Episode Details

Dennis Mortensen counted 1,019 meetings in his calendar from the past year. 672 had to be rescheduled. That pain led him to spend five years building an AI SaaS that most founders would never attempt. In this episode, Dennis reveals why he spent months trying to talk himself out of building x.ai, how he validated the idea by hiring a human assistant for 50 friends, and what building AI products really takes.

Dennis raised a $2M seed round with a radical pitch: the only outcome would be a thumbs up or thumbs down on whether the AI SaaS was technically feasible. No MVP. No customers. No revenue. Five years and $44M later, 70 annotators in Manila had labeled 33 million data elements through supervised learning before the AI startup reached its inflection point.

More than half of all x.ai signups came from people who received scheduling emails from the AI SaaS assistant - making the product itself the largest acquisition channel. But freemium failed because users could not self-onboard a machine learning product without hand-holding.

🔑 Key Lessons

• 🧠 Invalidate ideas before building AI SaaS: Dennis spent months trying to kill the x.ai idea. He invited friends to find fatal flaws, reasoning that if an idea survives active attempts to kill it, there is real value.
• 🎯 Use concierge MVPs before building AI products from scratch: Dennis hired a human assistant for 50 friends at under $10K to test whether scheduling delegation was genuinely valuable before any engineering investment.
• 🚀 Accept pure technical risk when building an AI SaaS: Dennis raised $2M with no MVP promise - just a feasibility test. He spent the first year defining scheduling intents, not building customer-facing software.
• 📉 Freemium fails if users cannot self-onboard your AI startup product: x.ai offered five free meetings but users needed too much hand-holding. The concept of emailing an AI agent was too new for self-serve.
• 🤝 Let your AI SaaS product sell itself through built-in virality: Over half of x.ai signups came from people who received scheduling emails from the assistant - every interaction is a demo for potential customers.

Chapters

• Introduction
• From Denmark to New York via Yahoo acquisition
• What x.ai does and the scheduling problem
• How counting 1,019 meetings sparked the AI SaaS idea
• Why Dennis tried to invalidate the idea
• The concierge MVP with a human assistant for 50 friends
• Technical risk vs market risk in AI startups
• Raising a $2M seed for a thumbs up or thumbs down
• How the seed money was spent on data labeling
• The annotation process and defining the scheduling universe
• Built-in virality and product-led acquisition
• Why freemium failed for building AI products
• The scale of engineering required - $44M and 120 people
• How supervised learning and data labeling work
• Lightning round
• Where to find Dennis Mortensen and x.ai

Resources

• Full show notes: https://saasclub.io/212
• Join 5,000+ SaaS founders: https://saasclub.io/email