Explainable AI

Know the Why and the How Behind Your AI Solutions

Model performance is reliant not only on metrics but also on how well a model can be explained when something eventually goes wrong. 
Fiddler AI analyze slice query with charts
"Fiddler's explainability capability is a real game changer, giving us a much better understanding of our models. It was the most convincing argument to use Fiddler instead of its competitors"
Richard Sieg
Lead Data Scientist, how.fm

Explainable AI techniques at enterprise scale

The Fiddler AI Observability platform delivers the best interpretability methods available by combining top explainable AI principles, including Shapley Values and Integrated Gradients, with proprietary explainable AI methods. Obtain fast model explanations and understand your ML model predictions quickly with Fiddler Shap, an explainable method born from our award-winning AI research.

Hear from our Data Science team about different Explainable AI concepts:

Build trusted AI solutions that delight customers

Fiddler AI feature attribution
Deploy responsibly

Gain visibility into your most deeply complex models

To ensure continuous transparency, Fiddler automates documentation of explainable AI projects and delivers prediction explanations for future model governance and review requirements. You’ll always know what’s in your training data, models, or production inferences.

  • Understand every single prediction made by your AI solution and spot discrepancies using out-of-the-box explanation methods.
  • Simulate ‘What If’ scenarios in tabular and text models to validate and build trust.
  • Bring in data and models from any platform and gain faithful explanations.
Minimize risk

Increase coverage, efficiency, and confidence in your models

You can deploy AI governance and model risk management processes effectively with Fiddler. 

By proactively identifying and addressing deep-rooted model bias or issues with Fiddler, you not only safeguard against costly fines and penalties but also significantly reduce the risk of negative publicity. Stay ahead of AI disasters and maintain brand reputation.

  • Understand model predictions at a global and local level before it’s put into production.
  • Explain complex recursive inputs, or bring your own explainers specific to your use case.
  • Generate and share any type of report required for MRM and compliance reviews using Fiddler Report Generator.
Fiddler AI prediction discrepency
Increase brand loyalty

Responsible AI principles make customers happy

It’s important to understand model performance and behavior before putting anything into production, which requires complete context and visibility into model behaviors — from training to production.

When areas of low performance or potential issues are rooted out before they are seen, the customer experience is improved, leading to higher Net Promoter Scores (NPS) and stronger customer recommendations. 

  • Compare distributions across training, test, and production data.
  • Explain performance discrepancies across groups by slicing data into groups.
  • Find anomalies and data drift by analyzing AI predictions in relation to an entire data set or specific groups.

Zoom in to the details

Explainable AI features

On demand explanations

Perform counterfactual analysis in real-time.

Gradient-based explanations

Get faster explanations than just perturbation-based explanations.

Advanced explainability

Explain complex recursive inputs for fine-grained attributions.

Shapley values (SHAP) and Fiddler SHAP

Increase your model’s transparency and interpretability using SHAP values, including our award-winning Fiddler SHAP metric.

Integrated gradients

Run deep learning models, including NLP and CV models, faster and comprehend how data features contribute to data skew and model predictions.

‘What-if’ analysis

Gain a better understanding of your model’s predictions by changing any value and studying the impact on scenario outcomes.

Global and local explanations

Understand how each feature you select contributes to the model’s predictions (global) and uncover the root cause of an individual issue (local).

Surrogate models

Improve the interpretability of your models before they go into production by using automatically generated surrogate models.

Custom explanations

Bring your own customized explainers specific to your use case, via APIs, to our patented UI.