What’s new at Fairgen: July 2024

We are excited to share the latest product releases at Fairgen!

Our team has been diligently incorporating feedback from customers and we are ready to share it with you. This last innovation batch not only supports our mission to transform survey sampling with statistically reliable synthetic data, but also underscores our team’s commitment to data security.

This month’s release includes:

Automatic constraints implementation, supporting skip-logic, looping, branching, and filtering
Support recoded variables
Parallel testing automated reporting
SOC 2 Type 1 certification

Before we dive into each release, be sure to subscribe for product updates delivered directly to your inbox.

Automatic constraints implementation

We have successfully implemented the first release of a series of upcoming features to ensure our synthetic sample boosters respect the same constraints and filters applied during real-world survey programming.

Users can expect a seamless application on a broader range of survey types. The following constraints are now supported:

Skip-logic: Fairgen automatically maps the presence of skip-logic on real data and enforces its synthetic data to always bypass certain questions based on a respondent's previous answers. Example: Respondents aged 18-24 should skip questions 1 to 4.
Loop skip-logic: Fairgen automatically understands the presence of looping logic, ensuring that each respondent answers only the relevant repeated questions based on previous answers. Example: Respondent only evaluates a set of brands if he/she has declared that they were used in the last year.
Branching: Synthetic data is enforced to always respect different paths respondents can enter during a survey based on answers provided. Example: If a respondent answers "Yes" to owning a pet, they are branched to a set of questions about pet care.
Filtering: Similar to branching, Fairgen automatically detects filtering patterns, ensuring no synthetic respondent who should be excluded from a set of questions has answers. Example: Only respondents who work in FinTech will see questions about the finance industry.

Mapping recoded variables

Users can now manually map recoded values to ensure generation will respect these patterns across the entire fairset. By simply informing the platform which question variables are associated, Fairboost will automatically map the logic and ensure its generation always respects the recoded patterns.

Self-serve parallel testing reporting

After hundreds of scientific evaluation processes internally and with our customers, we’ve streamlined the experience of parallel testing reporting. We now allow users to independently compare the performance of a fairset (synthetic boost), its training dataset to a ground truth dataset. The option to add a baseline dataset is also added to facilitate the understanding of boost factors. With this new experience, users can:

Calculate errors in relation to a holdout ground truth dataset.
Understand the boost factor by comparing synthetic boosts to real-world boosts.
Analyze at aggregate levels by looking at Mean Absolute Error (MAE) and error distribution per dataset.
Analyze at granular levels by visually comparing answers and Mean Absolute Error (MAE) across each question.

Data security certification: SOC 2 Type 1

We are happy to announce that Fairgen has earned SOC 2 Type I certification, reinforcing our commitment to data security and confidentiality. The certification, widely known as a standard for securing data, highlights the efficiency of our protocols and controls in protecting client information. It certifies that our customers’ information is safe.

This milestone demonstrates how we prioritize customers’ digital privacy. By getting SOC 2 Type I certified, we assure our clients about our control over data management and commitment to trustworthiness and openness. Fairgen continues following these principles as it guarantees AI-driven research solutions with reinforced security measures.