FACET: Open-Source Library for AI Model Explainability

Artificial intelligence can predict outcomes, but its inner workings can be a mystery. GAMMA FACET, our open-source library, cracks complex AI algorithms, uncovering how variables contribute to a model’s reasoning—and how you can get better results.

Request a demo

Predictive AI is not a magic crystal ball. It’s based on science and has a long track record. But for many business leaders, the analogy does raise a point. Without understanding complex machine-learning (ML) algorithms, how can we truly trust the models—let alone base strategies upon them?

GAMMA FACET, an open-source library by BCG GAMMA, takes a unique approach, showing how relevant features interact to determine key outcomes. This helps data scientists avoid common misinterpretations of ML models. And it reveals the variables that matter most. By explaining AI, FACET informs—and improves—decision making for both data science and the business.

How FACET Works

By analyzing patterns and making predictions, AI algorithms help advance everything from manufacturing efficiency to social impact. But as AI models become more complex, it is harder to understand how they arrive at their conclusions. That knowledge is crucial, for when you know what is driving a result, you know what to improve—and where to optimize—to get better results.

FACET delivers insights on the workings of AI in two ways:

Model Inspection

Applying a unique geometric interpretation of ML models, FACET assesses how the variables interact. This analysis transforms the explainability of ML models, as variables often do not act independently, but in a redundant or synergistic way. Understanding these dependencies has long been difficult, which could lead to significantly distorted interpretations of ML models. By uncovering these relationships, FACET reveals the true contributions of each variable.

Model Simulation

Once you know the critical variables and how they’re connected, you’re ready to optimize them. FACET’s simulation algorithms let data scientists run what-if scenarios—virtual experiments to discover how changing key parameters affects predicted outcomes. By helping users identify the parameters that matter most, FACET provides a roadmap for achieving even better outcomes.

The Benefits of Understanding Complex ML Models

We designed FACET for easy use by data scientists. But the benefits extend across an organization. Chief among them:

Our Client’s Success with FACET

BIOPHARMA INDUSTRY
Maximizing production output of a pharmaceutical drug is a crucial goal, as it expands access to treatment while creating more value. The company used FACET’s explainable AI open-source library to identify parameters with the biggest impact on output, understand their relationships and dependencies, and conduct what-if simulations to find the optimal values for maximizing production volume.

Our experts work across industries—and around the world—to help organizations leverage the power of advanced analytics. Here are some of our experts on AI algorithms.