How a Research Agent Runs Packaging Research

A pack redesign is one of the most visible bets a brand makes. NIQ reports that redesigns recognized in its Design Impact Awards are associated with an average volume increase of approximately 4% across nearly a decade of winners. Those are award winners, not every redesign, which is exactly why brands test before they commit.

This guide shows how a research agent links shelf stand-out, emotion, and purchase intent into one evidence chain. If your team works in consumer insights, it also shows where researchers stay in control.

What Packaging Research With a Research Agent Means

Agent-led packaging research is a study in which the agent plans, fields, analyzes, and recommends. It turns a business question into hypotheses, places pack designs in a competitive shelf, collects attention, emotion, and choice data, and drafts a recommendation. An AI research agent does this end to end, with a researcher reviewing each stage.

That differs from an AI moderator, which only runs the interview. AI-moderated packaging testing is one part of the work, while the agent also designs the study and combines the signals.

The researcher still approves the hypotheses, the stimuli, and the final recommendation. Adoption of AI is already broad: a McKinsey survey of 1,491 participants in July 2024 found 71% of organizations using gen AI in at least one business function.

The Packaging Evidence Chain: Attention, Emotion, Then Intent

A pack must be seen, then felt, then chosen. Miss any stage and the pack fails, so each stage gets its own question and its own metrics.

Attention comes first because shoppers give it so little time. Lund University research on shelf behavior found that a pack has about 1.5 seconds to reach the shopper before they look away, and that shoppers hunting for a specific product scan no more than half the shelf range.

Any single metric misleads in isolation. A pack can win attention and still repel shoppers, or please them and go unnoticed. Stated answers add their own distortion, a pattern covered in the say-do gap in CPG research.

How the Agent Plans a Packaging Study

The agent starts with the business question: a redesign, a new SKU, a line extension, or a claim change. It converts the question into hypotheses and success thresholds, such as "the new pack must be found as fast as the current one".

It then selects the test design. A monadic design shows each shopper one pack. A sequential monadic design shows several, one after another, in rotated order. A current pack versus new routes design compares the incumbent against each redesign.

Success thresholds are set before fieldwork. Examples include findability at least equal to the current pack, valence above the category norm, and choice share above an agreed level. Agreeing them upfront stops the team from moving the goalposts once results arrive.

Finally, it sets the sample and quotas using category buyers, since people who never buy the category cannot judge a pack. Our guide to package testing covers the fundamentals behind these choices.

Building Shelf Stimuli and Competitive Sets

Realistic shelves make results transferable. The stimulus should reflect the facings, shelf position, adjacent competitors, and price tags a shopper would see, in the shape of a real planogram.

The agent selects which competitors appear in the competitive set, usually the category leaders plus the brand's closest substitutes. It rotates pack position across respondents so that a pack never benefits from sitting at eye level for everyone.

Measuring Shelf Stand-Out

Stand-out is about getting noticed. The agent tracks four core metrics: time to first fixation, findability rate (the share of shoppers who find the pack), fixation share against competitors, and noticing order.

It reads heatmaps and gaze sequences against benchmarks, such as the current pack or category norms. Hardware and webcam setups differ in precision and scale, as explained in webcam vs hardware eye tracking.

Keep stand-out separate from appeal. A pack can be noticed quickly because it is garish, and a pack can be liked and never found. Stand-out says nothing about whether the pack is liked.

Reading Communication Hierarchy From Gaze Paths

Gaze paths show what shoppers see first, second, and third: the brand, the variant, the claim, or a visual cue. That order is the communication hierarchy. If the variant is read before the brand, shoppers may not know whose product they are looking at.

The agent also flags elements that draw attention but go unread, such as a decorative graphic that pulls the eye without carrying a message. These findings feed directly into package design fixes.

Measuring Emotional Response to Pack Designs

Facial coding reads emotion at first exposure and again at the close-up view. The agent tracks valence (positive or negative), engagement, and confusion, then compares each design route against the current pack.

Test both views, because emotion at first exposure and at close-up can differ. A pack may attract at a distance and confuse when picked up.

The agent also links emotional peaks to specific design elements. A spike in confusion when the claim comes into view points to the claim, not the whole pack.

Emotion carries signal that other measures miss. Kantar's analysis of digital ads, measured with facial coding, found that ads evoking strong emotions were about four times more likely to drive impact, and that facial expressiveness predicted outcomes better than clickthrough or completion rates. That evidence is from advertising, but the lesson transfers: reactions explain what ratings do not.

Our look at emotion AI in in-store shopping explains why shopper emotion matters at the shelf.

Measuring Purchase Intent Without Inflated Claims

Stated intent scales tend to overstate. Shoppers say they would buy, then pick something else at the shelf. A choice-based shelf task is more realistic: the shopper sees the full shelf and picks a product, as they would in a store.

The gap between stated and revealed intent is why choice tasks matter. A shopper can rate a pack highly on a scale and still leave it on the shelf when five competitors sit beside it. Reasons for rejection are often more useful than reasons for choice, because they name exactly what to fix.

The agent then probes the reasons behind each choice and each rejection, in the style of package design interviews. It also measures switching from the shopper's current brand, which shows whether a pack wins new buyers or only pleases existing ones.

Treat purchase intent as the last link in the chain, not a standalone score.

How the Agent Combines Signals Into a Pack Recommendation

The agent scores each design route against thresholds or category norms across all three stages. A route might need to match the current pack on findability, beat it on valence, and win a set share of choices. These are go/no-go thresholds, set in the plan and agreed before data arrives.

Not every metric deserves equal weight. For a line extension, stand-out within the brand's own portfolio may matter most. For a redesign, protecting recognition may come first. The agent can propose weights from the objective, and the team should confirm them before data arrives.

It then ranks the routes and points to the design element that needs fixing. The output is a recommendation with its evidence attached, for the researcher to sign off.

When Attention, Emotion and Intent Disagree

Signals will sometimes conflict. The three common patterns are below.

In each case, the signal that disagrees is the one that tells you what to change.

Teams comparing consumer research platforms for this work should check that attention, emotion, and choice data sit in one workflow, otherwise the combining step happens by hand in spreadsheets.

Packaging Decisions a Research Agent Can Support

Agent-led studies suit several decisions. Redesigns must protect existing brand equity cues, such as a signature color or shape, while improving stand-out. New variants need to stand apart within the brand's own portfolio, not only against competitors.

The same chain also tests on-pack claims, sustainability messaging, and ecommerce thumbnails. Sustainability messaging should be tested for visibility and credibility, since a claim that is noticed but not believed can hurt. At thumbnail size, fine print vanishes and shape and color carry the pack, so the agent tests a reduced view and checks that brand and variant stay readable. For current ideas, see CPG packaging trends.

Where Researchers Stay in the Loop

Some decisions should stay with people. Researchers approve the hypotheses, judge brand equity, and check regulatory claims. The final go/no-go call stays with the brand team. This division is covered in human-in-the-loop AI-moderated research.

Governance protects the investment. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, citing cost, unclear value, or weak risk controls.

Measuring Attention and Emotion in One Packaging Session

Running eye tracking and facial coding in the same session connects what shoppers saw to how they felt. A shopper who fixates on the claim and shows confusion has told you something neither metric could alone.

Decode by Entropik supports agent-led pack studies. It is trusted by 150+ global brands and backed by 17 patents.

Running attention and emotion in one consumer insights platform keeps the evidence chain connected from stimulus to recommendation.

Eye tracking reaches 96% accuracy for shelf stand-out and gaze paths.

Facial coding reaches 90%+ accuracy across 62 facial expressions.

With 70+ languages supported, the same design runs across markets. Teams can explore Decode's pack design testing to see how the studies run.

Frequently Asked Questions

1. How does a research agent run packaging research?

It turns the business question into hypotheses and thresholds, builds shelf stimuli, fields the study, measures attention, emotion, and intent, and drafts a recommendation. A researcher approves each stage.

2. What is the difference between a research agent and an AI moderator in packaging research?

An AI moderator runs the interview. A research agent plans the study, fields it, analyzes the data, and recommends, using the moderator as one tool.

3. How do you measure shelf stand-out for a new pack design?

Place the pack in a realistic competitive shelf and track time to first fixation, findability rate, and fixation share against competitors.

4. Which eye tracking metrics matter most for packaging?

Time to first fixation, findability rate, fixation share, and noticing order. Read them against a benchmark, such as the current pack.

5. How does emotional response relate to purchase intent for packaging?

Emotion often explains intent. Positive valence at first exposure and close-up can support choice, while confusion can undermine it. Read them together, not as a formula.

6. What should you do when attention and purchase intent results disagree?

Look at which signal departs from the others. Strong attention with weak intent usually points to a design element shoppers dislike or misread, so probe the reasons for rejection.

7. How many shoppers do you need for a packaging study?

It depends on the design, the number of routes, and the decisions at stake. Set cell sizes with category buyer quotas and confirm them with your statistician.

8. Can a research agent test packaging across markets and retail channels?

Yes. With multi-language support, the same design can run across markets, and stimuli can mirror physical shelves or ecommerce pages.