Measurement foundations and trustworthy sport data · Lesson 4
Segmenting people, visits and orders
11 min read
Lesson objectives
Separate buyer share from order frequency and limit disclosure of narrow segments.
Reading tools
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
A club team wants to know whether season members buy merchandise more often than single-match buyers. The comparison looks simple until you inspect the tables: one person can place several orders and attend several matches, while ticketing and retail systems may assign different identifiers to the same person. Choose the observation unit before segmenting. If the question is “What share of people bought?”, count each eligible person once during the window. If it is “How many orders per visit?”, visits or admissions form the denominator under an explicit linkage rule. Orders divided by people cannot be relabeled the “buyer rate.”
Construct a people table with identifiers the organization is permitted to use for this purpose. To measure buyer share over 30 days, define eligible people at the start of or during the period in a consistent way, then mark each person who completed at least one purchase. Keep returns separate. The number of people who placed an order and the number who retained a product may differ. If identity matching is unreliable, report separate platform records rather than inventing unique people from similar names or phone numbers.
A second problem appears in very small segments. A public dashboard showing that just three people in a narrow age and seating group bought a sensitive product may let a reader infer identity when combined with other information. Removing a name from the table alone does not settle the risk. Combine categories or restrict detail, and assess re-identification risk in light of who will see the report and what other information they may possess. This is a data-protection practice, not a universal numerical threshold. The UK Information Commissioner's Office guidance explains why context and other available data matter to effective anonymisation; its legal application has its own jurisdiction, but the analytical question is useful whenever a team shares a report.
Keep account, beneficiary and order identifiers separate when appropriately available. Preserve their relationships instead of combining them in one field. Check their direction: one account may have several orders, while one order may contain tickets for several beneficiaries. Understand this structure before calculating rates. Label numerator and denominator units beside every average. If the denominator changes between months, rebuild the comparison on a common unit or explain why the new figures answer different questions.
Select an order state appropriate to the time question. A purchase-conversion study may count confirmation on creation day, while retained-customer analysis waits for a defined return window. Keep both views when decisions require them. Inspect partial returns too: a customer may return one item and keep another. Fully refunded orders differ from orders with any refund. Define the rule before segmenting, so teams understand why shop and campaign reports can differ despite sharing source records.
Segments also change over time. An account may become a season member during the month. Choose a rule, such as status at study entry, and retain change dates separately when studying original groups. A description of today's customer base instead uses current status. Both views can be useful if labelled. Include a transitioning account in the test data and confirm that numerator and denominator use the same chosen classification rule.
Lesson summary
Buyer share uses eligible people while order frequency uses buyers and orders. Small segments require identifiability review before disclosure.