Ines Montani is the co-founder of Explosion AI, a digital studio specialising in tools for AI technology. Ines speaks from experience from spaCy and prodigy, tools for NLP and data annotation.
The "startup playbook" isn't the only way. You can be profitable early. You can keep the team small. Yu don't have to do anything sneaky, you can just make something good.
You don't need to run at a loss! Reasons to run at a loss may be network effects to get started, a large scale of operations, predatory pricing, or enterprise sales, all of which cost you money in one way or another.
But bigger isn't necessarily better. Software is more expensive to build at scale, and most businesses aren't "winner takes it all" (and those markets suck anyways).
Be aware of survivership bias! We don't really know what caused other companies to win out, so imitating them isn't possible in a meaningful way.
The good news is: there are lots of opportunities! People are drawn to "winner takes all" markets. This leaves many other high value opportunities untouched. Optimize for median (not mean!) outcome.
Good teams can be surprisingly small. You don't need to pass the bus test – excellence requires authorship not redundancy or design by committee. Building the right stuff matters much more than building lots of stuff. At a smaller scale, generalists make much more sense than specialists in general, but the best model is if everybody has some specialisation, but also a common skill set. This complementary team setup is hard to build in larger teams, and much easier in smaller teams. Think of it as tree-shaped skills (some trunk skills, lots of branches, and growth) instead of T-shaped skills.
In a lot of traditional environments, this is selected against, but if you leave these environments and perceptions, you can cover everything you need with few people (hence not running at a loss).
You can make good decisions without testing all of your assumptions. Don't fall prey to inverse survivorship bias either – just because you didn't do something and failed, it doesn't mean doing that thing would have saved you. Have a look at autopsy.io, collecting autopsies of failed startups.
You can't replace logic with data. Decisive data is the exception, not the rule. Decisions are mostly based on reason, and you'll win if you're mostly right. Build things you think are good.
The true value does not lie in your users' data. Data is not the new oil, wth. Monetize the money, not the users. Ship value, charge money (novel concept). Users appreciate cosftware that works. Users are not interchangeable test subjects, they're people and they remember things. Profit is the best KPI.