Great question, and definitely one of the operational complexities of the business. Our thesis is: by relying heavily on software from day one, we can actually

Great question, and definitely one of the operational complexities of the business. Our thesis is: by relying heavily on software from day one, we can actually do it with a small, super-highly-skilled team (rather than a giant army of folks.) If the task can really be done by an arbitrary person somewhere without context on the business itself, I suspect we can execute that same task ourselves in software (only with much higher consistency and much lower cost). Transaction categorization is a great example of this, actually. I think the alternative—scaling by just hiring a ton of people—is compelling because it's short-term easy but long-term hard. By analogy to a software system: if you scale with an army of people, you're building a giant distributed system of super-unreliable hardware (the error rate of people is pretty high!), and you can't reprogram it by running "git push". I'll venture to make a stronger statement: if you care at all about consistency of quality, I think you have to do this mostly in software -- because otherwise the quality of your service is going to be entirely the quality of the individual bookkeeper you use, which is fundamentally going to be pretty variable.