The consultancy wanted AI in their finance workflows but kept hitting the trap that sinks most such efforts. Either you keep doing everything by hand, or you attempt to replace an entire process with a black box — and discover too late which steps genuinely needed a human.

They needed a way to think about automation that was incremental, honest about where AI helps and where it doesn't, and grounded in their actual workflows rather than a vendor's promise. Not a roadmap on a slide. A way to reason, step by step, about their own process.

Each finance workflow is modeled as a graph. Every node carries a swappable implementation — Human, AI, or Program. The reader flips a node from human to AI, runs the workflow, and watches execution either block on a remaining human step or flow through to completion.

The argument is the artifact. You can see exactly where automating one slot speeds things up, and where a human is still load-bearing. This was a focused advisory deliverable — a way to make a strategic case concrete and let the client reason about their own processes node by node.

Automate one swappable step at a time, keep humans where judgment lives, and let the workflow show you the order. Stated as a sentence, it's a platitude. Turning it into something a finance team could operate — an executable graph where flipping a node has visible consequences — is what made the argument land.

The interactive model does the persuading that a deck never could. It lets the client run the experiment themselves, on their own workflows, and reach the conclusion rather than be told it. Conviction that you produce yourself is the only kind that survives contact with a real process.