FMCG research should build an evidence chain of "consumption scenario — choice — product experience — repurchase," rather than relying on a single brand attitude survey to simultaneously determine positioning, packaging, flavor, and sales volume. Consumers may like a concept but not notice it on the shelf; they may prefer a flavor in blind testing but reject the format or price; or they may make an initial purchase due to a promotion, but repurchase disappears as the usage experience falls short. Each product stage requires methods tailored to answer the decision at that stage, and must clarify what evidence is still needed for the next step.
The best starting point for research is usually not "what matters most when buying a beverage?" but rather the most recent specific purchase or usage: where it happened, for whom, the problem to solve, the available alternatives, and why this item was ultimately chosen. FMCG choices are influenced by inventory, promotions, companions, usage occasions, and shelf environment. Asking abstract preferences outside actual events tends to yield answers like "quality, health, value for money" that are correct but not actionable.
First Determine Which Layer of Evidence the Product is In
Different stages may all appear to ask "do consumers like it?" but the criteria for judgment differ:
- Opportunity exploration: Identify frequent consumption tasks that current solutions fail to address and that the brand is capable of solving, relying primarily on event interviews, diaries, and behavioral data.
- Concept screening: Compare relevance, distinctiveness, credibility, and conditional consideration intent within target scenarios, while controlling for price, format, and visual differences.
- Product experience: Validate taste, aroma, texture, perceived efficacy, and ease of use; blind tests and branded tests answer different questions.
- Packaging and information: Examine shelf findability, brand and category recognition, comprehension of key information, format judgment, and open/use behavior—not just aesthetics.
- Post-launch validation: Observe trial purchase, repeat purchase, usage frequency, returns, and complaints under distribution conditions, and explain gaps relative to earlier intent.
Each layer can overturn the optimistic conclusions of the previous one. Winning at the concept stage does not mean winning in the formula; satisfaction in product trial does not equal channel preference; and a first purchase does not guarantee stable repurchase. At project inception, only identify the single decision that must be made now, and define pass, modify, and kill criteria in writing. Doing so reduces situations where "everything was measured but we don’t know which one to choose."
Define Target Samples by Recent Behavior, Not Demographics
Age, gender, and city tier help structure samples but do not prove that a person can evaluate the category. Eligibility should include evidence of personal purchase or actual use within a defined reference period, frequency, primary occasions, brand repertoire, and decision-making role. In children’s food, the payer, chooser, preparer, and consumer may differ; in household cleaning products, the buyer may not be the frequent user. Different questions must be directed to those who have the relevant experience.
The reference period should match purchase frequency. For ready-to-drink beverages consumed daily, asking about the most recent purchase or the past few weeks is appropriate; for low-frequency household goods, a longer window may be needed, supplemented by order or occasion cues. The U.S. Census Bureau’s questionnaire design research indicates that ordinary events such as purchases are underreported as recall intervals lengthen, and event timing tends to be remembered as more recent than it actually was. Instead of uniformly asking about "the past year," first confirm the most recent event, then collect enough information to classify frequency.
Current brand users are suitable for studying usage and attrition but may overestimate brand familiarity; heavy category users can quickly reveal details but may miss barriers for light users. The sampling plan should combine current users, lapsed users, competitor users, and light category users based on the growth question at hand, and report sources and unweighted bases separately—do not label convenience-recruited individuals as "all consumers."
Concept Test with Six Hundred Respondents: Overall Inconclusive, Scenarios Point the Way
Below is an illustrative calculation. A team compared two main concepts for a sugar-free ready-to-drink tea: Concept A emphasized "everyday freshness with low burden," and Concept B emphasized "switching state when focus is needed." Six hundred recent category buyers were randomly assigned, each viewing only one version; packaging layout, volume, price, brand prominence, and message length were held constant. In Group A, 174 of 300 considered purchasing under the given scenario and price (58%); in Group B, 153 of 300 did (51%).
The overall difference was seven percentage points. Using a simple approximation for two independent proportions, the 95% confidence interval was roughly -0.9 to +14.9 percentage points, containing zero. This does not support "A is the clear winner," nor does it mean the two are identical; the current sample remains insufficiently certain about the population difference. If the team looks only at point estimates, it risks turning random fluctuation into a definitive answer.
The study pre-specified a key scenario before data collection: individuals who purchased sugar-free beverages in the morning at least four times a week. In Group A, 120 fell into this segment, of whom 84 considered purchasing (70%); in Group B, 118 fell into this segment, of whom 65 considered purchasing (approximately 55.1%). The difference was about 14.9 percentage points; the approximate interval ranged from 2.8 to 27.1. Among the remaining participants, Group A had 90 out of 180 (50%), and Group B had 88 out of 182 (approximately 48.4%), a very small difference.
The correct decision based on these results depends on the growth objective. If the product is specifically designed to serve the high-frequency morning scenario, A warrants moving to the next round of product and shelf verification; if the goal is an all-day mass-market product, the overall evidence is insufficient—one cannot declare universal superiority based on a single subgroup. Key subgroups must have business rationale and be pre-specified before analysis; scanning dozens of cross-tabs to cherry-pick significant segments amplifies chance findings and should be treated as new hypotheses requiring replication in independent samples.
How to Avoid Changing Five Variables at Once in Concept Experiments
Concept comparisons should ensure that versions vary only on the element under decision. If A uses green packaging, a lower price, shorter copy, and a "low-sugar" claim, while B uses red packaging, a higher price, and "energy-boosting," no result can be attributed. First compare the core value using a unified template, then test packaging, price, and product experience separately for winning concepts.
A monadic design, where each person sees only one version, is closer to real exposure and reduces cross-cueing; a sequential randomized multi-concept design saves sample size and provides within-person comparisons but requires controlling for fatigue, order, and contrast effects. The choice depends on the number of stimuli, sample budget, and whether cross-version influence is a concern. Whichever is chosen, randomization, stimulus version, and exposure failure records should be kept, and the two groups should be checked for approximate balance on key eligibility and background variables.
Purchase intent items should specify a sufficiently realistic price, volume, channel, and usage scenario. Without these conditions, "would you buy?" reads more like a polite response to copy. In addition to intent, measure what consumers think the product is, for whom and when it is used, the core promise, and their own words for reasoning; a concept liked but misunderstood is not a success.
Packaging Tests: First Task, Then Aesthetics
Packaging performs four tasks: recognition, selection, comprehension, and use. Test discoverability of the category and brand in a simulated shelf setting; test memory of key information after brief exposure; ask participants to judge format, flavor, and usage occasions without prompting; finally, observe real actions such as opening, pouring, storing, and disposal. Click maps can show attention location but cannot alone prove correct comprehension.
The U.S. Food and Drug Administration, when developing front-of-package nutrition information, used literature review, consumer focus groups, and randomized experiments with nearly 10,000 adults to compare how different labels helped consumers assess nutritional information quickly and accurately. This case shows that exploring language, comparing designs, and validating task performance are distinct steps; "which label do you prefer?" cannot replace "which is understood faster and more accurately?"
Packaging materials must be managed by version. Front and back panels, display size, lighting, adjacent competitors, rotatability, and mobile screen ratio all affect task performance. Online simulations are suitable for low-cost screening; real shelves or physical handling are closer to actual use. If packaging must carry allergen, nutrition, usage warning, or other legal information, the research team should have it reviewed by relevant legal and regulatory professionals; consumer comprehension testing cannot replace compliance review.
Product Experience: Distinguish Blind Testing, Branded Testing, and In-Home Use
Blind tests minimize brand and packaging expectations and are suitable for comparing sensory or physical performance; branded tests reveal whether promises align with actual experience; in-home use lets consumers use the product in real time, with real meal pairings, storage, and household contexts, suitable for repeated use or habit-formation products. It is reasonable for the three to yield different answers.
If blind results are strong but branded results decline, erroneous expectations may have been set by price, positioning, or packaging; if single-use lab trials are good but in-home repetition declines, issues with dosage, cleaning, portability, or long-term feel may exist. Reports should record batch, formula, temperature, preparation, number of usage occasions, and exposure conditions to avoid mistaking experimental differences for consumer differences.
Product claims must also be separated from experience judgments. Surveys can test how consumers understand expressions like "low burden" or "long-lasting freshness," but cannot prove nutritional, health, safety, or efficacy facts. The U.S. Federal Trade Commission’s advertising substantiation policy requires that objective express and implied claims be supported, prior to dissemination, by a reasonable basis consistent with the nature of the claim; high-risk health or safety claims typically require competent and reliable scientific evidence—"most people feel it works" is not a substitute.
Connect Intent to Observable Choices to Learn Prediction Bias
Concepts entering the next stage should record predictions: target population, scenario, price condition, expected choice rate, and failure signals. Subsequently validate stepwise with simulated shelf choice, physical trial, small-scale street placement, coupon redemption, or authorized purchase data. Each level is progressively closer to real behavior but also introduces new variables such as distribution, promotion, out-of-stock, and channel.
Build a simple calibration table: record denominators and time windows separately for early consideration intent, simulated choice, first purchase, second purchase, and stable usage. If consideration intent is high but shelf choice is low, first check findability, format, and price; if first purchase is high but repurchase is low, examine product experience and promise gap; if intent is moderate but actual choice is good, the survey may have failed to recreate shelf habits, promotions, or availability. Do not leave only a "concept score," otherwise it is impossible to learn which stage fails to predict the next.
Nine Checks Before Publishing FMCG Research
- The present study addresses one clear product stage and decision.
- The sample is verified by recent category behavior, frequency, scenario, and role.
- Sources and bases for current users, competitor users, lapsed users, and light users are visible.
- Concepts are comparable in price, format, layout, and information load except for the element under test.
- Randomization, order, exposure failures, and stimulus versions are traceable.
- Packaging metrics correspond to discovery, recognition, comprehension, or usage tasks, not just liking.
- Product tests record blinding status, batch, preparation, number of uses, and usage environment.
- Key subgroups are pre-specified, and exploratory findings are designated for independent replication.
- All purchase intent items specify price and scenario, and state the next behavioral validation step.
For developing testable concepts, refer to Product Concept Test Survey Design; for packaging visual tasks, combine with Image Click & Heatmap Analysis, but retain comprehension items; for price condition choices, continue reading Pricing Research Methods. Final conclusions should note the product stage, behavioral sample eligibility, stimuli and price conditions, unweighted bases, uncertainty, and behaviors still awaiting validation.
Methodological Foundations and Further Reading
- U.S. Census Bureau: Survey Questionnaire Construction: Forgetting of consumption events, telescoping, reference periods, and contextual cues.
- U.S. FDA: Front-of-Package Nutrition Labeling consumer research: Evidence chain of literature review, focus groups, and randomized label experiments.
- U.S. FTC: Policy Statement Regarding Advertising Substantiation: Prior evidence requirements for objective express and implied product claims.
- AAPOR Transparency Initiative: Disclosure of samples, eligibility, randomization, stimuli, quotas, and weighting.