the product only works because i do two hours of setup per client — how do you get that to zero

Do not aim at zero yet. Aim at knowing what the two hours are made of, because right now you are describing it as one lump and it is not.

Time yourself across the next five clients with a stopwatch and a list of tasks. Every time I have done this the answer was six or seven distinct tasks where two of them ate most of the minutes. In my case it was the import fixing and one specific configuration question, and the rest was small.

Then automate only the two. Not the flow, not the whole onboarding, the two. My equivalent went from two hours to about thirty five minutes with one importer improvement and one better default, and I had been planning a full self serve rebuild that would have taken a season.

The remaining thirty five minutes is a call, and the call is where I learn what to build next, so I am in no hurry to remove it.

Put a number on it before you decide anything, because right now it might be worse than you think.

Two hours per customer at any honest hourly value of your time is a real cost against a monthly subscription. Divide that cost by your monthly price and you get the number of months before that customer has paid for their own onboarding. Then compare it to how long your customers actually stay. If median lifetime is shorter than that payback, every new customer makes you poorer, and growth is the opposite of the cure.

At nine customers you cannot know your median lifetime yet, which is itself an argument for keeping the setup manual and the price high enough that it does not matter.

Screen record yourself doing it once. You will delete half the steps before you automate anything, purely from the embarrassment of watching.

The thing nobody says out loud: at nine customers the two hours is the research phase, not the problem.

If you automate the current process you are casting your least informed understanding of the workflow into code, and you will spend the next year fighting your own abstractions. I have done exactly this and had to delete a whole onboarding engine because it encoded assumptions from my first four customers, three of whom churned.

Keep doing it by hand. Write down what happens every time. Automate at the point where you are bored, because boredom means the shape has stabilised.

Intermediate step that helped me more than it should have: a template repository and a checklist document per client, with the exact steps in order and the commands to paste. It is not automation, it is just not thinking. Went from about ninety minutes to fifty, made the two most annoying steps obvious because they were the ones I dreaded on the list, and gave me something to hand to someone else eventually. Took an afternoon.