Specifics on some 2frame's Projects

2000

Client: It was developed for a research group at the University of São Paulo Faculty of Medicine Clinics Hospital (HC/USP).
Problem: This project was about speech disfluencies in several discourse types.
Main Techniques: Multivariate multilinear models and different types of correlations were the main techniques used for this work.
Other Information: The article was published.

2008

Client: The client was a European fast-fashion retail store chain, with almost 2,000 stores worldwide and more than 300 spread in the main Brazilian cities.
Problem: The client needed to forecast short, medium, and long-term sales for multiple aggregation levels (spatial and product types).
Main Techniques: Multivariate time series models were the main technique for this project.

2009

Client: Our client was one of the main Brazilian direct marketing/utility companies, carrying out a sales operation for a life insurance product targeting the Brazilian low-income population, for a Brazilian brand of an American S&P 500 index insurance company with a global presence.
Problem: Increase the efficiency of sales and the operation altogether.
Main Techniques: Propensity models for the leads and time-to-event models for the telemarketing operation were the main used techniques. Specific metrics to monitor the performance of the operation and its parts, as well as call allocation cost measurements, were derived from the time-to-event models. These measures were jointly used in creating an optimization for the telemarketing operation.
Other Information: The model successfully mapped the target consumers, and the methodology for the telemarketing operation was implemented by the contracting company as a permanent system and a new service.

2012

Client: Our client was one of the main Brazilian market research companies working for a children's television network with a global presence, owned by an American S&P 500 index media company.
Problem: To understand the parent-child relationships (parenting styles) groups and their consumption of media (newspaper, magazines, internet, etc.), leisure, health style, and children's TV content (including the network’s channel content), among others.
Main Techniques: It was used autologistic models and multiple multivariate techniques to find more or less homogeneous groups among the information provided, besides relationships and patterns in it, and likewise for within/between found groups. Of the 80 original variables, more than 70 were multivalued categorical ones.
Other Information: The study subsidized the modification of the planning of the network's programming in Brazil, as well as the rest of Latin America, and led to changes in the channel's programming and associated media content.

2013

Client: It was developed for a research group on ALS at the University of São Paulo Faculty of Medicine Clinics Hospital (HC/USP).
Problem: The researcher was developing a test that could lead to an early referral to tube feeding to ALS patients, preventing a late referral (and the involved suffering). It used measures and tests from the literature (ASHA, ALSFRS, and several derived metrics, taken in several stages of the illness) and new metrics being developed by the group (e.g., several measures related to tongue, cheek, and lip forces, besides measures related to patients’ weight, age, diet, saliva, etc., also taken in several stages of the illness).
Main Techniques: In this project, several non-parametric techniques and linear mixed models were used.

2013

Client: It was developed for a research group at the University of São Paulo Faculty of Medicine Clinics Hospital (HC/USP).
Problem: In this project, it was developed a study with head trauma patients' data on tests for the assessment of memory deficits and language-related memory deficits.
Main Techniques: Propensity, resampling, and nonparametric statistics.
Other Information: The study was published.

2014

Client: Our client was one of the main Brazilian loyalty market companies, developing a tailor-made loyalty marketing relationship program for a main Brazilian women's clothing store chain spread across the entire national territory.
Problem: To develop the program, our client needed to understand the store's customer profiles and their purchase tickets, how these customers buy, product grouping and basket analysis, each customer's spending behavior in the long term, and the expected value for each customer in the base.
Main Techniques: Repeated time-to-event models were used in this project. In addition, were developed metrics derived from the modeling results for the product grouping, spending behavior of each customer, and the expected value of each customer.

2018

Client: A Brazilian wine importer and a well-established Brazilian market research company.
Problem: In this project, the clients investigated how wine tasting differs between the wine-tasting experts' gender, age, and experience level (called groups, henceforth). The wine tasting records comprehended partially ranked multiple types of wine over about 20 years, and all wines' assessments (measures of satisfaction and measures of intensity for multiple characteristics) were ordinal.
Main Techniques: An optimization (crafted by us) extracted numerical measures from the assessments and other information provided in the records (partial ranks and ordinality of used scales, for example), producing ranking completions as a byproduct. Intergroup differences were accentuated using multivariate techniques. Multivariate techniques also were used to show patterns in the data and relationships between different assessments, and likewise for within/between groups.