Trade promotion is the second-largest line on most consumer goods P&Ls, and the number used to justify it is measuring something other than what people think it measures.
For two decades, the accepted wisdom has been that roughly three-quarters of a promotional bump comes from brand switching, meaning you take volume from rivals. That belief traces to a real body of scanner-panel research. It also rests on a units error that was formally corrected in 2003, and the correction changes the answer by a factor of more than two.
Answer first
Approximately 74% is the share of the promotion ELASTICITY attributable to brand switching. It is not the share of the extra units. In Is 75% of the Sales Promotion Bump Due to Brand Switching? No, Only 33% Is (Journal of Marketing Research, 40(4):481-491, November 2003), Harald van Heerde, Sachin Gupta and Dick Wittink demonstrated that the elasticity result "does not imply that if a brand gains 100 units in sales during a promotion, the other brands in the category lose 74 units." Applying their transformation to previously published elasticity decompositions, they found that approximately 33% of the unit sales increase is attributable to losses incurred by other brands in the same category.
So about two-thirds of your promotional lift is not taken from a competitor. It is your own existing and future buyers, purchasing earlier than they would have and in larger quantities than they would have. You did not win that volume. You borrowed it from yourself, mostly from next month.
The sign inversion that matters
This is the inversion that makes trade promotion measurement so persistently misleading, and it runs in the opposite direction to how the category is usually discussed.
When you research why shoppers choose one brand over another, you are studying volume moving between households and brands. Trade promotion analysis inherits that mental model and assumes the bump is competitive conquest.
But the dominant mechanism in a promotional bump is volume moving within the same household, across time. The causal question is not "whose buyer did I take" but "when would this buyer have bought anyway, and how much." Those are different questions, they need different data, and only one of them is answerable from shipment or POS totals.
Three components make up any promotional bump:
| Component | What it is | Is it incremental? | Visible in POS lift? |
|---|---|---|---|
| Brand switching (secondary demand) | A buyer chooses you instead of a rival | Yes, genuinely incremental to you | No, indistinguishable |
| Purchase acceleration (timing) | Your buyer buys this week instead of in three weeks | No. Borrowed from a future period | No, indistinguishable |
| Quantity increase (stockpiling) | Your buyer buys three instead of one | Partly, and often followed by a trough | No, indistinguishable |
The headline lift number sums all three and reports the total as success. The van Heerde correction says the genuinely competitive component is about a third of it.
Why the post-event analysis usually never happens
The industry figure most often quoted is that somewhere between 59% and 72% of trade promotions fail to break even once cannibalisation, pantry loading, distributor fees and deduction write-offs are netted out. That range circulates widely in consulting and vendor material. We could not trace it to a published primary source with a stated methodology, so treat it as a directional industry estimate rather than a measured statistic. The honest version of the claim is narrower and better supported: promotional lift as normally reported systematically overstates incremental volume, because two of its three components are not incremental.
The structural reason the error survives is mundane. The data needed to prove a promotion lost money sits in three systems that do not reconcile: trade spend in the TPM system, shipments in ERP, and consumption in syndicated POS. Reconciling them is nobody job title, so the post-event analysis that would settle it does not get done, and next quarter is planned on lift.
What actually establishes incrementality
There is only one class of design that estimates a counterfactual: something that creates a comparison group which did not receive the promotion. We covered the general form of this argument in marketing mix modeling vs attribution vs incrementality, and trade promotion is the same problem wearing a different hat.
- Matched-market or geo holdouts. Withhold the promotion in a set of comparable stores or markets. This is the only clean answer, and the reason it is rare is that retailers dislike it and sales teams dislike it more.
- Baseline modelling. Estimate what would have sold absent the promotion. Only as good as the baseline, and the baseline is exactly the thing under dispute.
- Post-promotion dip analysis. Measure the trough after the peak. If volume falls below baseline for the following weeks, you have direct evidence of acceleration rather than conquest.
None of these tell you why the bump had the composition it had, and that is the part that determines whether to repeat the promotion.
The question POS data cannot answer, and interviews can
Every method above works on aggregate transactions. None of them can distinguish these four shoppers, all of whom bought your promoted pack in the same week:
- Bought yours instead of the competitor because of the discount. Genuine conquest.
- Was going to buy yours in three weeks and pulled the purchase forward. Borrowed.
- Buys yours every week regardless and paid less this time. Pure margin loss.
- Bought yours because the competitor pack was out of stock, and never noticed the promotion. Not caused by the promotion at all.
In POS data these four are one number. In an interview they are four completely different accounts, and shoppers recall them accurately because a recent grocery trip is a concrete episode rather than an abstraction.
This is where AI-moderated research changes the economics. Historically, reaching several hundred promotion-exposed buyers within days of the event was impossible at a sensible cost, so nobody tried, and the composition question went permanently unanswered. Running it as an interview study is now a two-day exercise.
A Koji study for a post-promotion read uses all six structured question types in one conversation: single_choice for which pack was bought, yes_no for whether the shopper noticed the promotion at all (the answer is humbling more often than teams expect), scale for planned purchase timing before the trip, multiple_choice for what else was in the basket, ranking for what drove the choice, and open_ended with AI follow-up for the account of the trip. The yes_no promotion-awareness question alone reclassifies a meaningful share of "promotion-driven" volume every time it is asked.
Design notes that matter:
- Field within days. Purchase-timing recall degrades fast. This is the single largest quality lever.
- Ask the counterfactual directly and carefully. "When would you next have bought this if it had not been on offer" is answerable; "would you have bought it anyway" invites a yes.
- Watch for social desirability. Few people enjoy describing themselves as deal-driven, and stated price sensitivity is understated in almost every category.
- Recruit non-buyers too. Interviewing only people who bought during the promo window is textbook survivorship bias and will make any promotion look good.
- Pre-commit to the analysis. Promotional post-mortems are unusually vulnerable to analytic flexibility, because the window boundaries, the baseline period and the comparison set are all choices, and all three move the answer.
What this changes about promotion planning
Reframe the objective. If two-thirds of the bump is borrowed from your own buyers, then a promotion evaluated on lift is being graded on a metric that mostly measures its own cost. Three practical consequences:
- Promotion depth and frequency trade off against each other. Frequent shallow promotions train your best buyers to wait, converting full-margin volume into discounted volume with no share gain.
- The most profitable promotions are the ones reaching category buyers who do not currently buy you. That is a targeting question, and it needs to be researched before the promotion, not after.
- A promotion that fails on incrementality may still be justified by a trade term, such as distribution or display or a listing. That is a legitimate reason. It should be stated as the reason, rather than laundered through a lift number. The same disclosure problem appears in retail media measurement, where the party selling the media also reports the return, and in category captain arrangements, where a competitor advises the retailer on your shelf.
Get the composition of your next promotion with Koji
Koji is the AI-native customer research platform for teams who need the reason behind the number, not just the number. Launch an AI-moderated interview study within days of a promotion, reach hundreds of category buyers, and get quantified themes plus segment cuts in a one-click report. Structured questions deliver the counterfactual timing data; the conversational AI captures the account of the trip that no scanner file contains.
Your syndicated data will tell you the bump happened. It will never tell you what fraction you paid for twice. Start a free study on Koji and find out before you commit next quarter trade calendar.
Related reading
- Price Pack Architecture (2026) - why retail price setters invert your pack-price ladder on your best-selling size.
- Shrinkflation Research (2026) - why scanner data and fairness experiments disagree, and which shoppers each one measures.
- Personalized Pricing in 2026 - what individualised prices do to the reference price your research assumes.
Frequently Asked Questions
What percentage of a trade promotion bump is real incremental volume?
On a unit-sales basis, roughly a third of the increase comes at the expense of other brands in the category. Van Heerde, Gupta and Wittink (Journal of Marketing Research, 2003) showed that the widely cited 74% figure describes the promotion ELASTICITY, not units, and that the corresponding unit-sales figure is approximately 33%. The remainder is purchase acceleration and quantity increases among buyers you already had.
Why is the 74% brand switching figure wrong?
It is not wrong, it is misapplied. About 74% of the sales promotion elasticity is attributable to secondary demand effects, which is a statement about elasticities. It does not mean that a brand gaining 100 units causes rivals to lose 74. When the same underlying results are transformed into unit sales, the cross-brand loss is around 33% of the gain.
How do I actually measure trade promotion incrementality?
Only a design with a genuine comparison group estimates a counterfactual: matched-market or store-level holdouts where the promotion is withheld. Baseline modelling and post-promotion dip analysis are useful supplements, with the dip being direct evidence of purchase acceleration. None of these explain why the bump had its particular composition, which is what interviews add.
Do most trade promotions lose money?
Industry sources commonly claim 59% to 72% of promotions fail to break even once cannibalisation, pantry loading and deduction write-offs are counted. That range is widely repeated but we could not trace it to a published primary study, so treat it as directional. The better-supported statement is that reported lift systematically overstates incremental volume, because acceleration and stockpiling are counted as gains.
What should I ask shoppers after a promotion?
Field within days, and ask when they would next have bought the product absent the offer, whether they noticed the promotion at all, what else was in the basket, and what actually drove the choice. The promotion-awareness question routinely reveals that a share of supposedly promotion-driven volume was bought by people who never saw the offer.
Can AI-moderated interviews replace holdout testing?
No, and they should not. A holdout estimates the size of the incremental effect; interviews explain its composition and why it occurred. Run the holdout to size it, run interviews to understand it, and use the interviews to design the next promotion rather than to grade the last one.
