Got First 500 Users for your AI product as an Indie-（To be continued)

Today I checked our user data.
select count(*) from users

returned a single number:541.

You might not think that's much, but for my first indie project, it's a number I'm proud of. And since launching on October 10, 2024, this number has been growing steadily and organically — we haven't done much active promotion.

After reading this post, you may see:

• Ideation: How we turned a friend's casual request into an AI stock analysis product.

• 0 to 10 users: how we validate ideas with minimal cost, and even without put our product online

• 10 to 100 users: How to ride market momentum and media buzz for organic growth

• 100 to 500 users: how good user experience drive our user growth.

• monetization: Key points and how to start

It's a bit too long and I may need to split into 2 posts.

1. How It Started

Last July, an open-source project called Trading Agents suddenly blew up on GitHub. Its social media performance was equally dramatic — typical of their style. While I'm not a fan of this kind of hype-driven marketing, I have to admit it does make technical concepts accessible to the masses.

A friend of mine, a fund manager, asked if I could get this open-source project running and he wanted to utilize it for stock analysis .

You might think: just fork it, deploy it, run it — should be easy, right?

Yeah, I thought so too at first.

But I discovered the hardest part was data acquisition — you need real-time stock quotes, company financials, news data, and more. I spent a couple days figuring it all out and finally generated the reports.

After that, my friend asked me every few days to run analysis on different stocks.

That's when I realized there was a real gap: the technology to analyze stocks with AI exists, but regular investors can't access it. If I could package this capability into a product where people just open a webpage and use it, there should be demand.

In fact, I'm increasingly realizing:
technology only creates commercial value when it's properly packaged into good products.

The things tech nerds often disdain are exactly where the commercial value lies. I didn't realize this early on — I'm still in the process of this mental shift.

2. Decide to try

I discussed it with my partner. (Yes, we're a two-person indie dev team. My partner focuses more on technical architecture and back-end development, I focus more on product design and front-end development, but we don't have strict role divisions. How to find a co-founder is a topic for another time.)

After a short talk, we decided to start with the simplest version. We run some popular stocks daily and post them to stock forums to gauge reactions. Our reports back then were pure text as below:

We posted for over ten days straight. No traction. We started thinking about what was wrong.

3. The Pivot

Then we decided to change the format — converting numbers into visual dashboards.

Just that one change made a huge difference. People started paying attention, discussions picked up, and people began asking: "Bro, where did you get this chart?"

Yes, just that one change. For tech developers, it's trivially easy to implement, but from a user perspective, it created a completely different reaction.

This validated what I said earlier: "The things tech nerds often disdain are exactly where the commercial value lies."

So we moved into formal development.

4. Build V0.1

We decided V 0.1 would have only the core function: enter a stock ticker, wait a few minutes, read the report. No complex parameter settings, no fancy UI, not even a user system.

With AI coding tools, we estimated: one week to launch. (There are plenty of articles about using AI coding tools, so I won't elaborate here.)

However, even this simple version took us a month.

Why? Because while the features were simple, we spent considerable effort on data scraping and prompt engineering. Simply put: context engineering — sounds simple, actually quite tedious. We had three core tasks:

1. Stable data acquisition. Initially we used APIs to fetch data in real-time, but discovered it severely impacted performance. We switched to storing data, making performance more controllable. You can use web scraping or purchase data feeds.
2. Prompt optimization. We found issues in the original open-source project's prompts and made targeted improvements.
3. Landing page polish. The homepage is your first impression — you have 3 seconds to get users to click.

Time allocation for these three tasks was roughly: 60:20:30

On October 9, we finally launched our product online. Just posted it on our social media, telling friends: "Built something, give it a try."

5. First 10 Users

First week, 10+ people registered. Mostly friends or friends of friends.

I obsessed over the backend data — how long they stayed, how many stocks they looked up, which pages they lingered on.

The data looked good. Average 10 minutes stay per person, 7 stocks(at that time one page was for one stock) queried. This meant they weren't just clicking in and bouncing — they were actually using it, exploring.

6. First 100 Users

Second week after launch, a hot event happened on social media: the Alpha-arena ran an experiment where they gave several LLMs real $10,000 to trade Bitcoin and showed live results.

This experiment made "AI investing" blow up. Many people started believing, or at least getting curious: can AI really help with investment decisions?

We seized this moment.

We rode this wave and published a deep-dive article titled:

When AI Can Trade Crypto, How Should We Use AI for Stocks?

This wasn't just clout-chasing. We spent two full days writing this piece, took several stocks(APPL, CEG, KO), and backtested AlphaWiseWin's recommendations — meaning, what would our returns have been if we'd followed AlphaWiseWin's advice?

While we admittedly didn't have much data at the time, we objectively used about ten days' worth and found our returns still beat the market.

This article did get some attention. By early November, we crossed 100 registered users. You may see the 2 spikes are all due to our post on Alpha-arena's event.

Feedback and Iteration

With 100 users, we started getting feedback. Good and bad, via email.

Good feedback:

• Reports are genuinely useful, comprehensive angles, more reliable than randomly reading news
• Someone said they only realized how many risk factors their stock had after seeing our analysis

Bad feedback:

• Too slow — reports take several minutes to generate
• Sometimes the AI analysis is too academic and hard to understand
• Want to see more data, like financial metrics, valuation comparisons, etc.

We decided to tackle "too slow" first, compressing generation time from an average 10 minutes to around 4 minutes. Methods included:

• Workflow optimization, changing serial nodes to parallel
• Testing different LLM API response speeds
• Leveraging caching

These methods aren't hard, just require repeated refinement.

Note: up to this point, we didn't even have a user system, let alone commercial charging. That was our next consideration.

We entered November, we think there are more to improve in visualization, that's how we decide the coming version should be the real Version 1.0

7. Version 1.0

Version 1.0, we decided, should have one core principle: let users understand a stock at a glance.

We noticed many products give users tons of options, requiring them to ask questions:

• Include technical analysis?
• Deep dive into financials?
• Focus on short-term or long-term?

Sounds flexible, but it's actually creating friction. Because:

• Most users don't know how to comprehensively analyze a stock
• Multiple interactions are needed to build complete understanding

So our approach: let users get a complete report with one click. Two keywords:

• One click
• Comprehensive

This change had immediate impact. New user engagement clearly improved, returning user frequency increased. By mid December, we approached registered users with stable 20% weekly retention.

Till now, we think we have gone through the so-called MVP stage, we started to consider monetization. I will continue in next post.

What I learnt:

looking back, the most important learning is:

MAKE THE TECHNICALS EASY TO USE IS MORE IMPORTANT THAN I THOUGHT.

in next post， I will post more on monetization, although we are still on the way to monetization, but still there is something to share.