Overview of AI Usage and Controls in SupportLogic

Introduction:

SupportLogic is a SaaS Service Experience (SX) platform that uses AI, ML, and Natural Language Processing (NLP) to analyze support interactions and generate actionable insights.

The platform helps support teams proactively reduce escalations, improve CSAT/NPS, and shorten case resolution time, while operating under a governed framework focused on accuracy, fairness, privacy, and human oversight.

1. AI Model Framework:

The platform uses:

• Large Language Models (LLMs)
  
  

• Statistical and machine learning models
  
  

AI capabilities include internally developed models and third-party models from:

• OpenAI
  
  

• Anthropic
  
  

• Amazon Web Services (AWS Bedrock)
  
  

These models are integrated under defined governance and monitoring controls.

2. Data Privacy & Handling:

• Personally identifiable information (PII) is not stored for model training purposes.
  
  

• User inputs are not retained for generalized model retraining.
  
  

• External AI providers do not use customer data for independent training.
  
  

• Data minimization principles are applied during development.
  
  

Privacy-by-design policies govern AI system design and deployment.

3. Accuracy & Performance Controls

AI models are subject to structured evaluation before and after deployment. Controls include:

• Defined minimum accuracy thresholds
  
  

• Regular performance testing
  
  

• Benchmark comparisons against live models
  
  

• Rollback mechanisms if performance declines
  
  

Fallback models to maintain stability

4. Bias & Fairness Management

• Balanced training datasets are used during development.
  
  

• Bias testing is conducted during development and at defined intervals post-deployment.
  
  

• Detected bias triggers retraining and remediation.
  
  

• Users can report problematic outputs for review.
  
  

In limited cases, generative features may show bias toward certain products mentioned in case details. These instances are monitored and refined to maintain neutrality and fairness.

5. Human Oversight

SupportLogic follows a human-in-the-loop approach:

• AI generates signals and recommendations.
  
  

• Human reviewers evaluate and act on outputs.
  
  

• Decisions are not solely automated.
  
  

• Users can request human review.
  
  

Overrides and corrections are documented.

6. Risk Mitigation Summary

The AI system is assessed for risks, including:

• Bias and discrimination
  
  

• Lack of transparency
  
  

• Excessive data collection
  
  

• Model drift
  
  

• Inaccurate generative outputs
  
  

Mitigation measures include ongoing testing, defined accuracy thresholds, structured monitoring, human oversight, audit logging, and documented remediation procedures.