Guiding values
Measurable controls are necessary but not sufficient. Behind every control sits a value choice. When an edge case shows up no policy covers, we need a tie-breaker. Bizzi publishes five guiding values for AI decisions. They are not slogans. They are the criteria we use when the rulebook is silent.
Context
Section titled “Context”B2B finance runs on trust. Trust runs on consistency. A single AI mistake on a tax filing or a wire instruction erases years of brand equity. The values below exist so engineers, product managers, and the AI Board reach the same conclusion when they face a novel situation. Not because they read the same memo. Because they share the same priorities.
Definition. Customers rely on AI decisions and outputs.
Why it matters. B2B finance runs on trust. A single mistake costs an organization’s confidence for years.
How we enforce it. Every AI decision has a complete audit trail. Confidence scores appear in the UI. We publish model limitations rather than hide them.
Accuracy
Section titled “Accuracy”Definition. Technical quality clears the operational bar before a feature ships.
Why it matters. In accounting, a field-level error becomes a financial error.
How we enforce it. Extraction accuracy on critical fields (tax ID, invoice total) exceeds 99%. Every new model passes A/B and shadow-mode evaluation before it serves live traffic.
Accountability
Section titled “Accountability”Definition. Every AI decision has a named human who owns it.
Why it matters. AI is not a defendant. A system deciding without an accountable owner is a governance failure, not a feature.
How we enforce it. Every AI feature has a named Business Owner. The audit trail traces back to model version, dataset, and parameters.
Human-centricity
Section titled “Human-centricity”Definition. AI assists. Humans decide the consequential transactions.
Why it matters. Over-automating where financial risk is high is a category error.
How we enforce it. Human-in-the-loop is mandatory for high-value and anomalous transactions. Human-on-the-loop with a rollback dashboard covers the standard flow.
Continuous learning
Section titled “Continuous learning”Definition. Models, processes, and this framework adapt to reality.
Why it matters. Regulation, technology, and competitors move every quarter. Standing still is falling behind.
How we enforce it. The AI roadmap gets reviewed quarterly. BAGF receives minor updates every quarter and a major update every year. Internal red-teams probe for gaps on a fixed cadence.