Substitution analysis without loyalty cards
Substitution — what a shopper buys instead — is usually studied with identified shopper histories. In a channel without loyalty identifiers, that method is unavailable, and the question does not go away.
What remains is inference from baskets, from stockouts, and from cross-item behaviour over time. Each is weaker than panel-based substitution and each is genuinely informative.
What can a basket tell you?
That two items were bought together, which is the opposite of substitution. Items that co-occur are complements or independent; items that almost never co-occur in a category where shoppers buy one are candidates for substitution.
Absence is the signal, and absence is weak evidence on its own. A pair of items that never co-occur may be substitutes or may simply be bought by different people.
What can a stockout tell you?
Considerably more. When an item is unavailable in a store for a period and another item's sales rise in that store and not in comparable stores, that is direct evidence of substitution.
Natural stockouts are the closest thing to an experiment available in this channel. They are messy, unplanned and not randomly assigned, and they are still the strongest available signal.
How is a stockout identified without observation?
By a rule: consecutive zero-sales periods in a store that normally sells the item, above a threshold. The rule produces false positives — a store that delisted, a store that stopped reporting — so it needs a check against store-level reporting completeness.
Stating the rule matters. A stockout-based substitution study rests entirely on how stockouts were defined.
What about price-driven substitution?
Periods where one item's price changed and another's did not offer a similar natural comparison. Cross-price responses estimated this way are noisy and directionally useful, particularly when many such periods are pooled.
The caution is confounding: the price change may have come with display support, and the response then includes it.
What cannot be answered?
Whether the same shopper switched. Without an identifier, all of this is aggregate behaviour, and aggregate substitution is consistent with several different individual stories.
Saying so is part of using the method honestly. Tender type is a payment method, not a persona, and inferring household characteristics from how somebody pays is analytically sloppy and ethically wrong.
What is the practical output?
A substitution map at category level: which items' sales respond to which others' absence or price. It is enough to inform assortment and pricing decisions and it is not enough to claim individual switching behaviour.
Presenting it at that level of claim keeps it credible. Over-claiming from aggregate data is how a genuinely useful analysis gets discredited.
How should this be used in a range review?
To identify which items are genuinely differentiated and which are interchangeable. An item whose removal is fully absorbed by another is a delisting candidate; one whose removal loses category volume is not.
That is the most commercially valuable output of the whole exercise and it is answerable from aggregate data.
Frequently asked questions
Is basket analysis useless here?
Not useless — it answers complementarity well and substitution poorly. Use it for the question it fits.
How many stockouts are needed?
Enough that the effect is not driven by a handful of stores. Pooling across many events is what makes the estimate stable.
Can I plan a stockout to test substitution?
Deliberately withholding an item is a real cost and an ethical question with the retailer. Where a test is genuinely wanted, a controlled assortment test is the honest version.
Does this work across brands?
Yes, and cross-brand substitution is usually the point. The method does not care whose item is absent.
What about pack-size substitution?
It is one of the cleanest cases, because sizes within a line are close substitutes and stockouts are common. It is a good place to start.
Does seasonality confound it?
It can, if the stockout coincides with a seasonal shift. Comparing against stores without the stockout in the same period controls for it.
Is loyalty data worth buying for this?
Where it exists for your channel and the question is high-value, possibly. In a channel where it largely does not exist, the question is moot.
Should the substitution map be shared with retailers?
Often it is the most useful thing a brand can bring to a range conversation, because it speaks to category outcome rather than to brand preference.
What is the honest caveat to attach?
That this is aggregate substitution inferred from natural variation, not observed individual switching. Standard treatments of market basket analysis set out what the method can and cannot support.
How should the output be validated?
Against a second source where one exists — a panel read, a qualitative study, or the field team's own view. Agreement raises confidence and disagreement locates an assumption worth examining.
### What should a brand actually do with this?
Use it where the decision is about range rather than about price. Substitution evidence is strongest on the question of whether an item can be removed without losing category volume, which is exactly the question a range review asks.
Bring it to the retailer rather than keeping it internal. An analysis that speaks to category outcome is the kind of contribution that changes how a brand is treated in a review, and substitution evidence is rare enough to be noticed.
Working with retail data? See NRS Insights data solutions or contact NRS Insights.