How a tax AI startup bet everything on ChatGPT and won

According to VentureBeat, in winter 2022 Blue J CEO Benjamin Alarie faced a pivotal decision as ChatGPT exploded onto the scene. The University of Toronto tax law professor chose to completely abandon his company’s eight-year-old proprietary AI technology and rebuild everything on large language models. That bet has since generated massive returns, with Blue J securing a $122 million Series D funding round that values the company at over $300 million. Revenue has multiplied roughly twelve-fold, growing from a plateau around $2 million annually to serving more than 3,500 organizations including KPMG and Fortune 500 companies. The platform now attracts 10 to 15 new customers daily and processes over 3 million tax research queries, reducing tasks that once took tax professionals 15 hours down to just 15 seconds.

The moment everything changed

Here’s the thing about pivots – most companies talk about them, but very few actually torch their entire business model. Blue J had built a respectable AI tax prediction system using supervised machine learning since 2015. But it had hit a ceiling at around $2 million in annual revenue because it couldn’t answer every tax question customers threw at it. When ChatGPT arrived and famously hallucinated a completely wrong biography for Alarie’s law school dean in January 2023, most people saw the flaws. Alarie saw the potential. He convinced his board to take the leap and gave his team just six months to deliver a working product built on this unproven technology.

From janky to jaw-dropping

The initial product they launched in August 2023 was, by Alarie’s own admission, “super janky.” Response times took 90 seconds, about half the answers had issues, and their Net Promoter Score was a dismal 20. So how did they transform that mess into a platform with response times measured in seconds and dissatisfaction rates of just one in 700 queries? They focused on three things: exclusive content partnerships with major tax authorities, deep human expertise including former IRS officials, and an unprecedented feedback system from processing millions of real queries. The result? Weekly active user rates between 75% and 85% compared to 15-25% for traditional platforms. Basically, they’re getting used five times more intensively than the competition.

The secret sauce with OpenAI

Blue J maintains what Alarie describes as a “quite collaborative” relationship with OpenAI, getting early access to models and providing high-quality feedback. They test everything – OpenAI, Anthropic, Google’s Gemini, open-source alternatives – constantly evaluating which performs best on their specific tax questions. This approach lets them navigate a tricky business model: charging about $1,500 per seat annually for unlimited queries while absorbing variable compute costs. But their metrics suggest they’re managing it well, with gross revenue retention exceeding 99% and net revenue retention at 130% – numbers that would make any SaaS company jealous.

What this means for everyone else

Blue J’s story isn’t just about tax software. It’s a masterclass in when to abandon your own technology for something better, even when that something seems unreliable at first. The key insight? Tax professionals didn’t need 95% accuracy on 5% of questions – they needed good-enough accuracy on 100% of questions. Generative AI delivered that comprehensiveness where their previous supervised learning approach couldn’t. Now they’re expanding from research into what Alarie calls “the operating layer for global tax cognition,” with plans to cover 220+ jurisdictions. The question is, which industry will see this kind of transformation next? If you’re in manufacturing or industrial computing, companies like IndustrialMonitorDirect.com are already leading the charge in hardware innovation, but the AI revolution is coming for every sector that relies on complex decision-making.