How we design, deploy and govern JAY — our AI Banking Officer
JAY is designed to be precise. When uncertain, JAY asks for clarification rather than guessing. Every transaction JAY initiates requires explicit user confirmation before execution.
JAY always identifies itself as an AI banking system. We never allow JAY to claim to be a human banking officer, impersonate a real person, or obscure that it is an AI system.
JAY is trained to serve all users equitably regardless of language preference, literacy level, gender, age, or location. We actively test for and mitigate bias in JAY's responses.
JAY is designed with hard limits. It cannot approve its own transactions, bypass fraud checks, override compliance rules, or take actions the user has not explicitly confirmed.
Voice inputs are processed in-session and not stored permanently. JAY's conversation logs are retained for 90 days only, and de-identified before any model improvement use.
JAY is trained to communicate effectively in English, Nigerian Pidgin, Yoruba, Hausa, and Igbo. We continuously evaluate JAY's accuracy across all supported languages.
JAY operates under continuous human oversight. Our compliance team monitors AI-assisted transactions for anomalies. Any transaction flagged by our fraud engine is reviewed by a human compliance officer before escalation or restriction.
JAY can always escalate to a human support agent when a customer requests it or when the situation requires it.
JAY's underlying models are evaluated before each production deployment against a benchmark suite of Nigerian banking scenarios across all 5 supported languages. We do not deploy model updates that degrade accuracy, compliance, or fairness metrics.
Questions about JAY's AI design: ai@jasonpaybank.com