MAI case studies
MAI · Put your knowledge to work in context.
The MAI is a vendor-neutral certification-style credential for people who build, deploy, or govern artificial intelligence systems in an organisational setting. It suits machine-learning engineers moving from notebooks to production, data scientists who need to own the full lifecycle, technical product and project managers coordinating AI delivery, and risk, compliance, or governance staff who must review AI systems without writing the code themselves. The exam is organised around eight domains: AI and ML foundations; data preparation, feature engineering, and model evaluation; neural networks and deep learning; generative AI and large language models; AI ethics, safety, and governance; solution design and deployment; operations, monitoring, and maintenance; and business strategy and AI project management. Foundations carry the largest single weight, but the applied domains together dominate, so breadth matters more than depth in any one area. The MAI case studies are 5 cases and about 3,611 words.
1Default Model That Looked Excellent Until Production
Northwind Lending built a gradient-boosted default model on 18 months of loan applications. The portfolio default rate is 3.1%, and the team reports validation ROC-AUC of 0.94 using a random 80/20 split.
762 words, about 4 min2RAG Assistant With Weak Retrieval and Unsupported Citations
Helios Support operates a retrieval-augmented assistant over 12,400 help-center articles for a SaaS product. It handles about 400 conversations per day, with 38% of answers containing a citation and 22% escalated to humans.
686 words, about 3 min3Vendor CV Ranking in a High-Risk Employment Context
A European retailer receives 45,000 applications per year for hourly store roles. It licenses a vendor's CV-ranking model that outputs a 0-100 fit score.
695 words, about 3 min4Edge Vision Defect Detection Under Cost Asymmetry
A manufacturer wants visual defect detection on six assembly lines. Each line produces one part every 1.8 seconds and has four cameras.
742 words, about 4 min5Churn Model Monitoring When Labels Arrive Late
A subscription analytics company deployed a churn propensity model for its customer success team 11 months ago. At launch, holdout ROC-AUC was 0.82 and calibration error was 0.03.
726 words, about 4 min