Deploying AI in wealth management: predictable solutions for regulated firms

It fascinates me that the AI I use every day is agentic. But almost none of the AI I ship to my clients is.

My clients are wealth management firms: AFSLs in Australia, RIAs in the US, and the service providers that sit behind them. Some of my work is big-picture strategy, some is hands-on design and build. All of it has AI threaded through it.

In my own work, I live in an agentic AI world. I throw messy problems at AI models, let them explore, plan, suggest ideas, prototype, fix things, and converge quickly on good answers. It’s freeform, expensive in tokens, and unbelievably productive.

But that’s rarely what a regulated wealth management business wants from AI.

What my clients actually need — and what I ship — are apps that use AI predictably. Clear steps, defined boundaries, human oversight … but still hyper productive.

I deploy genuinely agentic AI judiciously, only where on-the-fly autonomous decision-making makes sense. One example is conversational experiences — say, allowing an advisor to interrogate their client data using natural language. That’s necessarily freeform.

But in most other circumstances, predictability is king.

I typically deploy AI only where it provides real magic that isn’t feasible with traditional code: extracting meaning from unstructured documents; interpreting natural language instructions; analysing large quantities of variable data.

For everything else, I use AI to design, build, test and deploy traditionally coded modules.

I’m finding this approach to be really powerful. Except where autonomous decision-making is absolutely critical, using AI to deploy AI predictably is a compelling pattern.