I’m using Gemini to improve my investment strategy, and it’s beating my portfolio advisor

As much as I dislike AI, it’s hard to disagree that it has its use cases. As a purely scientific tool, it’s a great way to sweep through troves of documents, research reports, news, and distill trends. While that makes it particularly useful for research, there’s one more sector that makes heavy use of it. The finance sector, particularly high-frequency trading, has been using AI and machine learning for a while now. Custom AI solutions are also used for predictive analytics and portfolio automation. However, those tools depend on highly specialized in-house AI tools. What I’ve been wondering is — can a general-purpose AI like Google Gemini be optimised to perform similar functions?

A few months ago, I decided to run a small experiment. Instead of blindly following every piece of advice from my portfolio advisor, I wanted to see what would happen if I brought Google’s Gemini AI into the mix. My idea was pretty straightforward. Take the same account statements I usually share with my advisor, feed them into Gemini, and ask it to give me an analysis. I also fed it mutual fund fact sheets and portfolio disclosures and asked it to run deep research surveys into the sectors and funds I was investing in. I was curious to know if an AI model could make sense of my investments, identify risks, and maybe even suggest strategies that were worth testing. And I’d put some money behind it as well. What happened next was rather interesting.

Using AI for investment advice largely works the same way as speaking to a financial advisor, and it starts with setting goals. Asking Gemini to just optimize my portfolio would invariably spit out vague advice because it has no grounding in what I expect from it or my risk appetite. So I spelled out what optimizing a portfolio meant for me. In my case, I wanted a strategy that leaned aggressively, prioritizing higher long-term returns over absolute safety. This would invariably result in an equity-heavy portfolio, but I was fine with that. I also defined my benchmark that the portfolio should aim to beat the Nifty 50 over a ten-year horizon while avoiding unnecessary duplication in the choice of sectors. I was fine with short-term volatility along the way if it meant better overall returns. Defining those goals upfront gave Gemini a direction to work with. It’s an excellent strategy to have when working with an LLM for any task, and particularly so if you’re working with financial data.

Next, I gave Gemini access to my holding statements. Based on my experience, Gemini didn’t fare too well with Excel documents, and feeding it raw data worked out much better; it was able to make sense of all the data in seconds. However, your mileage may vary. Regardless, I would highly recommend labelling essentials like the fund name, code, current value, invested amount, start date, returns so far, and more in a structured format. This makes it trivial for the LLM to understand your current holding statements and avoids unnecessary to-and-fro in cleaning up bad parsing of data.

Of course, Gemini doesn’t automatically know everything, or for that matter, anything, about each fund. So, your next step should be to enrich that data pool. A good place to start is to run a deep research scan on every single fund in your portfolio. This helps it fill the gaps in expense ratios, exit load rules, and category averages. I found this to be a fairly taxing process since Deep Research scans run separately from the regular chat, but the extra information is worth it, even though you’ll have to manually copy over the insights and learnings.

Once Gemini had a clear view of the data, the next step was to actually start making sense of it. To kick things off, I started asking it to analyze the holdings. It summarized my allocation between equity, debt, and hybrid and pointed out the risks of concentrating my investments in a specific type. It’s interesting to note that even when certain funds were underperforming, it didn’t ask me to make an exit. It understood the context of small-cap quant funds and suggested holding through the downturn. Broadly speaking ,though, the analysis was an interesting departure from the somewhat vague statements from my advisor. While he’s great at helping me generate wealth, it’s understandable that a human doesn’t always get down into the weeds with explaining why a decision is being made. Instead of the vague “stay invested in this fund” that I might hear from an advisor, Gemini showed me exactly which funds had lagged their category and explained why I should or shouldn’t hold on to my investments. Even as a purely educational exercise, this is very exciting.

Once I had the basics covered, I wanted to see what Gemini could do if I asked it to actively restructure my portfolio. My goals remained the same – focus on aggressive growth over a long-term horizon. Generally speaking, I’ve traditionally stuck to an 80/15/5 percent split between equity, debt, and hybrid funds. It can be a risky gamble with potential for volatility in the short term. Would an AI tool, especially one that isn’t specifically optimised for guiding financial decision making, give me such a risky suggestion? There’s only one way to find out.

Gemini’s restructuring was interesting in ways that I didn’t expect. To start with, it cut out funds that were duplicating exposure and increased allocation to funds that had historically outpaced their category averages. Sounds obvious, but it dug deep into the portfolios and helped me segment funds that I thought were fairly dissimilar, but clearly played in the same verticals. This bit caught me by surprise.

Moreover, it didn’t just shove me into the risky small-cap territory for the sake of being aggressive. It kept a strong base of large-cap index fund exposure while adding mid-cap and thematic equity funds that could realistically outperform over a ten-year horizon. Yes, it still gave me a risky portfolio, but with sufficient reasoning to believe that it might actually outperform my wealth manager’s advice. It also recommended adding in tangential sectors like renewable energy that I hadn’t invested in so far. That’s ambitious, but once again, grounded in research.

Unlike dedicated tools, Gemini has no way to backtest data beyond running a theoretical simulation on existing historical data. I asked it to show me potential returns based on historical data, and if I’d swapped out some of the existing holdings for its recommendations. Turns out, Gemini’s suggestions would’ve made me more money. But hindsight is 20/20, and I’ve tested Gemini’s suggestions with a separate portfolio. Looking at the ongoing XIRR over the past few months, Gemini’s picks are currently outperforming my advisor’s recommendations by just under 1%. That’s encouraging, though short-term outperformance doesn’t necessarily translate into long-term success. When it comes to investments, even a 1% difference compounds into significant wealth creation over time.

It’s been a few months since I’ve been running this experiment, and the results have been promising. But that’s not to say that you should go all in on Gemini or ChatGPT to manage your money. The fact of the matter is that it took a fair amount of handholding to get to this point. Sure, between the holding statements and all the research, Gemini was able to give me a lot of clarity and a way to get a better sense of my portfolio. It identified overlaps, reduced investments that weren’t giving me the results I wanted, and leaned harder into aggressive bets that actually lined up with my investment goals.

But, by and large, I’d have likely achieved the same outcome if I spent a sufficient amount of time researching the funds on my own. What Gemini did was accelerate the process dramatically. What would’ve taken me a weekend or more, it pulled off in minutes. Not just that, it was able to identify gaps that I, as a non-professional, wouldn’t have thought of. That said, Gemini is still limited by the data you feed it and its training cutoff, and it doesn’t have access to real-time mutual fund performance. It’s not a financial advisor and you shouldn’t treat it as one outside of trying to make sense of your portfolio.

If it wasn’t obvious, I’m not firing my wealth manager just yet. While LLMs run on data, a good advisor has insights into market trends and the pulse of the people, policies, and more, all of which have a direct impact on the markets. Still, as an experiment, it was worth the effort to tap into Gemini to analyze my investment portfolio. It helped me understand it better, and, to be fair, did actually give me suggestions that made me more money. Definitely something I’d recommend checking out next time you’re rebalancing your portfolio.