Responsible AI Policy

Last updated: 12/7/2025

InsightPilot is committed to the responsible development and deployment of artificial intelligence in our FX trade confirmation reconciliation platform. This policy outlines our principles, commitments, and practices for AI/ML technologies used in our services.

Our AI/ML Capabilities

InsightPilot uses AI/ML technologies to help treasury teams reconcile FX trades more efficiently:

  • Document Parsing: Extracting structured data from SWIFT MT300/MT304 messages and bank confirmation PDFs
  • Fuzzy Matching: Comparing FX trades with configurable tolerance bands for rates, amounts, and dates
  • Anomaly Detection: Identifying unusual patterns that may indicate errors or require review
  • Natural Language Generation: Creating human-readable explanations for reconciliation breaks

Human Oversight

AI/ML outputs in InsightPilot are designed as assistive tools, not autonomous decision-makers:

  • All AI-generated matches and recommendations require human review before settlement decisions
  • Flagged exceptions and anomalies are presented to treasury professionals for judgment
  • Users can override or correct any AI-suggested match
  • Audit trails capture both AI outputs and human decisions for compliance purposes

Data Privacy & Confidentiality

We maintain strict boundaries around customer data:

  • No Training on Your Confidential Data: We will NOT use your FX trade data, bank confirmations, or reconciliation results to train AI models that benefit other customers
  • Data Isolation: Customer-specific model improvements (e.g., learning your tolerance patterns) remain isolated to your account
  • Aggregated Improvements: We may use anonymized, aggregated usage patterns to improve our general algorithms
  • Explicit Consent: Any fine-tuning on your specific data requires explicit written agreement

Accuracy & Limitations

We are transparent about AI capabilities and limitations:

  • AI/ML outputs are not guaranteed to be 100% accurate and should be verified by qualified personnel
  • Machine learning models may occasionally produce incorrect matches or miss discrepancies
  • Document parsing accuracy depends on the quality and format of input documents
  • Natural language explanations are generated summaries, not definitive financial analysis

Reconciliation outputs do not constitute financial, legal, or investment advice. Users are responsible for verifying results before acting on them, particularly for settlement decisions.

Continuous Improvement

We are committed to ongoing improvement of our AI systems:

  • Regular model performance monitoring and evaluation
  • Feedback incorporation from user corrections and overrides
  • Transparency in significant model updates
  • Collaboration with customers to address industry-specific needs

Questions & Contact

For questions about our AI practices or this policy:

Email: [email protected]

Related policies: