Responding to GLP-1 demand shifts in assortment

Why this matters

Since late 2025 nearly every category review has included a slide about GLP-1 medications, and most of those slides are wrong in the same way. They take a measured finding about grocery panels, extend it to a channel it was not measured in, and convert it into an assortment recommendation with a confidence the underlying research does not support.

The finding is real and worth acting on. The action is not "cut snacks."

The best evidence to date comes from Sylvia Hristakeva at Cornell University, published in the Journal of Marketing Research on 18 December 2025, using Numerator panel data covering roughly 150,000 nationally representative US households. Households cut grocery spending by 5.3% within six months of a member starting a GLP-1, and by more than 8% among higher-income households. Savory snacks fell about 10%, with similar declines in sweets, baked goods and cookies. Spending at limited-service restaurants fell about 8%. A handful of categories rose, led by yogurt, then fresh fruit, nutrition bars and meat snacks.

This page is a method for turning that into a decision you can defend: what it licenses you to conclude, what it does not, and how to test whether the predicted shift is actually present in the categories you are responsible for.

The methodology

Step 1: separate the finding from the extrapolation

Write down, explicitly, which parts of your slide are measured and which are inferred. In practice the split looks like this:

The last row deserves attention because it is the most common slide and the least supported. The research found effects persisting at least a year among continuing users with magnitude decreasing over time. That describes a level shift that partially attenuates, not a compounding trend.

The convenience-channel row is the one that does the most damage. The panel measures grocery and foodservice transactions. Convenience baskets are smaller, more impulse-driven and more occasion-bound, and a household buying fewer groceries overall may or may not change its 3pm behaviour. That is a genuinely open question, not a settled one.

Step 2: test the shape in your own data

You cannot identify GLP-1 households in retail data. Nothing in a transaction file marks medication use, and attempting to infer it produces a segment that is mostly noise and raises questions no retailer wants to have to answer.

What you can do is test whether the aggregate shape the research predicts is present. The prediction is specific enough to be falsifiable:

• Savory snacks and sweets soft
• Protein-forward and better-for-you items firm or growing
• The effect larger in higher-income trade areas
• The effect present in units, not only dollars

That last one is the discriminating test. Inflation moves dollars, so a dollar-only decline proves nothing. If the pattern is real it appears in unit velocity, and if it appears only in dollars you are looking at pricing.

Step 3: check the trade-area gradient

The research found a steeper decline among higher-income households. If the pattern in your data is driven by this mechanism, it should be stronger in higher-income trade areas and weaker in lower-income ones.

This is the closest thing to a natural experiment available without shopper identification, and it is a real test: if savory snacks are declining uniformly across every trade area, something else is going on, because the mechanism under discussion is not uniform.

One caution on interpreting the income gradient. Higher-income households showing a steeper decline most plausibly reflects who could access and sustain the medication during the study window, rather than a property of income itself. That distinction matters if you are projecting forward, because access patterns change.

Step 4: respond on the growth side first

The instinct is to cut the declining categories. This is usually the wrong first move in convenience, for a structural reason: the categories the research found growing are ones a convenience store carries thinly or not at all.

Yogurt, fresh fruit, nutrition bars and meat snacks. Of those, most convenience stores carry meat snacks properly, nutrition bars adequately, and the first two barely. The upside is a distribution question you can act on immediately; the downside is a demand question you cannot control.

There is also a regulatory tailwind worth folding into the same decision. SNAP stocking standards are rising to at least seven varieties in each of four staple categories, including dairy and fruits or vegetables, with compliance required from 4 November 2026 for authorised retailers. An operator who accepts SNAP is already going to be adding fruit and dairy varieties. Doing that work once, with the demand shift in mind, is considerably cheaper than doing it twice.

Worked example

Sunrise Market, an illustrative eleven-store operator, testing the pattern across four quarters of unit velocity, indexed to Q1 = 100.

The shape matches the prediction: declines in savory snacks and sweets, growth in meat snacks and nutrition bars, and a control category that did not move. Because these are unit indices rather than dollars, inflation is not driving them.

Then the trade-area split, comparing the four highest-income locations against the four lowest:

The gradient is present and steep, which is consistent with the mechanism the research describes. Note the word: consistent with, not proof of. Higher-income trade areas differ in many ways at once, and this test can rule a hypothesis out more confidently than it can confirm one.

What the operator actually did

Not a snack delist. Four points of shelf in the highest-income four stores moved from the softest savory snack facings into nutrition bars and meat snacks, with the fruit and dairy expansion sequenced against the November 2026 SNAP deadline. The lower-income stores, where the gradient was weak, were left alone pending another two quarters of data.

That is the shape of a defensible response: act where the evidence is strongest, do nothing where it is weak, and let the categories that are growing tell you where the space should go. The assortment optimization method covers how to choose which facings give way.

Turning the GLP-1 assortment read into a review you can defend

A category review that mentions GLP-1 will be challenged, and the challenge is usually fair. Three things make the read defensible.

Show the control category. A page of declining categories proves nothing; declining categories exist every year. The packaged beverages row in the table above, flat across all four quarters, is what makes the other rows meaningful. Without a control, a reviewer cannot distinguish your hypothesis from a general softness in the store.

State the confound you could not remove. In the worked example, the higher-income stores are also the four newest and the two closest to a competing QSR opening. That does not invalidate the gradient, and it does mean the gradient has at least two other candidate explanations. Naming them yourself is both more honest and considerably more persuasive than having a reviewer find them.

Size the decision, not the trend. The finding is four points of shelf in four stores. That framing survives scrutiny in a way "GLP-1 is reshaping the category" does not, and it is also the only part anyone can act on.

What would falsify this

Worth writing down before the next review, because a hypothesis nobody can disprove is not doing any work:

The fourth row is the real test of the decision rather than the hypothesis. If four points of shelf move to nutrition bars and nutrition bar units do not respond, the space was not the constraint, and the correct action is to move it back rather than to add more.

A note on how long to wait

Two quarters before revisiting, which is why the lower-income stores in the example were left alone. Assortment changes take a full purchase cycle to register in the data, and a category read at six weeks mostly measures the disruption of the reset itself.

The brand-side version of the same question

If you are a supplier rather than an operator, the test inverts. You cannot see trade-area income for a retailer's stores, but you can see store-level velocity dispersion, and a widening spread between your best and worst stores in a better-for-you category is the same signal arriving through a different door.

The action differs too. An operator moves shelf; a supplier makes the case for it, which means arriving at the category review with the store-level evidence rather than with the Cornell headline the buyer has already seen four times this quarter.

Doing this in Scout

Every test on this page is a comparison across time and store groups on consistent definitions: unit velocity indexed by quarter, split by trade-area cohort, with a control category. The hard part is not the arithmetic, it is keeping the definitions stable across periods so the comparison stays honest as new data lands. Scout connects your retailer and distributor data and holds those definitions in saved views that update rather than being rebuilt.

Two honest limits. Scout cannot identify GLP-1 households, and neither can any other retail data tool, because the information is not in the file. And it reads the data you connect, so a category benchmark for the channel needs syndicated data alongside your own.

Summary and further reading

• The Cornell and Numerator research is solid and measures grocery and foodservice panels, not convenience baskets. That extension is an inference.
• Test the predicted shape in unit velocity rather than dollars, or inflation will answer the question for you.
• Use the trade-area income gradient as a falsification test; a uniform decline across all trade areas points to a different cause.
• Respond on the growth side first, because the growing categories are the ones convenience under-carries, and sequence it with the November 2026 SNAP stocking change if you accept SNAP.
• No retail file identifies these households, and inferring them creates a segment that is mostly error.

Further reading: assortment optimization and SKU rationalization for the space decisions, and the GLP-1 shopper glossary entry for the source figures.