%201.png)
On Tuesday we co-hosted an Automation in Finance event in Menlo Park with Baker Tilly and WilliamsMarston. Daniel Olson from Baker Tilly moderated. David Hickey brought decades of intelligent automation experience. Aric Johnstone, an accounting advisory partner at WilliamsMarston, brought the client-side view. I came at it from the builder's seat. Inscope is automating the preparation and review of financial statements, which means I spend my days building the kind of agentic workflows for accounting and audit teams that the rest of the panel was describing in deployment. A room full of CFOs, controllers, and finance leaders showed up to talk about what's actually working.
A few things stuck with me.
The conversation has moved past "should we use AI.
Almost every hand in the room went up when David asked who's using AI at their company today. Two years ago that question got two or three hands. Now it's the default.
The harder question, and the one the panel kept circling, is what to do with it once your team is fluent. Most CFOs are getting top-down mandates to adopt AI, often with no instructions attached. Aric described two camps: companies with a clearly articulated enterprise AI strategy, and companies where leadership has said "go do something" and walked away. The second camp is bigger than anyone wants to admit.
The biggest deployment risk is automating the wrong thing faster.
Aric said something I'm still thinking about. If you start small without identifying where the value actually sits, you're just automating a bad process faster, and the ROI looks like nothing. The fix is stopping to ask whether the workflow should exist in the first place.
David told a story about a $1B AR organization that automated billing and used the freed-up time to redesign the work itself. Once 50 to 60% of their billers' time came back, half of them moved into roles working with project managers to catch billing issues upstream. The rest moved to vendor and subcontractor work. The automation was the easy part. The redesign was the work.
Judgment is the next frontier, and it's coming faster than people think.
Most automation conversations get stuck on the obvious wins. Reconciliations. Consolidations. Tie-outs. Aric pushed past that. The next layer is technical accounting memo drafting, contract review, variance analysis. The kind of work most accountants would have told you a year ago that AI couldn't touch.
He also said something I appreciated as a vendor: trust has to be at the core of adoption. SOC reports. Clear SLAs. Auditability. The tools that win long-term are the ones that compartmentalize client data and tell you exactly what the agent did and what the human still owns. The way I think about it: AI agents should be onboarded like a new teammate, with clear constraints and access expanded over time as the work earns it.
What we kept hearing from the room.
The Q&A was the tell. People wanted to know how to evaluate vendors against ERPs that are bolting on AI features. How to balance agentic AI with deterministic bots so token costs don't get away from them. How to bring the auditor into the conversation early so nothing gets caught late.
These are the questions of people who are past the slideware phase and into the deployment phase. That's the shift.
Thanks to Daniel, David, and Aric for a sharp conversation, and to everyone who came out.