Navigating the Risks and Rewards of AI in Advice

As AI adoption accelerates across advice practices and licensees, so too does the opportunities and risks. While the technology holds great potential, managing its use responsibly is critical. 

Misuse and lack of guardrails are major risks 

The biggest threat is misuse. HUB24’s Head of Innovation Lab, Dr Evan Morrison, warns wrapping prompts around ChatGPT and calling it robo advice is risky: 

“You are effectively saying that humans don’t matter in this process and that chatbots can actually do what a human does.” 

Without proper oversight, such applications can cause lasting harm. Guardrails like accuracy checks, privacy protection, and compliance filters are crucial as experimentation expands. 

Data quality and clear processes are foundational 

Many firms still struggle with fragmented data and undocumented processes, limiting AI’s effectiveness. 

AGS Financial Group’s approach shows the value of building structured, logical systems before introducing AI. 

“There are so many things in our business that are just logical steps and for some reason we have a person doing them,” said Paul Bolstad, Financial Adviser, AGS Financial Group. 

Outsourcing helps manage cost and change 

The cost and pace of AI development leave many smaller firms uncertain about their path forward. 

“In the good old days, you could run an RFP [Request for Proposal] process, pick a piece of software and know that with a good process in place, you’re generally going to be ok,” said Nathan Jacobsen, CEO Vital Business Partners. “But now it’s out of date before you’ve turned it on.” 

Yarra Lane CEO Nicholas Perrett agrees outsourcing is key. 

“That’s why we get the big boys with thousands of staff and resources to stay ahead of the game for us.” 

Evolution of AI will help iron out is challenges 

Dr Morrison sees two key developments: generative AI and thinking/reasoning models that follow instructions and reason through tasks. 

“We also think RPA will move down the path… where you describe a task and then share access with a language model or agent,” said Morrison. 

These models aim to reduce hallucinations and improve reliability by reasoning before responding, but today HUB24 is also testing open-source and smaller models like Phi-4 to improve trust and accuracy.