Your elasticity is probably measuring your promo calendar

Ask a commercial team for the price elasticity of a given SKU and most can produce a number. Ask how that number was estimated and the answer is usually some version of the same thing: a model fitted to several years of price and volume history, by SKU, by retailer.

That method has a structural problem. Across Europe's six largest grocery markets, roughly 34 per cent of branded units sell on promotion. In some categories it runs considerably higher. When a third or more of the observations in a price series are promotional, a model fitted to that series is not measuring willingness to pay. It is measuring how shoppers respond to deals.

Two different behaviours, one number

Base price elasticity and promotional elasticity describe genuinely different things, and they are usually different in size by a factor of several.

Base elasticity reflects willingness to pay. When a shelf price moves and stays moved, some shoppers leave the brand, some trade to another pack, some trade to own label, and some accept the new price. That response tends to be moderate, and it plays out over months.

Promotional elasticity reflects deal seeking. A temporary price cut, usually with a display or a shelf barker attached, pulls forward purchases from shoppers who would have bought anyway, attracts switchers who will leave again when the deal ends, and prompts stockpiling that suppresses the following weeks. That response is sharp and short.

A single coefficient fitted across both produces a blend weighted by how often the brand promoted. That gives a number that is specific to the promotional calendar rather than to the product, which is exactly the wrong property for a planning parameter.

What goes wrong downstream

The consequences are predictable and they compound.

• Price rises get underestimated or overestimated. A blended elasticity usually overstates how much volume a base price rise will cost, because it carries the sharpness of promotional response. Teams then leave price on the table, or they build a business case that fails to survive contact with the finance director.
• Promotional plans get justified by the wrong number. Using a blended figure to model a deal understates the response, which makes weak promotions look adequate and keeps them in the calendar.
• Pack architecture inherits the error. Every ladder decision rests on relative elasticity between packs. If each pack's figure carries a different amount of promotional contamination, depending on how often that pack was promoted, then the relative comparison between them is not measuring what it appears to measure.
• Scenario tools look more precise than they are. A well built simulation running on a contaminated parameter produces confident outputs to two decimal places. Precision is not accuracy, and the interface rarely communicates the difference.

The competitor problem sits on top

There is a second contamination, less discussed and harder to fix. Most elasticity estimates treat the observed volume in a non-promoted week as the baseline. That baseline assumes nothing else was happening.

Something else usually was. If a competing brand ran a deep cut in the same week, your non-promoted volume was suppressed by their activity rather than by your price. The model attributes the dip to your own price position, and the resulting coefficient absorbs a competitor effect that has nothing to do with your shopper's view of your brand.

This is why competitor pricing and promotional data belongs in the estimation rather than in a separate tracking report. A price tracker that sits in its own platform, consulted when someone remembers, cannot correct a baseline.

What a defensible estimate requires

None of this is exotic. It is mostly a question of construction discipline.

1. Separate the promotional periods. Estimate base response on de-promoted series, and estimate promotional response separately by mechanic. Two parameters, not one.
2. Build the baseline with competitor activity in it. The counterfactual has to account for what else was in market that week, otherwise incrementality and elasticity both inherit the same error.
3. Handle the pull-forward. Stockpiling depresses the weeks after a deal. A model that ignores the post-promotional dip overstates promotional lift and understates base stability.
4. Estimate cross effects explicitly. Volume lost to your own larger pack is a different commercial event from volume lost to private label. Both show as a decline in the SKU.
5. Refresh more than annually. An elasticity refreshed once a year and used for twelve months of weekly decisions is a structural mismatch between how fast the analysis moves and how fast the market does.

The practical test

There is a quick diagnostic worth running before the next planning cycle. Take the elasticity figures currently in use and plot them against the promotional participation of each SKU over the estimation window.

If the two are correlated, and they usually are, the numbers are describing the promotional calendar rather than the products. That is a finding with a direct commercial consequence, because every pricing decision built on those figures has inherited the same distortion.