Mastercard operationalizes enterprise AI governance with a lean five-person team

Mastercard operates at global scale in payment and transaction infrastructure, with many regulated banking relationships. As AI adoption accelerated, the volume of systems requiring governance oversight doubled every year — then grew 60% in 2024 alone. Teams across the company were launching or procuring AI systems without always notifying the governance team, creating a backlog of legacy products with inconsistent documentation, monitoring, and ownership. Regulated banking customers expected detailed model performance disclosures, yet record-keeping could be sporadic. Without structured oversight, operational risk compounded.

John Hearty built Mastercard's AI governance function from a single-person operation into a lean five-person specialist team, anchoring the program in influence-building rather than enforcement. The team embedded governance earlier in the development lifecycle by introducing a risk scorecard — completed by product owners before any AI system is built or contracted — covering data understanding, techniques, and decision-making agency. The team co-created a model documentation template with data science peers, developed a bias-testing API distributed company-wide, and established mitigation requirements for LLM evaluation. Partnerships were structured as win-wins: governance work addressed developers' pain points while delivering the upfront model transparency regulated banking customers required.

A five-person team now governs AI at enterprise scale across Mastercard's global operations, covering bias mitigation, efficacy assurance, and transparency for regulated bank customers.

• 60% growth in AI systems under governance in 2024, following consecutive years of doubling
• 5 specialists cover architect, model risk, technology, communications, and R&D leadership roles
• Regulated banking customers receive detailed model documentation pre-contract, improving trust
• Internal cross-functional partnerships produced reusable tooling — bias-testing API, documentation templates, mitigation requirements for LLM evaluation

The program shifted AI governance left in the development cycle.

• Small, highly skilled teams can govern AI at enterprise scale when organized around influence and enablement rather than top-down enforcement.
• Shifting governance left via scorecards and tooling — completed before build or procurement — prevents the legacy backlog problem from accumulating.
• Win-win partnerships with developer and data science teams are more durable than compliance mandates.
• In regulated industries, upfront model documentation shared with customers pre-contract builds trust and reduces friction at audit or renewal time.
• Hiring for curiosity and self-awareness matters as much as technical credentials when assembling a governance function that must earn influence across a large organization.