At the end of the month, two teams at the same company drop two reports on the CEO's desk for the exact same sale. Marketing is beaming: conversion jumped, and orders are up 40%. Finance opens the profit statement and frowns: revenue dipped, and net margin went negative.
In a column on Cossa.ru (announced in Cossa's Telegram channel), Growth Marketing Lead Anastasia Kurnosova unpacks the root of this paradox. Marketing grades a campaign by conversion rate (CR — the share of visitors who pay), without asking how many of those people would have bought anyway at full price.
Why the Conversion Report Looks Great While Cash Drops
Let's walk through the math from her example. A product costs 800 ₽. In a normal period with no promo, 500 people buy it, bringing in 400,000 ₽.
The team launches a 30% discount. The price drops to 560 ₽, and over the same period, orders rise to 700. Marketing reports +40% in sales. Pull out a calculator: 700 orders at 560 ₽ equal 392,000 ₽ in revenue. You shipped 200 extra units and collected 8,000 ₽ less.
And that is before subtracting the cost of goods for those 200 extra units and your email or ad spend. In reality, the profit hit is much worse. Why? The promo brought in only 200 new buyers, yet all 700 got the 240 ₽ discount. The company handed 120,000 ₽ in margin to 500 'own' customers just to get 112,000 ₽ in revenue from new orders.
Here you have to separate two questions. Incrementality measures how many extra sales happened strictly because of the promo and would not have happened without it (200 in this case). Economic efficiency checks whether those extra orders cover both the promo costs and the margin given away to baseline buyers.
Play with the sliders below using your own numbers. Move unit cost closer to your real purchase price and see how many orders you actually need just to break even on profit.
Discount Incrementality Calculator: Profit and Break-Even Point
Notice how quickly the picture changes once unit cost is above zero. If sourcing and logistics eat even 300 ₽ per order, a 30% discount forces you to nearly double order volume just to keep the same gross profit. Any growth below that bar means you are paying for higher turnover out of your own pocket.
Anatomy of Cannibalization: How Discounts Subsidize Existing Buyers
The costliest mistakes happen when shoppers are already half a step from checkout. Picture a normal scene: someone visits your store, drops an item in the cart, and closes the tab. Not because it is too pricey — they just got distracted, wanted to read reviews later, or are waiting for payday in two days.
Then an automated CRM flow fires and sends them a 10% discount email. They come back and pay. In the CRM report, a green flag lights up: the email drove a sale!
That is where the gap between attribution and incrementality shows up. Attribution simply logs a sequence of events: 'there was a touchpoint, then a payment happened.' Incrementality asks the uncomfortable question: 'would they have bought without that email?' If yes, the business created zero new demand and gave away 10% of revenue for nothing. Mass promo codes for warm audiences, discounts for repeat buyers, and aggressive retargeting at people already ready to buy burn margin the exact same way.
Comparison of attribution and incremental lift in a single report (Source: AppsFlyer)
The issue goes beyond a single SKU: promos often eat into sales of neighboring items in your own catalog. This is called demand and margin cannibalization. John Dawes of the Ehrenberg-Bass Institute analyzed retail data across 132 SKUs from 45 brands in 12 categories (US, UK, and Australia) and published the findings in the Journal of Retailing.
Cannibalization and Post-Promo Dip from Price Promotions (John Dawes Study)
The chart is sobering: shoppers rarely consume more of a brand overall. More often, they simply switch from the regular pack to the discounted one. On top of that, future demand shifts forward: people stock up at the lower price, leaving the shelf quiet and sales lower in the weeks after.
To be fair, Anastasia Kurnosova notes that Dawes' study used retail observational data rather than a controlled experiment, so you should not treat those percentages as rigid benchmarks for every niche. Still, the takeaway is rock solid: judge a promo across your whole assortment, include the post-sale dip, and measure final margin rather than one SKU's spike.
How to Measure Incrementality Without Guessing: Holdout Tests
To stop guessing, you need a fair experiment — a holdout test. The idea is simple: take one relevant audience and split it randomly into two groups. The test group sees the discount; the control (holdout) group does not. Every other condition must match closely, and the promo code must not leak to the control group through retargeting or other emails.
Following incrementality measurement practices (such as AppsFlyer's methodology), teams typically use three formats:
- User-level
holdout— the classic choice for CRM, email flows, paid social, and retargeting. Usually 5–10% of the audience goes into the control group, which protects revenue while collecting enough data for statistical confidence. - Coupon
holdout— a slice of site visitors simply does not see the promo pop-up or banner, showing how clean traffic converts on its own. - Geo-experiment (
geo-holdout) — a lifesaver for site-wide sales and offline retail where you cannot isolate individual shoppers. Instead of users, you compare similar regions (analysts sometimes use Bayesian time-series models to adjust for seasonality, though the core logic stays the same: comparing against a baseline trend).
Once the test wraps up, you calculate net new orders with a simple formula that scales the control group to match the test group's traffic:
# Orders genuinely created by the promotion
Incremental_conversions = Test_conversions
- (Control_conversions * Test_traffic / Control_traffic)
Next, you convert those extra orders into money — incremental contribution margin — by subtracting unit costs, campaign spend, and the subsidy given to buyers who would have converted anyway.
Comparison of actual incremental lift and baseline demand in campaign analytics (Source: AppsFlyer)
Watch out for rushing the timeline. A standard window for this test is 14 to 30 days so it covers normal weekly buying cycles, plus mandatory tracking after the promo ends. If you call the results after one week, the test group may look like a clear winner. Wait a month, and the control group often catches up in total sales, proving the discount did not win new customers — it just pulled future purchases forward by two weeks.
Where Discounts Actually Work — and Where They Break the Business
You do not have to ban discounts completely. Sometimes taking a loss on the first order makes total sense.
First, it can be an LTV (Lifetime Value) play. If a promo brings in new customers rather than one-off deal hunters, the lost margin on the first order pays off through repeat purchases. Second, launching a new product or entering a new region where shoppers need a gentle push to try you. Third, operational goals: filling idle capacity, clearing warehouse space, moving perishable stock, or speeding up cash turnover. The key is locking in the goal and evaluation window before you launch.
Never spray one flat discount across your whole database, since segments differ in price sensitivity and buying cycles. Often price is not even the blocker: shoppers in a new region may not pay because cards fail technically or the site lacks reviews. Another promo code will not fix a broken checkout.
It also helps to separate B2C from B2B. In high-volume B2C, cycles are short, samples are large, and holdout tests are easy to run. In B2B, deals are fewer and a discount usually turns into 'special terms'. That creates a reference price trap: give a new client 25% off at the start, and they anchor on that lower price and demand it at renewal, while your sales team gets hooked on closing tough negotiations with discounts.
After measuring incrementality, your decision comes down to three options:
- Keep and scale — the promo generates net profit or brings in customers with a healthy
LTVpayback. - Narrow to specific segments — lift only shows up among new or price-sensitive cohorts, so remove the discount for warm buyers.
- Shut it down — the promo is just handing margin to people who were already going to pay.
Before running any discount, Anastasia Kurnosova suggests answering two questions: 'If we stop offering this discount tomorrow, how many sales will actually disappear? And how much profit will disappear with them?' Until you have those answers in numbers, your business does not know the real cost of its discount policy.

