Why What You Bought Last Month Predicts What You’ll Buy Next

If you want to know what someone will buy next month, the most reliable predictor is what they bought last month. Not what they say they value. Not their demographic profile. Not their stated intentions. Their actual, documented purchase history. This is one of the most replicated findings in behavioral science, and it highlights the gap between stated and revealed preference. Yet most product teams still build their customer understanding on surveys, interviews, and demographic segments rather than on the transaction data that would tell them more.

Why Purchase Patterns Are Stable

Most purchasing decisions are not deliberate evaluations. They are habits. People develop routines around what they buy, where they buy it, how much they spend, and how often they repurchase. These patterns are stable because they are driven by embedded preferences, lifestyle constraints, and cognitive shortcuts that change slowly.

A consumer who has bought premium coffee beans every two weeks for the past year will, with high probability, buy premium coffee beans in two weeks. Not because they have thought carefully about it, but because the decision is no longer a decision. It is a default. Understanding these defaults tells you more about likely future behavior than any attitude survey could, because you are observing the outcome of real trade-offs rather than asking someone to imagine hypothetical ones.

What Purchase Data Shows That Surveys Cannot

Surveys ask people to introspect about their preferences, and people are unreliable introspectors. They overstate price sensitivity because it seems rational to care about value. They understate impulse buying because it seems irrational. They claim to prioritize quality while their purchase history shows consistent selection of the cheapest option. This is not a minor methodological nuisance. It is a systematic distortion that affects nearly every consumer category.

Purchase data records what happened, not what someone thinks happened or wishes had happened. From transaction records you can extract: the price range within which a consumer actually buys, the brands and products they switch between, how often they purchase in a category, and what share of their spending goes where. Each of these dimensions is more predictive of future behavior than its survey-based equivalent because it reflects real choices made under real constraints.

Price sensitivity is a useful example. Surveys routinely find that a large majority of consumers say price is their most important consideration. Purchase data tells a different story: a much smaller share consistently chooses the cheapest option. The rest say they care about price but repeatedly pay premiums for convenience, brand familiarity, or habit. The survey gives you an inflated number. The purchase data gives you a segmentation you can act on.

Switching Patterns Reveal Market Structure

One of the most valuable signals in purchase data is how consumers move between brands. Within any category, buyers range from fiercely loyal to highly promiscuous. Loyal buyers repurchase the same brand consistently. Promiscuous buyers rotate through a repertoire, driven by deals, availability, or novelty.

This distinction shapes product strategy directly. If you are launching into a category dominated by loyal buyers, your challenge is breaking established habits; that requires a meaningfully better product or a compelling reason to switch. If the category is characterized by repertoire buying, consumers already rotate between options and are more open to trying something new. Your challenge is getting into the consideration set, not displacing an incumbent.

Purchase data surfaces these patterns without requiring anyone to describe their own behavior. You can see how often consumers switch, what appears to trigger switches (promotions, new product launches, stockouts), and which direction switches flow. If consumers consistently move from Brand A to Brand B but rarely the reverse, that reveals something about relative value perception that no survey would surface as cleanly.

Revealed Preference as a Research Foundation

Economists call this “revealed preference”: when a consumer chooses one option over available alternatives, they reveal that they prefer it given their constraints. The choice itself is the data. No self-report required.

Every purchase is a small experiment. When someone buys a $5 sandwich instead of a $3 sandwich from the same shop, they have revealed that the perceived quality difference is worth at least $2 to them. Aggregate these micro-decisions across thousands of consumers and you build a detailed picture of how a market actually values different product attributes, without asking anyone a question. This is why companies sitting on large-scale transaction data (retailers, payment processors, subscription platforms) have a structural advantage in understanding their markets. The insights available in that data are different in kind from what survey research produces.

Why This Matters for Synthetic Research

The predictive power of purchase history is what makes behavior-grounded synthetic research fundamentally different from asking an AI to role-play as a consumer. When a synthetic persona is calibrated against real transaction data, its responses are anchored in revealed behavior: documented spending patterns, price tolerance, category loyalty, brand switching. An ungrounded persona generates opinions. A grounded one reasons from a behavioral record.

Digital advertising has moved in the same direction. Behavioral targeting, showing ads to people based on what they have bought or browsed, consistently outperforms demographic targeting. This is the logic behind behavioral segmentation. A 28-year-old woman and a 55-year-old man who both regularly buy premium running shoes are more alike, for the purpose of selling running products, than two 28-year-old women where one runs marathons and the other has never owned a pair. Purchase behavior cuts across demographics to reveal actual preferences. The same logic applies to research panels.

The Practical Implication

If your customer research is built primarily on what people say, you are working with the weaker signal. Stated preferences, survey responses, and interview quotes all have value, but they should supplement behavioral data, not substitute for it. Ground your understanding of your market in what consumers actually buy: how much they spend, how often, which products they switch between, and what price points they consistently accept. This behavioral foundation is also essential for finding your ideal customer profile.

Past purchases are not a perfect predictor. People change habits, try new things, and respond to genuinely innovative products. But the base rate of behavioral consistency is high enough that any research approach that ignores purchase data is leaving its most powerful input unused. The question is not whether to use behavioral data. It is whether you can afford not to.