Matt Rosfelder
I help regulated mid-market firms put AI to work where it actually pays off — and keep it out of the places it creates more risk than value. Strong on the business and the regulatory reality, technical enough to know what's real: not a coder for hire or an ML researcher, but the person who keeps the investment aimed at value and out of trouble. The residual risk on every engagement is owned by a named human — this one. Solo is the feature: no offshore black box, no handoffs — the person you meet is the person who signs.
Most mid-market AI attempts stall — and not because of the technology.
Companies in lending, insurance, logistics, and other regulated spaces are under pressure to "do something with AI." The usual blockers are messy data, no governance, and pointing AI at problems a simple script could have solved. The result is wasted spend and compliance exposure nobody asked for.
Twenty-plus years where the documents have to be exactly right.
Business analyst and product owner inside regulated financial services: lending compliance at BMO (including Department of Defense rules for military lending), risk analytics at PNC (Basel II capital work after the financial crisis), and product ownership at Vertex. Audit trails, fair-lending scrutiny, and data that has to be exactly right — the world that makes AI feel risky to these firms is the world I come from, and that's why I can de-risk it.
Before that, a first career in investigations — surveillance and background work, then computer-forensics analysis and reports for law firms and federal investigators. The craft never changed: gather the evidence, build the case the patient way, present it so it holds up under scrutiny, and know — before anyone else does — when something doesn't add up. WTD is that discipline pointed at AI agents. The detective frame isn't branding; it's the operating model.
Four rules, applied in order.
- Say no first Most proposed use cases don't need AI. The ones a script or basic automation handles better get a script, so budget goes to the few problems that genuinely require it.
- Fix the foundation No AI on messy data. The data is confirmed ready before anything goes live.
- Keep humans accountable AI does the heavy lifting; people stay responsible for the decisions — non-negotiable in regulated work.
- Capture how your best people decide The goal is a system that reflects your firm's judgment, not a generic model's.
An engagement runs: readiness assessment → a focused pilot on one high-value workflow → ongoing fractional advisory as it scales. Fixed-fee projects and monthly retainers — not hourly. The pilot shape and terms are on the engagement page.
Start with one workflow.
Chicago, IL. The first conversation is about picking the workflow and the metric.