Two shoppers wearing winter clothing inside a store.

Purchase Intent vs Purchase Behavior: what each signal can prove in retail media

A shopper can say that sustainability matters and still buy the cheaper pack. They can intend to try a new brand and forget it in store. They can report no interest in a category, then buy it because a child asked, the usual product was unavailable or a promotion changed the decision.

The survey is not necessarily wrong. The transaction is not the whole truth either.

Purchase intent records what people say they expect, prefer or plan to do. Purchase behavior records what they did under the price, assortment, time, household and channel conditions that existed at the decision. Retail media uses both signals, but they answer different questions.

What stated intent is good for

Research can reveal information that a basket cannot contain:

the need or occasion behind a purchase;
brands considered but not bought;
barriers such as price, trust or confusion;
attitudes toward claims, packaging and creative;
future plans that have not yet produced behavior;
why a shopper rejected an option.

Intent is especially useful early in a category, before enough transactions exist, or when the decision is infrequent. It can also help interpret a behavioral pattern. A mixed basket may suggest a mission, but research can test whether shoppers recognize that mission and what language they use for it.

The limitation is that stated plans do not always become action. A meta-analysis of experiments that changed intentions found a smaller change in subsequent behavior.[1] The review spans health-related behaviors rather than retail purchasing, so it supports the general existence of an intention-behavior gap, not a universal estimate for shopper conversion.

What observed behavior is good for

Transaction data records an outcome with real consequences. It can show:

which products were purchased together;
frequency, recency and switching;
response under actual price and promotion;
differences across stores and channels;
repeat behavior after trial;
whether an exposed shopper later bought.

This makes behavior useful for audience qualification and outcome measurement. For near-term category eligibility, a recent observed purchase can often be validated more directly against the target outcome than a broad statement of interest. That is practitioner guidance, not a rule for every category or time horizon.

Behavior also has limits. A sale does not reveal every considered option. It may reflect stock, distribution, budget or household constraints. It may come from another person in the household. A loyalty identifier may cover only part of the shopper's spending. Observed purchase is evidence of what happened, not a complete explanation of preference.

Four different questions that teams often mix together

1. What does the shopper say they want?

Use surveys, interviews, search language and qualitative work.

2. What did the shopper do?

Use transactions, visits, browsing and basket data, with clear identity and coverage limits.

3. What is the shopper likely to do next?

Use a predictive model built for a defined event and time horizon. Past behavior can be an input, but a probability remains an estimate.

4. What happened because of advertising?

Use an experiment or credible counterfactual. Neither high intent nor a later purchase proves that media caused the outcome. Incrementality guidance from IAB and IAB Europe separates attributed outcomes from causal impact.[4] Closed-loop attribution can connect exposure to sales, but linkage alone is not causal proof.

Keeping these questions separate prevents a common reporting error: using one data source to answer a question it was not designed to answer.

A worked example: a plant-based product launch

Pre-launch research finds strong interest in reducing meat consumption. The brand is tempted to target everyone who agrees with that statement.

A better design combines evidence.

First, research identifies motives and barriers. Some shoppers want health benefits, some want variety, and some worry about taste or price.

Second, retailer behavior identifies practical audiences: shoppers buying adjacent plant-based products, shoppers alternating between conventional and plant-based options, and shoppers with baskets that fit the intended meal. Basket influence analysis can help build these hypotheses without claiming to reveal motive.

Third, creative maps to the barrier. Taste reassurance should not be shown merely because someone bought vegetables. Price communication should not be used merely because someone agreed that value matters.

Fourth, the launch holds out eligible shoppers or stores. The campaign measures first purchase, repeat, category expansion and substitution. It does not call a survey shift or an attributed sale incremental growth.

The intent data informs the message. The behavior data improves eligibility. The experiment estimates effect. Omnichannel journey planning then decides where each signal can change a decision.

A decision guide for signal selection

Freshness is category-specific. The guidance below provides operating starting points to validate, not universal standards.

What barrier should creative address?

Preferred signal: Survey or qualitative research.
Minimum freshness or coverage: Current enough to reflect the offer and market.
Known bias: Stated preference and sampling bias.
Activation use: Message development.
Proof method: Creative test plus later behavior.

Who is eligible for a near-term category campaign?

Preferred signal: Recent observed category behavior.
Minimum freshness or coverage: At least one credible purchase cycle and known channel coverage.
Known bias: Missing cash, household and competitor purchases.
Activation use: Audience eligibility.
Proof method: Holdout-based conversion lift.

What happens in a sparse or new category?

Preferred signal: Intent plus contextual signals.
Minimum freshness or coverage: Adequate sample and documented transaction gaps.
Known bias: Intent may not become action.
Activation use: Early audience and message hypotheses.
Proof method: Validate against later purchases.

What happens next?

Preferred signal: Predictive score.
Minimum freshness or coverage: Recalibrated for the event and horizon.
Known bias: Model drift and label bias.
Activation use: Ranked eligibility.
Proof method: Out-of-time validation and campaign holdout.

What did advertising change?

Preferred signal: Experiment or credible counterfactual.
Minimum freshness or coverage: Predefined assignment and outcome window.
Known bias: Noncompliance, spillover and attrition.
Activation use: Budget and effectiveness decision.
Proof method: Incremental effect with uncertainty.

Intent can override weak transaction coverage when a category is new, infrequent or poorly observed. Behavior can override broad stated interest when the decision is immediate eligibility for a well-observed repeat category. In both cases, the team should state the missing evidence.

When baskets deserve more weight

Retail media should generally give observed behavior more weight when the decision requires:

recent category eligibility;
purchase cadence;
brand switching evidence;
product attachment;
repeat purchase;
sales outcome measurement.

That does not mean "baskets over people." It means matching evidence to the decision.

A survey can be more valuable when the team needs to understand a new need, diagnose a barrier, test language or explore a market where transaction coverage is weak. Intent and behavior should be combined when each contributes something the other cannot.

Data quality and privacy still matter

Behavioral data can feel objective because it is recorded automatically. It is still produced by a system with blind spots.

Teams should document identity coverage, household sharing, cash transactions, returns, channel gaps, product mapping and observation windows. A retailer should not infer a sensitive condition simply because a basket pattern makes the inference possible.

As a UK example, the Information Commissioner's Office treats behavioral prediction for direct marketing as profiling and emphasizes fairness, transparency, proportionality, accuracy and objection rights.[2] Exact obligations depend on market and implementation, so legal and privacy review is required before activation.

How to measure without overstating the evidence

The IAB/MRC guidelines call for clear data sources, methods and outcome definitions in retail media reporting.[3] For intent and behavior work, a report should state:

whether a measure is stated, observed, predicted or experimental;
the population and time window;
how shoppers were identified;
what channels are missing;
whether the metric is attributed or incremental;
whether price, promotion and availability changed;
which claims are descriptive and which are causal.

A brand-lift result can show a change in awareness or intent. A sales result can show a purchase among exposed shoppers. A controlled test can estimate what changed because of the campaign. These are related outcomes, not interchangeable ones.

Common mistakes

Treating an attitude as an audience. Agreement with a statement may be too broad or too old to guide activation.

Treating a purchase as a stable preference. One basket may reflect a deal, a stockout or another household member.

Using claimed behavior as observed behavior. "I bought this last month" remains self-report.

Calling prediction fact. A propensity score estimates a future event; it does not record it.

Calling attributed sales causal. Buyers selected from transaction data may already be more likely to purchase.

Collecting more data than the decision needs. More fields can increase privacy risk without improving the campaign.

Buyer checklist

Ask these questions before approving an audience or result:

Is this signal stated, observed, predicted or experimental?
What exact decision will it change?
How recent is it?
What population and channels does it cover?
Could availability, price or promotion explain the behavior?
Is the signal being used to infer something sensitive?
What comparison will estimate campaign impact?
Which result would change the next action?

FAQ

Is purchase behavior always more reliable than intent?

No. Behavior is stronger for what happened in the observed system. Intent can be stronger for motives, consideration, barriers and unobserved future plans.

Can survey intent predict sales?

It can contribute information, but intention does not convert perfectly into behavior. Prediction should be validated against later outcomes for the specific category and context.

Are baskets free from bias?

No. Baskets reflect assortment, availability, price, promotions, identity coverage and household behavior.

Should retail media stop using brand-lift studies?

No. Brand lift answers upper-funnel questions that transaction data cannot. It should not be presented as sales incrementality.

What is the best combined approach?

Use research to understand the need and message, behavior to define eligible audiences and outcomes, and controlled tests to estimate causal impact.

Bottom line

Intent explains what shoppers say, expect and consider. Behavior records what they chose under real conditions. Prediction estimates what may happen next. Experiments estimate what advertising changed.

Retail media gets better when those four forms of evidence keep their own names.

Sources

[1] Rhodes and Dickau, Experimental evidence for the intention-behavior relationship in the physical activity domain: a meta-analysis.

[2] UK Information Commissioner's Office, Collect information and generate leads.

[3] IAB and Media Rating Council, Retail Media Measurement Guidelines.

[4] IAB and IAB Europe, Guidelines for Incremental Measurement in Commerce Media.

Ready to see how this works in practice?

Footprints AI helps brands and retailers measure what matters. See our customer success stories or get in touch to discuss your retail media strategy.