Advanced Analyses for Product
and Concept Tests

The "Penalty" (customer loss) caused by a poor product feature (e.g., packaging quality) and the "Reward" (customer acquisition) created by its excellence are not linear and symmetrical. Our advanced Kano modeling maps this non-linear asymmetry by testing the effect of product features on overall liking using Dummy Variable Regression.

Selecting the 5 "most popular" products (or flavors) to launch does not maximize your market reach; because the consumer base that likes these 5 popular products often overlaps. By calculating tens of thousands of possible combinations, the TURF algorithm finds the most optimal "unique product lineup" (unduplicated reach) that will reach the maximum percentage of the market without stealing each other's customers (without cannibalization).

Especially in physical prototype and food (FMCG) testing, the fact that consumers find a feature "too sweet" or "not dense enough" (JAR deviations) is not a sufficient insight on its own. Sensory Penalty Analysis simultaneously models the size of the audience declaring these specific sensory deviations (Frequency) and the "Penalty Score" (Mean Drop) this deviation wipes from the overall product liking.

Manually reading hundreds of pages of open-ended feedback collected about a new concept creates a human (cognitive) blindness. Using Natural Language Processing (NLP) and word co-occurrence algorithms, we transform the unstructured text responses consumers give to the concept into a mathematical Network Graph topology.

In categorical (non-numeric) product test data, finding out which prototype or competitor brand is identified with which specific sensory profile (e.g., "Crispy", "Classic", "Heavy") is a complex matrix analysis. Multiple Correspondence Analysis (MCA) creates a flawless "Perceptual Map" by translating multidimensional categorical (nominal) data onto a two-dimensional spatial plane.

In product tests, the contribution of each feature to satisfaction has a directional character. Our proprietary Potential Impact Analysis (PIA) algorithm calculates the regression coefficients of linear deviations in features on the Overall Liking score; making it definitively clear whether the product's feature dosages (e.g., Acidity, Hardness, Brightness) should be increased or decreased.

One of the most challenging measurements in consumer research is finding the "true competitive power" of prototypes. The 3P Model (Paired Comparison Tournament Design) pits all tested prototypes against each other in simulated matches (tournaments), calculating a "Win-Rate Preference" and cross-tabulating this with an external "Performance Index".

While ordinary regressions focus on the consumer, Econometric Discrete Choice models (Conditional Logit) directly target the "product" and its "Choice Probability". This advanced model calculates how many times product features (JAR) and the product's own brand/prototype constant (ASC) increase the likelihood of the product being picked off the shelf (Odds Ratio).

In physical product tests where the consumer classifies the product in binary scenarios like "Like / Dislike", classical correlation or linear regression models lose their statistical validity. At Datametri, we deploy Machine Learning libraries (caret::varImp) and Logistic Regression (Binomial GLM) algorithms in such cases. Holding all other features constant, this analysis conclusively ranks which sensory attribute holds the highest Z-score (statistical power) in creating the "Overall Liking" classification.