Matthew Murray

I put AI into production in places where a bad release has consequences. Right now that means PMI, where I lead product for Infinity, PMI's enterprise AI platform; before that it meant Dow Jones, JPMorgan Chase, and federal work at Booz Allen Hamilton. Four industries with four sets of rules, and the same question underneath every one of them: does this work well enough to ship, and can we prove it?

Models are rarely the hard part. Shipping is, because Security has to sign off on how retrieval is scoped and Legal on how data is handled, and then an executive has to look at the evaluation numbers and be willing to defend the release. Most of my time since 2025 went into building that machinery for Infinity: the evaluation and observability standard every release runs through, the governance policy for regulated content, the build-versus-buy calls across the tooling stack, and the team that operates all of it.

I came to product from data science, which is why I still read the eval results myself instead of waiting for the summary. At Booz Allen I built forecasting and NLP tools and a natural-language command interpreter for simulation workflows, years before LLMs made that easy; at JPMorgan Chase I modeled customer behavior and brand perception; at PMI I built the member recommendation platform; at Dow Jones I led the team that put ranking, risk, sentiment, and proposal-generation models in front of the advertising sales organization every day. PMI's Chief AI Officer asked me to come back in 2025 and build the platform organization, and I took the product seat to do it.

Before any of that, I spent a decade running restaurants. I ran dining rooms for Stephen Starr and The Ritz-Carlton with full P&L ownership, and I opened The Chelsea in Atlantic City from pre-opening through launch. Two things carried over: systems and tracking are what keep standards high after the opening-week adrenaline is gone, and nobody creates a good experience alone.

Leaving took a while. I started with math classes at Montgomery College while I was still running restaurants, then left the industry to study applied statistics full-time at UMBC, graduated magna cum laude at the end of 2015, and Booz Allen hired me the following February.

Idle Numbers is the notebook I wish someone had handed me in 2016. I write about what shipping AI inside a regulated enterprise takes once the demo is over: the decisions a spec never makes, evaluation you can defend in a room full of skeptics, how a team stays aligned after the PRD is signed, and when to build instead of buy. It's written for product managers and AI leads, and for the executives who sponsor them and have to answer for the results.

I live in Devon, Pennsylvania, and you can reach me on LinkedIn (opens in a new tab) or by email; I read both.