How Julius AI reached 2 million users by staying focused

Executive overview

General-purpose AI tools like ChatGPT handle everything but do nothing deeply. Julius AI bet on the opposite: become the best AI data analyst, nothing else.

That focus drove 2 million users in 18 months and 10 million data visualisations. When ChatGPT's plugin store shut down overnight and erased Julius's main acquisition channel, the team pivoted to product-led, word-of-mouth growth by building sharing into the core experience.

Focused tools beat general-purpose ones — and sharing is the compounding loop that replaces paid acquisition.

The case for radical focus

• ChatGPT handles video, images, search, and writing — data analysis is an afterthought
• Users hit ChatGPT's limits on real data (shallow insights, bad charts, no collaboration) and Google "AI data analysis"
• Julius is the top search result for that query
• One specialist outperforms one generalist with decent range — the same holds for AI agents
• "Google/OpenAI just killed your startup" is overblown; if you solve the problem better, it doesn't matter

Lessons from earlier failures

• Waterview (a managed backend service for hackathons) solved a real pain point — but only once per weekend project; users never retained
• Core rule: if users only have the problem once a month or less, they won't retain to the product
• At Uber and Facebook, one "no" anywhere in the chain kills an idea; at a startup, one "yes" is enough

Launch fast, learn fast

• Julius ships features when they're "barely working" to get early feedback before over-investing
• NBA/Hoops GPT was built and launched in two weeks before the season ended
• Result: sports fans aren't data-savvy; the only engaged users were bettors — not the audience Julius wanted
• Learning what doesn't work is as valuable as learning what does

Rebuilding growth after the plugin store closed

• Julius grew early through the ChatGPT plugin store; OpenAI shut it down overnight
• Forced the team to find organic, sustainable acquisition
• Insight: when users get a data insight, they want to share it — they don't keep it to themselves
• Built sharing directly into the product; analysis outputs became the acquisition loop
• Word-of-mouth replaced the lost channel; solve the problem well and users become champions