Generative AI and large language models offer enormous potential, but organisations deploying them face risks that traditional governance frameworks were never designed to address. An AI assistant that helpfully locates a policy exception can, with one follow-up question, become an accomplice that helps an employee circumvent controls. Guardrails that pass laboratory testing fail unpredictably when confronted with real users. Validation methodologies proven over decades for statistical models fall short when applied to systems whose outputs cannot be enumerated in advance.
Preserving the ROI of AI provides a practitioner's framework for managing these risks without paralysing adoption. Drawing on hundreds of hours of discussions with regulators, operations managers, and frontline users, Joseph L. Breeden, PhD, bridges model risk management, technology risk management, and business strategy into a unified approach for governing generative systems in regulated industries.
Original frameworks introduced in this book:
- The Batman Principle — a classification system that governs AI by use case, not by technology
- The LLM Validation Impossibility Principle — why traditional validation fails for generative systems, and what to do instead
- Staged deployment modelled on drug trials — a phased adoption methodology from sandbox to full release with defined exit criteria
- Assertions-based continuous monitoring — automated compliance checking across regulatory, ethical, and business policy dimensions
- Risk-adjusted ROI — treating governance as competitive advantage rather than overhead
Dedicated chapters address the distinct risk profiles of fine-tuned LLMs, retrieval-augmented generation (RAG), and agentic AI, each with specific validation and monitoring guidance. Real-world risk incidents — from the Sydney chatbot crisis to shadow AI exploitation and copy-paste infiltration attacks — illustrate how failures arise not from dramatic malfunctions but from subtle interactions between capable systems and the humans who use them.
Written for: Chief risk officers, model risk managers, compliance professionals, AI practitioners, and board members responsible for AI oversight — particularly in banking, financial services, and other regulated industries.
"Dr. Breeden charts a practical, easy path. Preserving the ROI of AI is the first book I've read that treats AI implementation and governance not as media or compliance theater, but as a genuine competitive advantage."
— Anju Patwardhan, Independent Board Director, FinTech Investor and Former Banking Executive
"A much needed honest discussion on the risks and major issues associated with Gen AI. Readers will find value in the structured approach to dealing with AI risk management."
— Naeem Siddiqi, Senior Risk Advisor, SAS Institute
