Incremental sales lift: The path to profitable promotions

In this blog post, we will look at how to measure incremental sales lift and turn it into a profitability decision, including:

According to Harvard Business Review and Talon.One, a survey of 420 respondents found one clear result. 65% of organizations planned to increase their focus on promotion profitability. That reflects a broader shift away from blanket discount strategies. Incremental sales lift starts to answer the profitability question, and incremental contribution is what finishes it.

Incremental sales lift is the additional sales a promotion generates above the baseline sales that would have occurred without it. The calculation is actual sales during the promotion window minus baseline sales. Everything else in the observed bump is not incremental. Some customers would have bought at full price, a promotion may pull future demand forward, or it may shift volume away from non-promoted products.

The same problem shows up in loyalty spend. ROI is easier to claim than to prove, which is exactly the gap a holdout test is built to close.

Lift on its own does not tell you whether the promotion made money. A promotion can generate positive incremental sales and still lose money once discount cost, fulfillment and payment-processing costs, and cannibalization are factored in. Lower contribution margin on promoted units and loyalty-point liability can erode the result further. The formula that actually answers the profitability question is:

Incremental contribution = incremental revenue − incremental COGS − discount and reward cost − other variable campaign costs

A basic example makes the distinction concrete. A campaign reaches 80,000 customers while 20,000 comparable customers remain in a control group. During the test, 5% of the control group buys, a rate equal to 1,000 control-group buyers.

If the promoted group had followed the same baseline rate, 4,000 customers would have bought without the offer. Instead, 4,800 customers buy.

The incremental sales figure comes from a simple rate-based formula:

Incremental sales = N(treatment) × (purchase rate treatment − purchase rate control)

For this example: 80,000 × (6% − 5%) = 800. The campaign creates 800 incremental sales, a 20% lift above the 4,000-sale baseline, even though all 4,800 transactions appear in the promoted group's report. Finance can then compare that incremental revenue and margin against the total promotion cost. This comparison is more useful than treating every redeemed offer as a new sale.

A 2023 BCG analysis found that 30% to 40% of retail promotions are either inefficient or unprofitable. It also found that 40% to 60% of promotions retailers expect to generate high ROI end up delivering low ROI instead.

A related gap shows up in promotion engine software research. Roughly 18% of revenue at a typical brand goes to discounting with no clear attribution or ROI. Gross lift hides these problems because it combines several kinds of non-incremental volume.

These hidden effects explain why a credible baseline matters. Without one, a team cannot separate new demand from subsidized sales or account for timing shifts and substitution.

Run a holdout test. Withhold the promotion from a randomly selected slice of the eligible audience, run it for everyone else, then compare the groups. Withholding 10% to 20% of the eligible audience is a practical starting point. This range is a vendor recommendation, not a universal statistical rule.

The right share depends on the baseline conversion rate, the minimum effect worth detecting, and risk tolerance. A power analysis should confirm the size before launch rather than defaulting to a fixed percentage.

A holdout creates a credible counterfactual, but it does not automatically capture every economic effect. Define the outcome at the right level and measure long enough to catch what a simple treatment-versus-control comparison can miss:

A few design decisions determine whether the result means anything. Build the test and control groups from the same eligible population. Mismatched buying history or customer status can otherwise look like campaign impact on its own.

Give both groups enough conversions to support a reliable comparison, since very small holdouts create uncertainty that can look like a real difference. Choose the test window before launch and avoid stopping a fixed-horizon test because a variant briefly leads. Repeated unplanned looks increase false-positive risk. If a team needs to monitor results early, use a pre-specified sequential-testing method and decision boundary instead of an ad hoc stop.

Avoid giving the holdout the same offer through another channel. That kind of crossover contaminates the comparison and can bias the estimated effect toward zero. If contamination happens anyway, log it and decide in advance how it will be handled in the analysis rather than discarding the test.

Teams should also distinguish percentage-point changes from percentage lift. Suppose the control group converts at 5% and the promoted group converts at 6%. The absolute difference is one percentage point, and the relative lift is 20% because the one-point gain sits on a 5% baseline.

This distinction matters when comparing campaigns with different baseline rates. Moving from 1% to 2% is a 100% relative lift. Moving from 20% to 21% is only a 5% relative lift, even though both improve conversion by one percentage point.

Report an estimated effect alongside a pre-specified confidence interval, such as 95%. Statistical significance addresses uncertainty under the test's assumptions. It does not establish that the promotion clears the commercial hurdle, and a narrow interval around a small contribution loss is still a losing campaign.

Choose the holdout format that fits how customers buy:

Card-linked offers can support the same method when the program records offer exposure and settled transactions. Dagrofa runs card-linked loyalty across 520 grocery stores on this kind of setup. Teams can compare exposed and withheld cardholders against one consistent transaction record instead of reconciling separate systems.

These strategies share one principle. They make the promotion conditional on a customer, behavior, basket, or timing signal instead of broadcasting value indiscriminately.

A grocer may test basket thresholds by market. A QSR brand may test daypart offers by location, and an ecommerce retailer can randomize at the customer level. The mechanic changes, but the underlying causal question stays the same.

Begin with a campaign whose result would change a real budget decision. Do not try to redesign the entire measurement system at once. Choose a high-spend campaign or one that receives regular scrutiny, define the holdout before launch, and complete one clean test.

Write down the decision rule before results arrive. For example, expand the campaign only if incremental margin exceeds promotion cost by an agreed amount. Redesign the offer instead if lift appears positive but fails to cover the reward expense. A pre-agreed rule reduces the temptation to reinterpret weak results after the campaign.

The same underlying logic applies to loyalty program ROI: incremental revenue minus total program cost, divided by total program cost. Incrementality is the distinction that makes the number meaningful in both cases.

Track the calculation from eligibility to financial outcome. Record how many customers entered each group, how many bought, and the revenue each group generated. Estimate the promoted group's baseline from control-group behavior, then subtract that baseline from the actual promoted-group result. Check post-campaign and category sales for timing shifts and substitution before drawing a conclusion.

If incremental sales come in well below what redemption counts implied, the campaign is a reallocation opportunity. If the result is strong, the team has a credible case for continued investment. Document each test's audience, exposure, outcome definition, financial result, and decision, so the next campaign improves on evidence rather than redemption volume.

Running this kind of test depends on connected data more than on statistics. Acquisition offers, clearance markdowns, member promotions, and free shipping often live in separate tools. Different teams can end up drawing from the same margin without seeing the combined effect on one customer.

A promotion measurement program needs three things in place. It needs transaction records connected across systems, since every additional platform and data handoff adds a place where identifiers or timestamps can stop matching. It needs preserved holdout assignments, since a clean test requires one consistent record of eligibility, exposure, redemption, and transaction outcome across the test window.

It also needs exposure compared directly with outcomes. When offer-exposure events reach the same system as cart data, teams gain one comparison record instead of reconciling several.

Joe & The Juice connects personalized offers to checkout in real time across an omnichannel setup. In 2024 the company reported a 17% revenue increase and 33% digital sales. That result illustrates the kind of data connectivity a measurement program needs, not proof that the connected setup itself caused the lift.

Depending on the stack, a CDP, engagement platform, commerce platform, and promotion engine all need consistent identifiers and exposure events for this comparison to work. That requirement varies across promotion management platforms.

Talon.One's Experiments feature runs two promotion rule variants inside one campaign and reports confidence scores for each. A confidence score describes the statistical precision of a comparison. It does not by itself prove commercial profitability, and it works alongside randomization and a defined outcome rather than replacing either one.

This later figure comes from the same survey referenced earlier. According to HBR and Talon.One, only 48% of executives rate their organization's promotion execution as effective. The same survey found most organizations consider profitability a growing priority.

Teams that can produce incrementality and contribution numbers have a more specific conversation with finance. They can show which campaigns create demand, which shift timing, and which need redesign.

Start with one recurring, high-spend promotion. Randomly withhold the offer from a pre-specified control group and define the outcome at the basket or category level where relevant. Continue measurement after the offer ends, and calculate incremental contribution, not only incremental orders or revenue, before deciding to scale. If the treatment cannot beat the control after discount cost, variable costs, and post-promotion effects, redesign it or stop it.