XAI (explainable AI)
In finance, explainability and verifiability are the basis of trust.
NoahAI's XAI exists to explain "why this choice was made"; it does not explain, organize, or steer the user's judgment.
XAI is not just an explanation feature; it is a protection layer that understands personal financial context and detects risk signals that differ from usual patterns, prompting extra confirmation. The goal is not to explain or organize judgment or guide behavior but to give the user enough context and reasoning to make their own sound decision.
NoahAI is a transparent AI system that can explain and trace every decision. Explainability is the core mechanism for trust, audit, and reproducibility, not a mere feature.
Value of explainability
Trust
Scenario:
When the user asks "Why was this trade made?"
XAI solution:
XAI combines market analysis, pattern recognition, risk assessment, and past performance to give a clear explanation.
Benefit:
The user can understand and trust AI judgment.
Audit and trace
Scenario:
When a regulator or internal audit reviews transactions
XAI solution:
Every judgment is recorded regardless of whether a trade occurred; input → reasoning → risk → options presented are traceable in time order.
Benefit:
Full audit trail supports compliance.
Reproducibility
Scenario:
When verifying that the same conditions yield the same result
XAI solution:
Input data, decision process, and execution result are all recorded so conditions can be reproduced and results compared.
Benefit:
System consistency and reliability can be verified.
Improvement and learning
Scenario:
When analyzing failed trades to improve
XAI solution:
XAI logs allow precise identification of failure causes and reflection in policy improvement.
Benefit:
Structured learning and improvement are possible.
Fraud and phishing defense
Scenario:
When the user receives abnormal financial requests or suspicious guidance
XAI solution:
Past behavior, fund-flow context, and mismatch with normal procedures are detected, flagged as risk, and additional confirmation is guided.
Benefit:
Protects users from voice phishing, fraud, and financial crime.
Personalized anomaly detection
Scenario:
When attempting a transfer with unusual amount, recipient, or time
XAI solution:
XAI compares the user's past fund flow, regular spending, and key-relationship patterns in anonymized context, explains anomaly, and guides extra confirmation.
Benefit:
The user understands deviation from their usual pattern and can recognize mistake or fraud risk in advance.
Log structure
Every decision is recorded with the following:
[analysis]
Market analysis log: data analyzed, pattern results, signals, technical indicators
[trade]
Execution log: entry signal, order creation, execution result, slippage, position check
[order]
Order management: TP/SL, order status, Algo Order API, retry logic
[monitor]
Monitoring: position tracking (e.g. 2s), market change, PnL, dynamic threshold checks
[exit]
Exit log: reason (TP/SL, dynamic threshold, external change), result, P&L, learning data
[safety]
Anomaly and abnormal flow: phishing suspicion, excessive urgency, skip-procedure requests, etc.
Real-time logging: Every decision-support step is logged in real time; the user can review the full AI judgment process. Logs are categorized for easy trace and verification.
Implementation examples (explanation and protection)
XAI works with the same explain-and-record principle not only for investment judgment but also for everyday finance guidance and safety.
Everyday finance dialogue example
All conversation and AI reasoning are logged and reviewable by the user at any time.
※ Examples are for illustration; we do not recommend specific financial actions or decide for the user.
Asset management judgment log example
Every decision-support step is recorded in explainable form; the user can see what context and reasoning AI used at any time.
Transparent AI decision process
Signal generation basis
All indicator values (RSI, MACD, Bollinger, etc.) are exposed
Pattern analysis
Comparison with similar past patterns and outcomes
Dynamic thresholds
How thresholds are adjusted by market regime
Learning data
Trade history and performance analysis
Safety reasoning
Anomaly detection criteria, comparison with normal procedure, reason for extra confirmation
Version tracking
AI model version and training history are tracked so each decision can be tied to a specific model version. This allows analysis of how updates affect performance and rollback to a previous version if needed.
Version tracking structure
- Model version: Version info for each AI model
- Training history: Data and time of training
- Performance comparison: Per-version performance metrics
- Rollback: Restore to a previous version when needed
Verifiability: All AI call logs and learning data are stored locally for external verification; we provide real-time AI analysis, not pre-stored data.
Related technical docs
Explain, record, and learning structure linked to XAI are described in the documents below.