I Heard a CEO Make Sense About AI at Yale

I spent a day this week at "The Great Redesign," an AI conference at the Yale School of Management hosted by the Data & Trusted AI Alliance and Yale's Program on Stakeholder Innovation and Management. It was built for business leaders, and it went straight at a problem anyone with an AI budget is starting to feel: adopting AI, scaling it and getting value from it are three different things. The organizers showed a PwC survey of CEOs from earlier this year: only about 12% say AI has both cut their costs and grown their revenue.

There was a lot of good material, and I'll write several pieces from it. This one is about one speaker.

I hear a lot of business leaders talk about AI, and I read a lot of their announcements. Most of it is platitudes, and I'll leave the names out. It reads like a press release and doesn't get at what's going on underneath. So I'll admit it: I walked out with a bit of a corporate crush on Rob Fauber, the CEO of Moody's. He was specific and clear, and on staff and cost he said things most CEOs don't, and he was right. It wasn't pitch perfect. But he sang a pretty good tune.

Moody's isn't replacing its analysts with AI

Fauber says he gets asked all the time whether Moody's will replace its credit analysts with AI. His answer, in effect, is no, because the analysts are the advantage. They talk with nearly every company that issues debt and every investor that buys it, and in those conversations they learn things the market doesn't know. Until now, most of that went into analysts' heads, notebooks and Word documents. Moody's is starting to capture it in its AI platform.

He put the cost side bluntly: "We're not going to cost-cut our way to greatness."

I'd take it a step further than he did. That information comes out of relationships - the trust, the discretion, and the judgment about what a CFO's comment means in context. You can write some of it down, but notes go stale fast, and a stack of notes plus an agent doesn't add up to an analyst. What AI does here is take over the mechanics of getting, filing and finding the information. Analysts get more time for what they do best, which is building those relationships and the trust, context and depth that come with them.

A CEO who knows what a knowledge graph is

Fauber is the first CEO of a big company I've heard use the term knowledge graph, and use it correctly. A knowledge graph is a map of facts and how they connect. Over the past decade Moody's has spent billions building and buying what he described as the world's largest company knowledge graph, connected to its credit, catastrophe and cyber risk models.

Knowing you're sitting on valuable knowledge is one step. Understanding how to make it usable is another, and he seems to understand both. On the ratings side, what analysts learn goes into a shared knowledge system on Moody's AI platform, which he says makes every analyst smarter over time. I'd add that a system like that only works if people keep it current and supervise it. Once knowledge lives there instead of in separate heads and teams, more people can use it, not only the specialists. And connections show up that couldn't be seen when the knowledge was split up.

Weighing the cost of doing nothing

His phrase for balancing speed against safety was "fast and safe." He said Moody's has "been very good at analyzing the risk of action, but not the risk of inaction." I like that concept - risk of inaction. I'm using that. Perhaps I'll try to measure it.

In my corporate experience, risk ran in one direction. The default was no, and an idea had to prove it cleared the bar. Weighing the cost of not acting against the cost of acting is a balance, and both sides cost something. Later that day, Sarah Bird, Microsoft's chief product officer for responsible AI, gave an example of the other side: companies that spent billions building their own AI models from scratch, then watched a smarter, safer wave of models arrive a few months later. That investment was lost. Go too slowly, though, and someone else can step in between you and your customers, or your team falls behind how the work gets done.

His fix was practical. Borrowing a phrase from J&J's CIO, he called it "secure by design": bring legal, compliance and security in at the start of a project instead of the end, and put both speed and safety on the company's scorecard so people see that both count.

Keeping junior analysts

Fauber said the ratings staff is shaped like a pyramid, the way it is at banks, law firms and consultancies, because rating debt takes a lot of administrative work. That work isn't where the value is. He expects Moody's to move toward something closer to one junior per senior, more like an apprenticeship, and said they still need to train "the analyst of the future."

That's one of the few times I've heard the CEO of a major company say out loud that you still need junior people. It ties straight back to that knowledge system. Captured knowledge goes stale unless people keep it up, handle the cases it doesn't cover, and answer for it. Those people have to come from somewhere.

With the clock running out, he didn't get to the harder part: whether a thinner bench of junior people is a business risk, and what kind of juniors, how many, and when to invest in them. I'd want to hear that plan.

Bonuses and the view from the top floor

Fauber told his board that "the technology is no longer the gating issue with a lot of what we want to do." Adoption is. In early 2023 he told all 14,000 Moody's employees that he expected every one of them to be an innovator with AI. Moody's raised bonus funding for everyone, 97% of staff finished an upskilling curriculum, and he built a working prototype himself in about an hour and showed it on a company-wide call. More recently, Moody's changed how it splits bonuses to pay more to what he calls "5x" performers.

This is where I break from him. Incentives move behavior, but they're a blunt tool. In his closing advice, he started with empathy for employees worried about their careers - "how could you not be worried about it?" - then told them they have agency and the tools, so they should be one of the 5x people building with AI.

That advice comes from the top floor. Senior leaders tend to be proactive, confident and well resourced, and from there it's easy to assume tools plus agency is all anyone needs. On a working team it takes more. Before you use a carrot or a stick, find out why people aren't there yet:

• Do they feel safe trying, and safe getting it wrong while they learn?
• Have they been trained on their own work, not a demo?
• Do they trust where this is going, including what happens to their job?
• Do they have the tools, the licenses and access to the data?
• Do they have the time?
• Are the expectations realistic?
• Does their own manager use it?

Only after all of that is it fair to conclude someone won't or can't adjust. Companies tend to rush past most of that list and go straight to "get on board."

Personal note: I tried that approach with a senior member of my team, and it backfired on me. Not recommended.

What Moody's has right, and what's left

Fauber knows where his company's advantage comes from: analysts, the relationships they keep, and a system that turns what they learn into something everyone at Moody's can use. That's further than most leaders I've heard. What's left is the employees he's asking to change how they work. Moody's has given them tools, training and a bigger bonus pool. How much it gets back from AI depends on finding out, team by team, what's still in their way.