Feature and display: inferring support from data
Transaction data records what sold and at what price. It does not record whether an item was on an end cap, in a flyer, or on a shipper by the door. In large-format retail those facts are collected separately through audits or retailer feeds; in the independent channel they are frequently unavailable.
That gap matters, because support is one of the main explanations for why a week was unusual.
What can be inferred from price alone?
A price change, and only that. A week where the observed price falls indicates a price promotion; it says nothing about whether the item was displayed, featured, or simply repriced.
Two executions with identical price changes and very different support will show identical price data and very different volume, which is the whole reason inference is limited.
What signals hint at display?
A volume lift much larger than the price change would predict, given a reasonable elasticity. That discrepancy suggests something other than price was operating, and display is the usual candidate.
It is a hint rather than a measurement. A distribution gain, a competitor stockout or a seasonal event would produce the same signature, so the inference needs corroboration before it is stated as a finding.
Why does the independent channel differ?
Because there is no central promotional calendar. A large chain plans support centrally and can supply the data; an independent store decides locally, and thousands of local decisions have no single record.
That is a genuine limitation and it should be stated plainly by anyone reporting on the channel. It is also why store-level execution questions in this channel are usually answered by fieldwork rather than by data.
What should a brand do about it?
Collect what it can at the point of execution. A field team, a photograph, a signed confirmation — any record made at the time is worth more than an inference made afterwards.
Where fieldwork is not feasible, the honest position is that support is unobserved and the analysis carries that caveat. A model that assumes uniform support across stores is making an assumption that is certainly false.
Does modelling help?
It can separate price effects from unexplained residual effects, which is useful and is not the same as measuring display. A residual is a measurement of ignorance, and labelling it "display" converts an unknown into a claim.
The careful version reports the residual as a residual and names display as a plausible contributor. That distinction survives scrutiny; the confident version does not.
How should support be reported when it is unknown?
As unknown, with the consequence stated: that the effect estimated may include support the analysis cannot see. Readers can weigh a stated limitation and cannot weigh a hidden one.
A provider that names what its data does not cover is telling you it understands its data, which is a better signal than a report with no caveats at all.
What does this mean for comparing executions?
That two promotions with similar price changes and different results are not evidence about price. They may be evidence about execution, and without support data there is no way to tell.
Comparing across a large number of executions helps, because unobserved support averages out. Comparing two does not.
Frequently asked questions
Can I buy display data for the independent channel?
Coverage varies and is generally thinner than in large-format retail. Ask a provider specifically what is observed, how, and at what frequency rather than assuming it exists.
Does a price drop always mean a promotion?
No. It can be a list price change, a mix shift within an item's stores, or a data artefact. A sustained fall is more likely a price change; a single week is more likely a promotion.
How do I detect a distribution-driven lift?
Check the store count alongside the volume. A lift accompanied by more selling stores is a distribution effect until shown otherwise.
What is the value of photographs?
They are the cheapest execution evidence available and they are time-stamped. For a brand paying for support, a photograph is what turns a claim into a record.
Should unexplained lift be reported?
Yes, as unexplained. That is a legitimate and useful finding, and it prompts the right next question rather than closing it prematurely.
Does this affect elasticity estimates?
Directly. Elasticity estimated from promoted periods with unobserved display will attribute display's effect to price and overstate sensitivity.
What about competitor support?
Equally unobserved and equally capable of explaining your week. Symmetry matters here: an unobserved factor that could flatter you could also have hurt you.
What is the honest one-line summary?
Scan data measures outcomes precisely and inputs poorly, and a report should be explicit about which it is describing. Industry context on the channel is collected by NACS.
### What does a brand lose by ignoring this?
Misattributed learning. A price test contaminated by unobserved display teaches the wrong elasticity, and that number then informs pricing decisions for years.
The cost is not the single analysis; it is everything built on it afterwards. That is the argument for collecting execution evidence even when it is inconvenient.
How much execution evidence is enough?
Enough to characterise the range rather than to audit every store. A sample of executions, photographed and dated, tells you whether support was broadly delivered, and that is usually the question a model needs answered.
Working with retail data? See NRS Insights data solutions or contact NRS Insights.