Base and incremental volume: separating the two
A promoted week sells more than an ordinary week. Some of that would have happened anyway, and some of it is attributable to the promotion. Base volume is the first part; incremental volume is the second. Without the split, a promotion's total sales are just a number that looks large.
This is the single most consequential decomposition in trade analysis, and it is a modelled estimate rather than an observation.
How is base volume estimated?
From the behaviour of the same item in comparable non-promoted periods, adjusted for seasonality, trend and whatever else the method accounts for. The estimate answers the counterfactual question: what would this item have sold in this week without the promotion.
Because it is a counterfactual, it cannot be observed and it cannot be verified directly. It can only be assessed on whether the method is sound and the assumptions stated.
Why does the method matter so much?
Because the entire incremental figure is the residual after base is removed. A base estimate that is ten percent too low makes the promotion look better by exactly that amount, and nothing in the output signals the error.
That is why the right question about a lift number is never "how big is it" first. It is "how was base estimated," and a provider or an agency that cannot answer plainly is presenting an opinion as a measurement.
What confounds the estimate?
Seasonality, competitor activity, distribution changes, weather, a calendar shift, and any other promotion running at the same time. Each of those can move actual sales in the promoted week, and a base model that does not account for them attributes the movement to the promotion.
In the independent channel, distribution change is the one most often missed. An item that gained stores during the promotion period will show incremental volume that is a distribution effect.
What about pantry loading?
A promotion that sells volume forward borrows from future weeks. The promoted week's incremental figure is real and the following weeks' base is depressed, so a measurement window that stops at the end of the promotion overstates the net effect.
Reading the weeks after a promotion is part of measuring it. A report that shows only the promoted period is showing half the arithmetic.
How should incremental volume be presented?
With its assumptions, its window, and its uncertainty. A single number with a decimal place implies a precision that no counterfactual estimate has, and a range communicates the actual state of knowledge.
Presenting a range is commercially uncomfortable and analytically honest. The discomfort is a reason to explain it, not a reason to drop it.
Does this apply outside promotions?
Yes. The same decomposition underlies any incrementality question: a new listing, a display, an advertising burst. Each asks what would have happened otherwise, and each depends on a defensible counterfactual.
That is worth saying because the promotional version is the most familiar and the reasoning generalises. Anyone comfortable with base and incremental already understands most of what makes a media test credible.
What should a brand ask before funding a promotion?
What the base method is, what window will be measured, what will be controlled for, and what result would count as a failure. The last question is the one that turns a measurement into a decision rather than a justification.
Agree it before the promotion runs. Any figure chosen after the numbers arrive is a rationalisation, however reasonable it sounds in the meeting.
Frequently asked questions
Can base volume be observed directly?
No. It is an estimate of what did not happen, which is why the method and its assumptions carry the whole weight of the conclusion.
Is a bigger lift always better?
Not if it was bought expensively or borrowed from future weeks. Lift is a volume measure and the decision is usually a profit decision, which needs the cost of the promotion alongside it.
What window should be measured?
Long enough to include the post-promotion period where pantry loading shows up. Stopping at the last promoted day flatters nearly every promotion.
How do competitor promotions affect the estimate?
They can depress your base or your actual sales depending on timing, and a method that does not observe competitor activity will absorb it into your result. Ask whether the model sees it.
Does this work in a small-format channel?
Yes, with the same caveats and one additional one: store counts are smaller, so store-level noise is larger relative to the effect. That argues for more stores or longer windows rather than for skipping the analysis.
What about promotions with no price change?
Display and feature support can drive incremental volume without a price move, and the same decomposition applies. What changes is that price is no longer the explanatory variable.
Should I compare lift across categories?
With caution. Base volatility differs by category, and a lift figure in a stable category is a more reliable number than the same figure in a volatile one.
What is the most common mistake?
Treating total promoted sales as the result. It is the number that is easiest to produce and the one that answers no question anybody asked.
### What does a credible lift report contain?
The base method, the window, the controls, the point estimate and a range. Five elements, and a report missing any of them is presenting a conclusion without the evidence that would let somebody check it.
The range is the element most often dropped and the one that most changes how a result is received. A lift stated as a range invites a decision about risk; a single figure invites a decision about nothing. External context on retail measurement practice is collected by trade bodies such as FMI.
Working with retail data? See NRS Insights data solutions or contact NRS Insights.