Back to home

Play Video

How We Work

Knowledge and Expertise

Our research team possesses extensive expertise not only in consumer intelligence processes but also programming knowledge and academic experience. This enables us to solve business problems across many fields by applying analytical methods that work best for each specific challenge.

Statistical Methods

Machine Learning

NLP Techniques

Deep Learning

Large Language Models

Qualitative Methods

Workshop Methods

Classic Statistical Methods

These techniques form the foundation for understanding data structure, hypothesis testing, and modeling simple relationships. They require deep methodological knowledge for proper application and interpretation.

Time Series Analysis

Studying data collected over time to identify trends, seasonality, cyclicality, and forecasting.

Helps understand patterns in temporal data and make predictions about future values.

Applications:

Sales forecasting
Market dynamics analysis
Evaluating the impact of events over time

Structural Equation Modeling (SEM)

Advanced methods for modeling complex relationships between observed and latent (hidden) variables.

SEM tests theories, while Partial Least Squares SEM (PLS-SEM) is more predictive.

Applications:

Building and testing customer satisfaction models
Analyzing loyalty drivers
Studying the impact of image on behaviors
Analysis of complex causal pathways

Linear & Logistic Regression

Modeling the influence of one or more independent variables (predictors) on a dependent variable.

Linear regression is used for continuous outcomes, while logistic regression is used for categorical/binary outcomes.

Applications:

Identifying and quantifying drivers (factors affecting satisfaction, loyalty, purchase intent)
Forecasting continuous or binary outcomes (e.g., whether a customer will make a purchase)
Determining statistical significance of relationships between variables

Factor Analysis

Dimensionality reduction; grouping correlated variables into latent factors.

This technique identifies underlying variables that explain patterns observed in complex datasets.

Applications:

Simplifying complex data
Creating perceptual maps
Identifying key dimensions of perception (e.g., brand image dimensions)

Conjoint Analysis

Measuring the value consumers assign to individual features (attributes) of a product/service.

Helps understand trade-offs consumers make between different product features and price points.

Applications:

Product/service optimization
Pricing strategy
Analysis of preferences for attributes

Cluster Analysis

Grouping objects (e.g., consumers, products) into homogeneous segments based on their characteristics.

Examples include two-step or hierarchical clustering techniques.

Applications:

Market segmentation
Grouping competitive products/brands
Identifying natural groupings in customer data

Classical Machine Learning Methods

ML algorithms allow learning patterns from data to automate predictive, classification, and grouping tasks on a larger scale, often on data with greater complexity than in pure classical statistics.

CHAID (Chi-squared Automatic Interaction Detection)

A decision tree algorithm for segmentation and classification using statistical tests for splits.

Particularly useful in market research for readability of segmentation results. Can perform multi-way splits.

Applications:

Market segmentation based on readable tree structure
Segment profiling
Identifying interactions between variables

ML-based Cluster Analysis

Application of ML algorithms for automatic data grouping.

Examples include K-Means and DBSCAN clustering algorithms.

Applications:

Scalable segmentation on large datasets
Discovering hidden behavioral groups
Identifying natural patterns in complex data

Classification Models

Modeling the influence of one or more independent variables (predictors) on a dependent variable.

Examples include CHAID decision trees, Random Forests, Support Vector Machines (SVM), and logistic regression.

Applications:

Churn prediction
Customer segmentation
Conversion prediction
Automatic data categorization

Regression Models

Building models that predict continuous numerical values.

Examples include Random Forest for Regression and Gradient Boosting Machines

Applications:

Precise sales/demand forecasting
Modeling drivers with non-linear relationships
Customer Lifetime Value (CLV) prediction

Dimensionality Reduction

Algorithms that reduce the number of variables, often as a preprocessing step for other ML models or for visualization.

Principal Component Analysis (PCA) is a common example.

Applications:

Data simplification for models
Complex data visualization
Feature extraction and selection

Structural Equation Modeling (SEM)

Advanced methods for modeling complex relationships between observed and latent (hidden) variables.

SEM tests theories, while Partial Least Squares SEM (PLS-SEM) is more predictive.

Applications:

Building and testing customer satisfaction models
Analyzing loyalty drivers
Studying the impact of image on behaviors
Analysis of complex causal pathways

NLP Techniques Generating Quantitative Data

Methods for creating vector representations of text that become quantitative data for further analysis by ML/statistical models. The goal is to prepare textual data for quantitative analysis (e.g., classification, clustering).

N-grams

An extension of BoW that treats sequences of N consecutive words as units instead of single words.

Unigrams are single words, bigrams are 2-word sequences, trigrams are 3-word sequences, etc. This partially preserves information about local word order.

Applications:

Phrase-level analysis
Capturing multi-word expressions
More contextual text analysis

Sentiment Analysis

The process of automatically determining subjective attitude (positive, negative, neutral) or emotions expressed in text.

Classical approaches often relied on sentiment lexicons (list of words with assigned emotional 'charge') or simple ML classifiers trained on features such as BoW or N-grams.

Applications:

Brand perception monitoring
Customer feedback analysis
Social media monitoring

Bag of Words (BoW)

The simplest method of text representation, where a document is presented as a set (vector) of words.

This approach disregards word order or grammatical structure. Values in the vector can be a simple count of word occurrences or weights (e.g., TF-IDF).

Applications:

Basic text classification
Document comparison
Simple sentiment analysis

TF-IDF (Term Frequency-Inverse Document Frequency)

A word weighting technique that assigns greater weight to words that are rare in the entire document collection but frequent in a given document.

This is an improvement over simple occurrence counts in BoW, giving greater importance to words characteristic of a given text.

Applications:

Keyword extraction
Document relevance scoring
Improved text classification

Topic Modeling

Identification of abstract topics in a collection of texts and representing each document as a quantitative probability distribution of topics.

Latent Dirichlet Allocation (LDA) is a common technique.

Applications:

Content organization
Trend discovery in large text collections
Document summarization and classification

Structural Equation Modeling (SEM)

Advanced methods for modeling complex relationships between observed and latent (hidden) variables.

SEM tests theories, while Partial Least Squares SEM (PLS-SEM) is more predictive.

Applications:

Building and testing customer satisfaction models
Analyzing loyalty drivers
Studying the impact of image on behaviors
Analysis of complex causal pathways

Deep Learning Methods

Deep Learning techniques, based on deep neural networks, revolutionize data analysis, especially when data is complex, high-dimensional, or initially unstructured (text, image, video), from which quantitative features or results are then extracted.

Sequential Analysis

Modeling data arranged in sequences using RNNs, LSTMs, and Transformers.

These models can capture patterns that evolve over time or in sequence.

Applications:

Precise forecasting of next events (e.g., next purchase)
Conversion path analysis
Anomaly detection (e.g., fraud in transactions/online behaviors)
Advanced time series analysis and forecasting

Word Embeddings

Techniques that create dense numerical vectors for words, positioning words with similar meanings close together in vector space.

These capture semantic and syntactic relationships (e.g., 'king' - 'man' + 'woman' ≈ 'queen'). Word2Vec was one of the first popular models of this type.

Applications:

Semantic search
Content recommendation
Identifying related concepts and terms

Image and Video Analysis

Analysis of visual content to extract quantitative data using Convolutional Neural Networks (CNN).

These networks excel at recognizing patterns in visual data through multiple layers of processing.

Applications:

Product recognition on shelves (count, presence/absence)
Customer movement and behavior analysis in physical stores
Advertising material analysis (element identification, faces, objects)
Emotion analysis from facial expressions in videos

Advanced Text Analysis

Processing text to obtain precise quantitative and categorical data using contextual embeddings and Transformers.

Models like BERT from Amazon AWS Bedrock can understand context in ways earlier models couldn't.

Applications:

Very accurate, contextual sentiment analysis (including aspect-based)
Precise key information extraction (Named Entity Recognition)
Automatic, granular coding of open-ended responses and qualitative data
Advanced topic modeling capturing nuances

Topic Modeling

Identification of abstract topics in a collection of texts and representing each document as a quantitative probability distribution of topics.

Latent Dirichlet Allocation (LDA) is a common technique.

Applications:

Content organization
Trend discovery in large text collections
Document summarization and classification

Structural Equation Modeling (SEM)

Advanced methods for modeling complex relationships between observed and latent (hidden) variables.

SEM tests theories, while Partial Least Squares SEM (PLS-SEM) is more predictive.

Applications:

Building and testing customer satisfaction models
Analyzing loyalty drivers
Studying the impact of image on behaviors
Analysis of complex causal pathways

Large Language Models (LLMs)

LLMs are a specific, cutting-edge category of DL models that revolutionize the possibilities of working with textual data, allowing for extraction and generation of quantitative insights with much greater freedom and contextual understanding.

Insight Synthesis from Multiple Text Sources

LLMs can 'read' a large collection of documents and identify key themes, conflicts, and opinions.

These can then be summarized and delivered in a form that enables their quantitative analysis (e.g., a list of key arguments with the number of documents supporting them).

Applications:

Market research synthesis
Competitive intelligence
Trend identification from multiple sources

Synthetic Data Generation

LLMs can be directed to create textual data that simulates specific distributions or characteristics.

For example, generating realistic reviews with specified sentiment or survey responses representing a particular segment.

Applications:

Creating training data for other models
Testing hypotheses with synthetic scenarios
Augmenting limited real-world data

Structured Data Extraction

LLMs can understand complex instructions in natural language and extract specific data from text, formatting it in JSON, tables, etc.

We use Amazon Nova via Bedrock for Natural Language to Data conversion.

Applications:

Automatic database population with information from documents
Gathering specific market data from unstructured text sources
Extracting prices, product specifications from online descriptions

Advanced Categorization and Coding

Instead of training on labeled data, LLMs can categorize or code texts based on category descriptions or examples given in natural language (prompts).

This allows for very fast, flexible, and granular categorization of open-ended responses and qualitative data.

Applications:

Rapid analysis of open-ended survey responses
Social media data categorization
Review and qualitative data coding

AI Agents

The most advanced level of our technological expertise, using the reasoning and planning power of LLMs to create intelligent agents.

We use Amazon Bedrock Agents to build systems capable of independently executing complex data-related processes.

Applications:

Automated market/reputation monitoring
Combining classical data analysis techniques with AI reasoning
Scaling research activities that previously required human intervention
Testing scenarios and experimenting in secure environments

Structural Equation Modeling (SEM)

Advanced methods for modeling complex relationships between observed and latent (hidden) variables.

SEM tests theories, while Partial Least Squares SEM (PLS-SEM) is more predictive.

Applications:

Building and testing customer satisfaction models
Analyzing loyalty drivers
Studying the impact of image on behaviors
Analysis of complex causal pathways

Publications and Expert Appearances

Publications

Mundialowa klęska urodzaju (World Cup Abundance Failure)

Aleksandra Smoter, Nowy Marketing

READ ARTICLE

Kiedyś "grey", dzisiaj "silver" (Once "Grey", Now "Silver")

Aleksandra Smoter, Media Marketing Polska (Silver Generation)

READ ARTICLE

Presentations

Łże-wykresy czy lepszy sort danych? (False Charts or Better Data?)

Aleksandra Smoter, Forum IAB

WATCH PRESENTATION