Repeat rate and the repeat window
Repeat purchase is the single most informative thing about a new product, because trial can be bought and repeat cannot. The measurement is straightforward where shoppers are identified and indirect where they are not, which is the situation in most of independent retail.
What survives without identifiers is a store-level proxy, and it is more useful than it sounds.
What does the store-level proxy look like?
Sustained velocity after trial decays. An item whose sales per stocking store settle at a level well above zero and hold there is being bought again by somebody, even if the data cannot say by whom.
That is a weaker claim than a household repeat rate and it answers the commercial question: does this item keep selling once the novelty passes. For a range decision, that is the question.
Why does the window matter so much?
Because a repeat window shorter than the category's purchase cycle records a non-repeat for a shopper who simply has not returned yet. A window that is too long includes people who would have bought anyway and inflates the figure.
The window should come from the category's own purchase frequency rather than from convention. A fortnightly category and a quarterly one need different windows, and using one number across a portfolio produces incomparable figures.
How do you estimate the purchase cycle without identifiers?
From the category's velocity and the typical basket quantity, which gives an average interval at store level. It is a rough estimate and it is enough to set a defensible window rather than an arbitrary one.
Where a panel or identified data exists for a comparable channel, borrowing its cycle estimate is reasonable as long as the borrowing is disclosed and the channel difference is acknowledged.
What does a poor repeat signal look like?
Velocity that decays toward zero after the trial period, across many stores, with no distribution explanation. That pattern is about as clear as commercial evidence gets, and it usually arrives earlier than anybody wants to act on it.
A single store showing it is noise. Twenty stores showing the same shape is a finding, and the shape is visible before the total sales figure turns.
What confounds it?
Distribution changes, promotional support that continues past launch, and seasonal drift. Each can hold velocity up while genuine repeat is weak, and each should be controlled before drawing a conclusion.
Support is the most common. An item on continuous promotion has not yet demonstrated repeat at full price, and a range decision taken on supported velocity is a decision about the support.
How should it be reported?
As velocity per stocking store over time, with support periods marked and the trial window shaded. One chart, honestly annotated, carries the whole argument.
Adding a line for the category's typical settled velocity gives the reader a benchmark, and a benchmark is what turns a chart into a judgment.
Frequently asked questions
Can repeat rate be calculated at all without loyalty data?
Not as a household rate. What can be calculated is persistence of velocity, which answers the commercial question about whether an item keeps selling without answering the behavioural question about who is buying it. Being explicit about which one you have produced is the difference between a defensible finding and an overreach.
How long after launch should repeat be assessed?
After at least two purchase cycles for the category, so that a shopper has had the opportunity to return and the trial effect has decayed. In a frequently bought category that may be a matter of weeks; in an infrequent one it can be several months, and applying the same date to both produces two incomparable assessments.
Does promotional support invalidate the measure?
It postpones it. Velocity under support tells you what the item does with support, which is useful for planning and is not evidence about unsupported repeat. Measure a clean period before making a range decision, and if no clean period exists, say that rather than reporting the supported figure as though it were one.
How does seasonality interfere?
A launch into a rising season shows velocity holding up for reasons unrelated to repeat, and a launch into a falling one shows decay that is not a repeat failure. Comparing against the category's own seasonal shape in the same stores is the correction, and it is usually available.
Is a high settled velocity always good news?
It is good news about the item and it does not settle the commercial question, which also involves margin, cannibalisation and the space the item occupies. An item with strong repeat that displaces a stronger item has still made the shelf worse.
What is the single most useful output?
Velocity per stocking store, plotted from launch, with support marked. It is one chart, it distinguishes distribution from demand, and it answers the question the launch was funded to answer. General context on retail measurement practice is collected by FMI.
### How does this inform a delist decision?
Directly. An item whose settled velocity sits below the category's working threshold, with distribution stable and no support running, has demonstrated what it is going to do. Waiting longer collects more of the same evidence.
The discipline is to agree the threshold before launch. A threshold negotiated after a disappointing settled velocity is a threshold chosen to protect the decision that created the item.
What should be reported at each review point?
Distribution, velocity per stocking store, brand total in stocking stores, and support. Four numbers, the same four each time, so that consecutive reviews are comparable rather than each being a fresh argument.
### What should a brand actually do with this?
Build the two-line ramp chart as a standing output for every launch rather than as a one-off for the ones that attract attention. A standing format means the awkward launches get the same scrutiny as the successful ones, which is where most of the learning sits.
Then keep the charts. A library of past ramps is the only realistic benchmark a brand has for what normal looks like in its own categories and its own channel, and it is considerably more relevant than a general figure borrowed from elsewhere. Channel context is collected by NACS.
Working with retail data? See NRS Insights data solutions or contact NRS Insights.