Why purchase data is so attractive
Surveys ask people what they do. Transaction records show what they paid for. That difference is why sales data is popular with analysts: nobody has to remember or report anything, the records are time-stamped, and the volumes can be very large. Retailers, payment providers, banks and loyalty schemes all generate this kind of information as a by-product of ordinary business.
Because it records actions rather than intentions, it can avoid some of the problems of self-reported behaviour. A person may say they buy more fresh produce than they do, but a till receipt does not care about flattering answers. That strength is real. The limits are just as real, and they tend to be less visible.
What transaction records can reveal
Used carefully, purchase data can often describe:
- Timing and seasonality. When purchases rise and fall across days, weeks and years.
- Mix. Which categories, sizes or price tiers are gaining or losing share within a store or platform.
- Frequency. How often the same customers return, if records can be linked to a loyalty account or payment card.
- Geography. Where, broadly, purchases occur, depending on how the data was collected.
- Responses to events. How spending changes around holidays, price changes or promotions.
For habits that run through payment systems, such as cashless payments or online shopping, the records can be a natural source. They can show whether a behaviour is growing within a defined set of customers and how fast, which is hard to learn from recollection.
What they cannot say about motives
A purchase is an outcome with many possible causes. Someone buying oat milk might prefer the taste, be catering for a visitor, follow a medical recommendation, respond to a discount, or buy it for someone else. The record shows only the item and the price. Stories that read a motive into a sales line, such as “shoppers are choosing ethical brands”, go beyond what the data holds. Motives require other evidence, such as interviews or carefully worded questions, and even those have their own limits.
A change in purchases can also follow from a change in what is on offer. If a shop widens its range, moves a product to eye level or runs a promotion, sales may rise without any shift in consumer taste. The data records behaviour inside a store designed by someone, not behaviour in a neutral setting.
Who is not in the data
Every dataset has a boundary, and sales records have several.
- Non-buyers. The data shows people who bought. It says little about people who considered the product and declined, or who never encountered it. A rise in sales to existing customers can coexist with no change in the wider population.
- Other channels. One retailer, bank or payment app covers only its own customers. Cash, other retailers, markets, home production and secondhand purchases may all sit outside the frame. Growth at one company may mean customers moved from another rather than that demand grew.
- Unrepresentative customers. The people who use a particular card, app or loyalty scheme are typically not a cross-section of everyone. Their age, income and location may differ from the wider public.
- Gifts and shared use. A purchase may be made by one person for a household, so individual habits are hard to read from household spending.
- Free or informal exchange. Things passed on, borrowed or shared produce no transaction.
This is a sampling problem in a different form: the question of who is on the list applies to commercial data just as it does to surveys, a theme developed under sampling frames.
Counting sales versus counting people
Sales totals can rise because more people buy, because the same people buy more, or because prices went up. These are different stories. A figure in money terms can also be distorted by inflation, while a figure in units may hide a shift to larger or smaller pack sizes. Reading a trend from sales therefore often requires knowing whether the figure counts customers, items or money, and whether it has been adjusted for price changes.
Similarly, an increase in the number of orders on a delivery app does not by itself show that more households use food delivery. It may reflect a smaller group ordering more often. Without customer-level information, the two cannot be distinguished.
Practical caveats about access and definitions
Much transaction data is owned by companies, and outsiders typically see only summaries chosen for publication. Category definitions are set by the data holder, so “healthy snacks” or “sustainable” may mean different things in different datasets. Companies may also have reasons to highlight favourable figures. Where the data is shared as a commercial report, the method is often only partly described, which makes independent checking difficult.
Markets for resale and secondhand goods illustrate the issue: many sales happen between individuals or across small platforms, and no single record captures them. A trend in resale markets measured from one platform’s sales may understate or overstate the whole.
A balanced way to use it
Transaction records are strong evidence about what happened inside a defined set of purchases, and weak evidence about why or about people outside it. They pair well with other sources: surveys can ask about reasons, and independent datasets can show whether the pattern appears outside one company. When two quite different methods point in the same direction, a trend claim becomes more credible. When only one company’s sales figures support it, caution is usually sensible.