Zerodha has integrated Claude AI to help investors analyze and manage their portfolios through automation (similar to Chatgpt).
That’s a pretty big leap in making professional-grade advice accessible for retail folks, and hats off to you and your tech team Mr Nadh for pushing the boundaries!
My question is:
Do you see Claude eventually evolving to handle tax optimization and personalized goal-based planning, or will that still need a human advisor alongside the AI?
As more investors start trusting AI tools, how will Zerodha ensure the advice stays unbiased and doesn’t accidentally push people toward overtrading or excessive risk-taking?
Super excited after trying his out, and curious about the longer-term roadmap! Thanks for driving such innovation.
If we look closely, a lot of credit for this
goes to the poor quality of advice provided by most professionals currently involved in the business of providing retail individual investor financial advice.
Simply regurgitating the popular fads seen in the “News” and social media,
is a task easily accomplished by LLMs trained on the same public sources.
LLMs can already do that. Here’s a recent ongoing topic-thread How to invest 40 lakhs for retirement corpus in 20 years. For now, the challenge continues to be the individual though. This is similar to the consultation services offered by numerous institutions for ITR filing. The individual continues to be responsible for any errors/omissions.
LLMs customized for the purpose (training, fine-tuning, custom system-prompt) to make them less sycophantic and more likely to provide comprehensive financial advice, can definitely replace the human in the loop. However, human financial advisors continue to be cheaper (for now), unless one has a huge user-base (like Zerodha does) to amortize the cost of training/running LLMs over lacs of users. Also, humans are required in the loop legally (eg. for representative tax filing).
Couple of ways this is possible already even with current-gen LLMs -
Add guardrails (static checks on user-input or LLM-output) around user-LLM interactions to avoid potential red-flags.
Note: The above comments about LLMs are based on experience of having done these and evaluated the results. However, am not a financial advisor. Also, am not looking to create a roboadvisor product/service using LLMs. So currently not pursuing this any further.
I keep asking K about what all this progress in AI means for our business. It feels like we’re at the very beginning of a massive shift in how financial services will work.
At some point, I think all of it: investing, trading, banking, payments, will happen through custom AI-powered apps built by users themselves.
In that world, what’s the role of a broker? Likely, we get reduced to just being the “pipes” connecting users to exchanges and back-office systems. The only way to stay relevant is to make sure we’re the best pipe: fast, efficient, reliable, and invisible when it matters.
That’s why over the years, K and the tech team have been obsessively making our APIs faster, more scalable, and future-ready. Even if these improvements don’t immediately change a customer’s trading or reporting experience, we’ve chosen to fix every possible bottleneck today, not later.
Tools like ChatGPT and Claude make it clear this shift isn’t an if but a when. It might take a year or a decade, but it’s inevitable. Human advisors will still have a role, mainly to help people stick to what these tools recommend.
As for how things will evolve, the answer is grey. No one knows. Our approach: stay curious, watch for trends, and act where it makes sense. For example, we’ve intentionally held back on enabling AI-driven order placement.
In a future where everything is automated, trust and infrastructure will be our only real moats.
Curious to know what all usecases are you all finding interesting with the claude integration. Honestly I couldn’t find anything insightful that I can’t get from console.
In fact found a lot of factual errors when I asked claude to critique/comment on my option trades… https://claude.ai/public/artifacts/32bac6ad-8dad-4d92-8d88-eea85c8040f1
For instance, the drawdown mentions are incorrect. Current market levels at 27k as per claude, again not sure are basis which index. Feedback is very generic (had the same feeling when I asked for comments on my stock portfolio). I mean it was fun playing around with the assistant, but for anyone comfortable with excel sheets would rather download console reports and do analysis than depend on claude’s blackbox analysis and generic inputs.
PS: I know the analysis/ insights quality will drastically improve with time, but my current experience was not anywhere near all the hype around AI as personal finance assistants.
Kite API has been reliable and excellent for many years. I genuinely appreciate Kite, still use it actively, and currently have no intention of switching. I agree Kite API has improved.
However, when I recently tested Upstox’s Plus API against Kite, I noticed Upstox’s API was faster—both in terms of order placement speed and in returning more timely price data (sometimes ahead of Kite by 500 milliseconds, and many times even a second or more).
You used to be ahead by a significant margin, but that’s not the case anymore. The competition has caught up and is becoming increasingly tough. I kindly urge the Kite team to significantly improve the API to outperform the competition once again. @nithin
Can you offer capabilities for analysing Family portfolio as well. At the moment it is limited by a single login. Given you already have this capability in Console, it may be a case of exposing this via an API for Claude to access.
