Why Companies Are Already Abandoning LLMs — with FICO's Mike Trkay

Everyone's racing to bolt AI onto what they already do. Mike Trkay's argument is that this is the exact mistake companies made with the move to cloud — and it's about to cost them millions to learn it again.

Mike is CIO and Chief Customer Officer at FICO, where the analytics run billions of decisions a day and power roughly 80% of fraud detection. So when he says 95% of AI projects never make it to production, it's not a hot take — it's the view from inside a company that has to make this work at scale, under real latency and regulatory constraints.

We get into why companies are already pulling workloads back off large language models, what actually separates automating a process from transforming it, and why "you're just doing a bad process faster" might be the most useful sentence in enterprise AI right now.

Chapters

0:00 — Introduction

1:23 — Why FICO's CIO Is Also the Chief Customer Officer

4:52 — The Office Space Problem: Why CIOs Are Really Translators

6:02 — "Conduit in Chief": The CIO's Real Job

6:45 — How CIOs Accidentally Become Cost Centers

7:40 — What FICO Actually Does (It's Not Just Credit Scores)

12:59 — How Missile Guidance Tech Became Fraud Detection

17:59 — When LLMs Can't Meet 150-Millisecond Latency

21:32 — Why AI Strategy Is Harder Than It Looks

24:50 — The Cloud Parallel: 5% Alignment and Why AI Projects Stall

27:08 — "You're Just Doing a Bad Process Faster"

31:44 — LLM Repatriation Is Already Happening

37:40 — A FICO Score for AI Decisions

43:10 — Global Regulation and Staying Ahead of Compliance

45:22 — The Australian Banking Example: Focused Language Models in Practice

52:08 — From Cost Center to Innovation Driver

54:40 — What's Next: Agentic Architecture and Focused Sequence Models

57:49 — "What's the Opportunity Today?": The CIO Mindset Shift

Guest: Mike Trkay, CIO & Chief Customer Officer, FICO — LinkedIn