How to DIY AI for financial advice firms

Back in Sydney after meeting some very creative financial advice firms who are dabbling in DIY AI. I love the entrepreneurial spirit, and I shared a few tips from a technologist’s playbook to make early tinkering smart, secure and scalable.

👉 Invest in a dev environment

If you’re experimenting with AI integrations — whether to your CRM, document store, or advice platform — don’t do it in production. Use a sandbox org, or even a dummy account set up purely for testing (I’ve done this with Microsoft 365 to safely prototype AI workflows with Teams and OneDrive).

A little upfront investment here avoids the headache of corrupted client files or accidentally triggering live processes.

👉 Create safe data to work with

It kind of obvious, but don’t feed PII into an experimental AI workflow. Instead, use AI to redact beyond recognition, or better still, generate fully synthetic client files: fake names, mocked-up advice documents, dummy fact finds and meeting transcripts.

With realistic fake data on hand, you can go hard, experiment freely, fail fast and iterate. Then you need only engage compliance seriously once you have a working proof of concept.

👉 Use multiple AI agents as a team

One of the most effective patterns I’ve seen is to split the work between agents, in your own virtual dev team. Make one the solution architect, another the developer, and another the QA specialist.

The architect writes a spec, the developer builds to it and documents its work, and the QA agent creates and runs tests. Feeding the outputs back into the architect gives you a self-checking cycle — modular, well documented, and typically far stronger than asking a single agent to “just build it.”

These are early-stage habits, but they make the difference between a fragile experiment and a foundation you can build on. 👍