AI and financial planning: what to hand over, and what to keep
AI can crunch the numbers and read the fine print. It cannot carry the responsibility, or read a family. Where the line falls.
AI is upon us, and financial planning is changing fast. Anything with a process can be automated and delegated away, so many tasks will be. But accountability and responsibility cannot be delegated away. When things go badly wrong, in a court of law, a legal person is held accountable, not an algorithm.
That means humans need the competence to check what AI produces, and to take responsibility for it. If AI-driven advice leaves a client worse off, who is liable? The developer, the financial institution that deployed it, or the representative who relied on its output? It is like trying to assign blame when a self-driving car crashes.
AI and the planning process
Much of financial planning in Singapore runs as a Proposal-Acceptance Arrangement. The representative proposes; the client must accept before anything is executed. If there is a dispute, the client carries some responsibility for accepting. If there was a factual error behind the proposal, the representative is at fault. Given that, there is no harm in getting additional proposals from AI to consider, is there?
Financial planning follows a process, and parts of it AI can already do faster, cheaper and often more thoroughly than a person: collating information, crunching numbers, rule-based analysis, mass customisation at scale. With enough context, it is like a personal finance coach who never sleeps and never judges your late-night online shopping. That makes basic guidance more affordable and widely available, which matters most for people who felt planning was out of reach.
Use it to read the fine print
AI can find patterns and correlations a human would miss. It is becoming a formidable co-pilot, a research assistant working tirelessly in the background.
You can give it a product's documents, such as the benefit illustration and product summary, and ask the questions that matter to you. Tools that answer from the documents you upload, rather than from the open internet, suit this well. They still make mistakes, so ask it to cite the page and paragraph, and check. Two questions I use for any product:
- What are the key features and benefits of this product, and its downsides?
- Give me a worked example of who this product would suit.
If you want to compare plans, pair this with the FDM method: let the AI do the heavy lifting on the documents, and keep the weighting of what matters to you.
One caution. A sensible rule is not to upload personally identifiable information to external tools, and to anonymise anything sensitive if a tool must be used. We cannot be sure how these systems handle what we give them.
Where people still matter
Life does not fit neatly into an algorithm. Divorce, legacy planning, business succession, care for elderly parents: these situations are personal and emotionally charged. They need nuanced discussion, ethical judgement and an understanding of human relationships. How intense a sibling rivalry is, or who the family's preferred carer is, can change how a legacy plan should be allocated. AI is not equipped for that kind of judgement.
At its core, financial advice is built on trust. Clients confide their hopes and fears about money. That trust is earned through consistent contact, empathy and reliable guidance, which AI cannot replicate. You trust an algorithm with data; you trust a person with your future.
So split the work. Let AI handle the cognitive parts, and the person handle the emotional parts. With less time wrestling spreadsheets, a representative can spend more time listening.
PS: This piece includes some work from AI, heavily edited for authenticity.
Questions
Can AI do my financial planning?
It can do much of the processing, but not the accountability: when a decision goes wrong, a person or firm is responsible, not an algorithm.
How can I use AI to read an insurance product summary?
Give it the product summary or policy document and ask it to list the benefits, exclusions, charges and what happens if you stop paying, then check each point against the document itself.
First published on isaacfang.com on 18 July 2025, with a tip from Isaac's post of 9 September 2025 on reading product documents with AI. Tool names removed. Lightly edited for this site.