Large Tech Company X needed to accelerate code deployment across its internal cloud, and reduce manual overhead. Using machine learning to optimise deployment processes had the potential to accelerate rollouts while minimising the 'blast radius' and impact of bugs, but service owners were concerned about maintaining the availability of revenue-critical systems.
Effective model management, including identifying key failure modes and more transparent reporting, enabled closer monitoring of deployment processes, turning a 'block box' process into one that service owners could trust. This unblocked adoption with the owners of the some of the highest revenue systems in the industry, accelerating deployments while reducing overheads by 75%.
Regtech Startup Y targeted Tier 1 banks. But these large financial institutions require their vendors to meet the same Federal Reserve Model Risk Management standards applied to their internal models (SR 11-7) These regulations can include providing a full audit trail and human-readable explanations of all decisions. A modular AI design provided accurate and flexible decisioning with full human-readable explainability, while also reducing the volume of hard-to-gather training data required to reach accuracy targets. And adopting a common documentation standard enabled the product to pass internal model management review with Tier 1 customers, and external review with US and UK regulators.
This level of transparency also gave customers enough confidence in the model outputs to reduce QA rates below that required of equivalent decisions from human analysts, further increasing RoI.
Fintech Z uses AI throughout their successful product suite, but blue chip investors required a robust AI governance policy as a condition of their investment. We worked with legal and product leads to define a suitable policy, worked with engineering squads to define a lightweight process they were comfortable with adopting, and gave them guidance on prioritising risks, reducing the the overall engineering overhead of getting new models into production.