Attention, preference and purchase need different tests

In brief

Record what a person looked at, what they chose and what they purchased as separate outcomes. To estimate additional sales caused by an intervention, define the comparison that represents what would otherwise have happened. Your team's dashboard can display all these measures, but it should preserve the question behind each one.

What the measurement debate changes

The IAB guide is a draft open for public comment. Public comment is open through 4 November 2026. It distinguishes methods such as marketing mix modelling, incrementality testing and attribution, including cases where their findings conflict. The useful implication for a research brief is to specify the decision before choosing a metric.

An attention study might help a designer locate an overlooked label. A preference task might compare two packaging alternatives under defined conditions. A sales experiment might assess whether a campaign generated additional purchases. Those are different commissions. Calling all of them effectiveness makes the evidence harder for a client to interpret.

What the research can tell a brand team

In a 2010 eye-tracking study, Ian Krajbich, Carrie Armel and Antonio Rangel modelled relationships between visual fixation and binary choice. The inspected primary abstract describes a model tested in human choice tasks. It supports studying the relationship between looking and choosing in that setting; it does not establish that a gaze heat map measures a customer's private intention or later retail spending. The record also links an erratum, so we use no detailed numerical estimate from this abstract.

A separate question is whether observed advertising exposure estimates a causal effect. Gordon and colleagues' 2019 study compared observational estimates with randomised experiments at Facebook. It covered 15 US experiments, 500 million user-experiment observations and 1.6 billion ad impressions. Observational methods often failed to recover the experimental estimates, even with extensive measured customer information.

The observation count is not a count of unique people. Nor does this US platform study quantify the error in a Caribbean campaign. Its relevance is the comparison of measurement methods: knowing who saw an advertisement is insufficient to establish what the advertisement caused.

Lewis and Rao's 2015 published abstract reports 25 large field experiments involving US retailers and brokerages and $2.8 million in digital advertising spending. The median confidence interval for return on investment was wider than 100 percentage points. It illustrates how uncertain a causal estimate can remain even in a large experiment. It is not an expected return or uncertainty estimate for Orbital.

Together, these inspected sources support separating the measurement tasks and checking the design. This is a focused narrative synthesis, not a systematic review or a new meta-analysis. We have not extracted compatible effect sizes and variances or pooled the studies. The laboratory choice task and advertising field experiments would not become a common outcome merely by placing their figures on the same chart.

Write the outcome beside the instrument

For your eye-tracking study, specify the task, image or display, viewing conditions and area being measured. A fixation measure can answer whether attention reached a location under those conditions. If the commercial question concerns whether people understood a price or an offer, add a direct comprehension measure. Looking at a number does not by itself establish understanding.

For your preference task, describe the available alternatives and whether a choice has a real consequence. An answer about which design someone likes and a purchase with their own money need different interpretation. Keep your task wording and product context with the result so a later presentation does not remove the conditions that made the finding meaningful.

For your purchase study, define the unit of assignment and the outcome window. If the aim is incremental sales, the analysis needs a credible estimate of purchases without the intervention. Random assignment can supply that comparison when it is feasible and correctly implemented. Where it is unavailable, record the assumptions required by the observational method and the reasons they may fail.

A proposed Orbital assistant could organise these fields before drafting a client summary. It would retrieve the study question, design, outcome and limits, then flag a claim that goes beyond them. This describes a possible research workflow. It does not claim that a released assistant has already passed a client evaluation.

What this means for Trinidad and Tobago brands

For a hypothetical Trinidad and Tobago event sponsor, start with the decision the evidence must support. A visible logo in a Carnival photograph establishes the presence of a logo in that photograph. It does not establish audience recall, product preference or additional sales. The credited festival image provides local context and depicts no Orbital study or endorsement.

Your Caribbean research brief could ask whether a revised offer is understood by the intended audience. Define your audience's eligibility before recruitment and record the language and context of the task. If a later commercial study concerns sales, plan how permitted transaction records could be linked to the assigned intervention without exposing customer identities.

Apply Adrian Dunkley's People First, Problem Second, AI Last approach. The person buying the research needs evidence for a business decision; a consumer needs a clear and honest offer. Define the decision and the misunderstanding being investigated before selecting eye-tracking, a survey or an AI model.

The official tourism page dates Trinidad Carnival 2027 to 8 and 9 February. For a proposed Carnival campaign study, that event provides a concrete time boundary. Record whether a response was collected before, during or after those dates. Distinguish residents from visitors through a permitted, self-reported field when that distinction serves the research question. Pre-event costume choice and an on-the-day purchase answer different questions; combining them can obscure what a sponsor needs to learn.

A measurement audit this week

You can test the clarity of the evidence brief before commissioning another dashboard. The following is our editorial proposal. It checks documentation and comprehension; it cannot establish campaign lift or determine a scientifically adequate sample for a sales experiment.

1. Write the decision and its primary outcome. Replace a broad instruction such as measure engagement with the exact observable result the decision requires.
2. Audit 10 existing campaign claims by 3 November 2026. For each, record the measured outcome, population, comparison and time window. Count claims that lack one of those fields.
3. Ask 5 permitted volunteer reviewers to read the same offer and explain its price and conditions. Record misunderstandings verbatim and avoid collecting personal information the check does not need.
4. Compare the claims with their source records. Withdraw a causal sales claim when only exposure or clicks were observed, and plan an appropriate comparison before making it again.

For your future experiment, predefine the main analysis and meaningful effect before reviewing outcomes. Keep a distinction between checking whether data were collected correctly and checking whether a commercial claim succeeded. Repeatedly changing the outcome after seeing results can leave the published finding disconnected from the original decision.

Risks in inferring what a customer wants

Behavioural data can be presented with more psychological certainty than the study supports. Consumers bear a privacy cost when detailed observations are retained without a clear purpose; clients bear a commercial cost when a weak measure becomes a confident sales claim. Limit the fields collected, explain the task and preserve the right to stop participation. Do not describe an individual as having an intention that the study never measured.

Local applicability remains untested. None of these sources establishes a Trinidad and Tobago sales effect for Orbital. A local study would need its own recruitment, comparison, analysis and uncertainty. A brand manager can use the commissioning brief to identify the decision, then choose a measurement capable of answering it.

Keep the units with the evidence

The quantities have different units and must not be combined into a performance score. None is an Orbital result.