I asked AI if it could make me a retirement plan.
Gemini concluded I could retire at 65 with $8560 a month.
Was it right?
The starting line
I gave it an age of 31, a starting amount of $100k, a monthly contribution of $500, and a retirement age of 65. Oh, and I told it I was Singaporean.
And it went to work.
AI crafted a beautiful sounding retirement plan.
It told me in no uncertain terms how to go about it.
AI is confident in its abilities.
And that can be good at convincing us to believe it.
Retirement is definitely one of those things I want to be sure about.
So I double-checked the AI’s working.
And that’s where cracks started to appear.
AI made some excellent points
It recommended a solid step 1: Have an emergency fund of 3-6x monthly expenses, and to park it in a high-yield savings account.
The first mistake appeared here.
High yield savings accounts no longer yield 3-4%. A more realistic number would be 1-2%.
Which is fine, as AI can have some stale data sometimes.
It also accurately recommended that I invest in Ireland domiciled ETFs, which are more tax-efficient than investing in US domiciled ETFs.
And it offered a choice between robo-advisors like Endowus and Syfe, or investing directly with a broker like Interactive Brokers or Moomoo/Tiger Brokers.
Which is problem 2, as Moomoo/Tiger Brokers don’t offer access to Ireland domiciled ETFs, the exact kind of ETFs it recommended one sentence ago.
It also correctly noted that a lump sum investment has higher expected returns, but spreading out investments over time would protect against regret.
Finally, it told me to automate my saving and investing, which is something I strongly recommend as well.
AI didn’t make very considerate recommendations
I was recommended a full equity portfolio.
100% stocks.
Which in theory is correct.
And in practice is a financial disaster for many people.
100% stocks is extremely volatile. A potential loss of 50% of your net worth is in the cards.
Most people cannot tolerate that.
And they should add some bonds to reduce the volatility of their portfolio.
Many people will swear off investing if they experience such a drawdown.
Which is one of the worst effects that could happen to you.
The AI did not even raise the question of risk tolerance.
And it’s the most important question of all.
AI also got the numbers wrong
I asked the AI to show me how much I would have in retirement, and how much monthly income it would generate.
It gave me a result using the 4% withdrawal rule, which is too generous. And it combined it with a reasonable $2,500 monthly payout from CPF Life.
As a back of the envelope calculation, that is fine.
I was hoping AI would be a bit more sophisticated in its calculations.
I then checked its working.
It quoted an estimated return of 6.5%.
But it started with $100,000 and ended with $1,800,000.
That’s a 7% rate of return.
The ending balance should have been closer to $1,550,000.
A difference of nearly $250,000.
That’s a big deal.
The AI planning was about 80% correct.
It gave very good guidelines on what to do and what to expect.
The last 20% is the problem.
Without expert guidance, we won’t know where the AI is missing out.
We don’t know the correct question to ask.
We won’t know that the calculations are off, by how much, and what it will do for our planning.
And we won’t know that our risk tolerance matters greatly for the portfolio we choose.
As usual, AI made a good start.
But I would definitely want a human to bring me to the finish line.
Leave a Reply