,000: RUB 35,000 against RUB 40,000. Revenue: 175 × 800 = RUB 140,000 against 100 × 1,000 = RUB 100,000, that is +40%. Profit: 175 × (800 − 600) = RUB 35,000 against 100 × (1,000 − 600) = RUB 40,000, that is −RUB 5,000. +RUB 35,000 is the whole profit of the promotion, not the gain over the baseline; +40% appears if revenue growth is taken for profit growth. The needed growth in units at this discount is 400 / 200 − 1 = 100%, the actual one is 75%.

  2. Price RUB 4,000 excl. VAT, cost RUB 1,600, discount 25%. By how many percent must sales in units grow so that the profit of the promotion does not fall (returns, losses and costs are not counted)?

    • A. 25%
    • B. 33.3%
    • C. 71.4%
    • D. 171.4%

    Correct answer: C. 71.4%. Profit per unit in the promotion is 4,000 × 0.75 − 1,600 = RUB 1,400, without the promotion RUB 2,400. The needed growth is 2,400 / 1,400 − 1 = 71.4%. 33.3% (0.25 / 0.75) pays back only revenue, not margin; 25% is the depth of the discount itself; 171.4% appears if one forgets to subtract one.

  3. Before the promotion the stores in the promotion sold 120 units a week, the control stores 100. During the promotion: 180 and 125 units a week. What is the effect of the promotion by the control group?

    • A. +20%
    • B. +44%
    • C. +50%
    • D. +25%

    Correct answer: A. +20%. Before the promotion the ratio is 120 / 100 = 1.2, so without the promotion the stores would have sold 125 × 1.2 = 150 units. The actual is 180: the effect is 180 / 150 − 1 = +20% (+30 units a week). +44% appears if compared with the control directly (180 / 125), +50% if compared with their own earlier level (180 / 120) without removing the common growth; +25% if the gain of 30 units is divided by 120.

  4. A promo code gives a 20% discount: 800 orders, an average basket before discount of RUB 5,000 excl. VAT, a margin at full price of 40%, distribution costs of RUB 20,000. What share of buyers who would have bought anyway still does not make the code loss-making?

    • A. 40%
    • B. 50%
    • C. 20%
    • D. 48.75%

    Correct answer: D. 48.75%. The order margin at full price is 5,000 × 0.4 = RUB 2,000, with the code 5,000 × 0.8 − 3,000 = RUB 1,000. The threshold = (800 × 1,000 − 20,000) / (800 × 2,000) = 48.75%. 50% appears without the distribution costs (1,000 / 2,000); 40% is the margin and 20% the depth of the discount: they have nothing to do with the threshold. Above the threshold the discount given to “own” buyers eats the margin of the new orders.

  5. The first-order discount is 30%, the basket is RUB 4,000, the margin at full price is 55%. 1,000 new customers were acquired for RUB 1,500,000. A repeat order without a discount gives a margin of RUB 2,000; each customer makes 0.2 repeat orders in 90 days and 0.35 in 180 days. When does the customer pay back?

    • A. From the first order
    • B. Within 90 days
    • C. Within 180 days
    • D. Does not pay back in 180 days

    Correct answer: C. Within 180 days. CAC = 1,500,000 / 1,000 = RUB 1,500. The first order: 4,000 × (0.55 − 0.30) = RUB 1,000, below 1,500. In 90 days 1,000 + 0.2 × 2,000 = RUB 1,400: still not enough; in 180 days 1,000 + 0.35 × 2,000 = RUB 1,700, above 1,500. “From the first order” would come out if the discount were not subtracted from the margin (4,000 × 0.55 = RUB 2,200). The cohort profit over 180 days is 1,000 × (1,700 − 1,500) = RUB 200,000.

  6. When does a price have no interior optimum at constant elasticity?

    • A. When the cost is above the price
    • B. When VAT is below 22%
    • C. When the elasticity is one or less in absolute value
    • D. When the volume exceeds a thousand

    Correct answer: C. When the elasticity is one or less in absolute value. The formula p* = c ÷ (1 − r) × ε ÷ (1 + ε) requires ε below −1. Covered in: A price has an optimum if the elasticity exceeds one in absolute value.

  7. What is “without the promotion” taken from when there is no control group?

    • A. Last year’s sales in the same weeks adjusted for trend growth
    • B. The weeks directly before the promotion
    • C. The sales plan
    • D. A competitor’s sales

    Correct answer: A. Last year’s sales in the same weeks adjusted for trend growth. The weeks before carry other seasonality, last year in the same weeks keeps it. Covered in: A promotion is assessed against a baseline, not the weeks before it.

Common mistakes

  • Mistake: Judging a promotion by revenue

    What it costs: 44% revenue growth hides a loss

    How to avoid: Count incremental profit.

  • Mistake: Taking the baseline from last year without adjustments

    What it costs: Range, prices and calendar have changed

    How to avoid: Use a control or seasonal adjustments.

  • Mistake: Not measuring the post-promotion dip

    What it costs: Lost weeks are written off as a “normal slump”

    How to avoid: Look at sales two to four weeks later.

  • Mistake: Choosing successful stores for a promotion

    What it costs: Self-selection overstates the effect

    How to avoid: Assign stores randomly or in pairs.

  • Mistake: Overlapping promotions and not separating effects

    What it costs: Unclear what worked

    How to avoid: Separate promotions in time and by group.

  • Mistake: Not doing a post-mortem

    What it costs: Mistakes repeat next season

    How to avoid: Record the conclusion and what was measured poorly.

  • Taking the elasticity from one promotion

    What it looks like (illustrative figures): The optimum becomes random.

  • Ignoring the dip after the promotion

    What it looks like (illustrative figures): The uplift looks larger than it is.

Russia and the EAEU

In Russia evaluating a promotion depends on accounting across channels. In stores the discount shows in the receipt and the till; for marked goods the receipt is issued with the marking code, so a discounted sale can be matched with the code and the date. On marketplaces discounts, coupons and points are often set by the platform, and part is co-funded by the supplier: in the platform’s reports see what is deducted from your revenue and check against the contract. Returns on marketplaces are higher, and counting “sold” without returns overstates the effect.

Truthfulness of information about a discount and the terms of a promotion is required by Article 5 of the advertising law (38-FZ), and the comparison base must be a real price; agree texts with a lawyer. Mailings with promo codes are allowed only with the recipient’s consent (Article 18 of the law), and online advertising of a promotion is marked and reported to ERIR. Promotion expenses (discounts, bonuses, advertising) are booked under tax accounting rules; check the qualification of discounts and bonuses for VAT with an accountant. Promotions in “hot” periods (Black Friday, 11.11, start of the season) overlap with platform promotions: count the baseline within one calendar.

The method of evaluating a promotion (baseline, control, dip afterwards) is the same in all countries, while taxes, platforms and the calendar differ. VAT rates by country are on the page “EAEU and CIS countries” (a guideline as of 14.11.2026, verify), so margin excluding VAT per unit and the payback threshold of a discount differ. In Kazakhstan the Kaspi ecosystem with instalments is strong, where discount and payment terms are mixed, and evaluation is done on the order margin. “Hot” periods differ too: Nauryz, Kurban Bayram, national holidays, and the school season starts differently. Check advertising and consumer protection laws with a local lawyer.

Categories and audiences

Apparel
In apparel promotions are natural at the end of the season, and the main risk lies not in the discount itself but in the dip afterwards and in training customers to wait for sales. Self-selection is stronger on fashion items. Evaluate a promotion by margin profit including returns, which are especially high in online apparel.
Footwear
In footwear promotions work on size tails and season leftovers, and the effect depends on which sizes are in stock. A discount on core sizes wastes margin, a discount on tails unloads the warehouse. Size returns are higher, and the dip after a promotion is weaker because purchases are less “for stock”.
Accessories
Accessories are bought on impulse, and a promotion effectively lifts purchases at the till and in sets. The dip after a promotion is weaker because people buy less for stock, and the gift season makes the baseline depend on holidays. A control is all the more necessary: the rise before 8 March is easily mistaken for the effect of a promotion.
Kids
In children’s goods promotions are tied to the school season and the child’s growth, so pulled-forward demand is pronounced: parents stock up, and the following weeks sag. Measure the dip for longer (four to six weeks). A discount on children’s goods is especially sensitive to truthful information and advertising requirements.
Men
The men’s audience responds to promotions less often and buys more out of necessity, so the effect of a promotion is smaller and self-selection higher: the discount goes to those who would have bought anyway. Basics are easier to evaluate by repeat purchases. Clear limited promotions work better than long discounts.
Women
The women’s audience responds to promotions more actively and compares prices more often, so the effect is higher, but so are cannibalization of neighboring goods and getting used to discounts. Pulled-forward demand and shifting a purchase to the promotion are noticeable. A control and a measured dip are especially needed, and a post-mortem helps choose the depth.

Questions this topic answers

  • Did the promotion pay back once you subtract what would have been bought anyway?
  • How much of the uplift did we lose to the post-promotion dip?
  • What does the control group show?
  • Does a promo code or a first-order discount pay back?
  • Repeat the promotion, change it or stop it?

Key measures and terms

Study first

What next

Related: “Markdown and exit price”, “Season, price and discounts”, “Marketing budget and testing impact”.

Documents on this topic

Open interactively →