What does last purchase date (R) mean in a target group profile?

A customer who purchased thirteen days ago has a different purchase history from one whose last recorded purchase was months ago. Without a product category, reference date and reliable history, however, that difference says little about the next decision. Last purchase date (R) records how recent a relevant purchase is and the limits of what you can infer from it.

Last purchase date (R) records in the target group profile when the most recent relevant purchase occurred, within a defined customer, supplier or product context, with its source, reference date and the scope of the available history.

This is Neurofactor’s practical profile definition. The purchase date is a calendar date. Here, recency is the time between that date and a chosen reference date. An R score is an optional classification of that interval. Keep these three items separate.

An audience generally contains several buyers and therefore several last purchase dates. Summarise the data as a distribution or a carefully defined statistic, with counts and missing information. Do not assign an invented purchase date to an entire audience.

The first three entries form a fictional calculation example. A budget round provides purchase context; it is not a purchase date. A quote request or website visit may be relevant behaviour, but should not count as a purchase without changing the field’s meaning.

The unit of analysis must also be clear. An organisational purchase is not automatically a personal purchase by the marketing manager being interviewed. State whether the unit is a person, household, account or organisation.

R stands for recency within RFM: recency, frequency and monetary value. The framework belongs to the analysis of recorded customer behaviour. The sources below provide historical and scientific context; we do not attribute the invention of the entire model to a single publication.

Bult and Wansbeek studied direct-mail selection based on expected returns and costs. Their 1995 publication is an early scientific contribution to customer selection. Bult & Wansbeek, 1995.

Fader, Hardie and Lee connect RFM to a customer-value model. Recent purchasing is interpreted alongside other purchase information. Fader, Hardie & Lee, 2005.

Check the measurement convention: Fader and colleagues use tₓ for the last purchase’s time since the start; elapsed time is T − tₓ. Author manuscript.

We therefore explicitly write ‘days since last purchase’ in this profile. A low value means a recent purchase here. A score may run in the opposite direction. The documentation and research guidance below is our practical application, not a new validated RFM scale.

Set the scope before calculating. Otherwise, the same field may accidentally answer different questions.

For subscriptions, deliberately distinguish taking out a contract, renewal and automatic payment. An automatic debit may count as a financial transaction, but does not establish a new deliberate purchase decision. Retain contract status and transaction history separately when both matter to the question.

For the profile, the best scope is the one that answers the research question. State it in the profile sentence and supporting record so readers understand what the result covers.

1. Define the purchase. Record coverage, product category, unit and counting rules.
2. Choose one reference date. Use the same date for records being compared. For timestamps, also specify time zone and rounding.
3. Check the history. Investigate missing channels, duplicate accounts or a system migration that cut off older records.
4. Select valid purchases up to the reference date. When reconstructing a historical analysis, do not use future orders or information unavailable at that time.
5. Take the latest valid date for each unit. Keep missing and inapplicable cases separate.
6. Calculate the date difference. In this workflow: recency in calendar days = reference date − last purchase date.
7. Summarise the audience. For example, show a distribution across defined time bands, together with counts, coverage and the proportion unknown. Explain any exclusions.
8. Check the interpretation. Consider product context, frequency and purchase rhythm before attaching an action or prediction to the result.

If the dates are the same, the difference is zero days. That means a purchase on the reference date, not a missing purchase. Negative results require checking the selection, dates or time zone.

If you use classes or scores, record their boundaries, direction and reference group. A score of 5 might mean ‘most recent’, but that convention is not self-evident. Ranking customers within your own database is also not a comparison with the entire market.

A prospect who has not purchased from your organisation has no last purchase date there. That person may still have purchased elsewhere in the category recently. Record the two scopes separately.

Interview questions might include: ‘When did your organisation last purchase in this category?’, ‘Was that from this supplier or elsewhere?’ and ‘What are you basing that period on?’ Preserve the precision of the answer. ‘Not known’ is useful information about the limits of the research too.

Fictional example — no customer data or sector benchmark. On 23 September 2026, we examine paid, non-cancelled orders in one product category from one supplier. The unit is the customer account; available history starts on 1 January 2026.

The profile sentence could read: ‘For these four fictional accounts, two last purchase dates are known as of 23 September 2026, with recencies of 13 and 92 days; one account has missing data and one has no purchase in the available window.’

An average calculated from A and B would concern only those two known values. It must not be presented as the average for all four accounts. Nor is D automatically a former customer: the information does not establish whether the account purchased before January or was a customer at all.

No R score has been assigned and no purchase probability calculated. The table illustrates how to keep measurable and unmeasurable cases alongside each other.

An associative target group is a bounded group of people whose similar associations guide decisions within a specific choice context.

Recency describes purchase history, not the association pattern behind a decision. Two people can purchase on the same day after very different considerations. You can use purchase timing to describe research groups or discuss experiences surrounding a specific purchase, but investigate associations separately.

One possible question is: ‘Which associations with this supplier played a part when you decided to buy?’ Do not assume that a recent purchase establishes trust or that a long gap means resistance. Those could be hypotheses; the date itself does not provide that explanation.

Likewise, do not infer BIS, BAS or delay-discounting scores from purchase recency. Time since an event differs from sensitivity to approach, avoidance or delayed rewards.

• Omitting the reference date. The same purchase becomes less recent each day; a stored value needs a measurement date.
• Equating a recent purchase with readiness to buy. The field describes the past; a prediction needs further support.
• Showing known buyers only. Make missing and excluded records visible.
• Storing a budget cycle as a purchase date. Keep decision rhythm as separate context.
• Mixing dates, days and scores. Specify the unit and scoring rule for each result.
• Confusing purchase recency with a memory effect. The recency effect in memory is a different concept.
• Assigning fixed customer value or loyalty. One date is not a measured preference, motive or valuation.

With purchase data:

‘Within [category and supplier scope], [defined group] has a recency distribution of [result in days or time bands] as of [reference date], based on [source and counts], with [coverage gaps or exclusions].’

Without purchase data:

‘Last purchase date is [unknown/not applicable]; [source] does describe [reported period or purchase rhythm], retained separately and not treated as measured R.’

Behind the sentence, keep the event definition, unit of analysis, history, date precision, calculation rule and any score boundaries. A periodic update needs a new reference date; retain the previous position as a version. Correct an erroneous source date with provenance rather than silently overwriting earlier results.

Last purchase date becomes useful when you state which purchase counts, the level of measurement and the available history. Keep the date, elapsed time and any score separate. Purchase behaviour can then enrich the profile without filling data gaps with assumptions about people.