Section 1
The constraint is cash, and every tool bills monthly
A capable stack assembles quickly and then quietly compounds. Four subscriptions, a usage-based bill that scales with volume, an add-on seat for a contractor, and an annual plan you took for the discount. Individually each is defensible. Together they are a salary you never approved. So bootstrapped automation has a rule the funded version does not need: every tool has to name the hours it returns or the revenue it protects, and get reviewed on that basis. If you cannot say what a subscription earned in the last quarter, that is the answer. The second-order effect matters too. Usage-based pricing means your best month is also your most expensive one. Model the bill at three times your current volume before you build a workflow you cannot afford to succeed at.
Section 2
The three jobs worth automating when you are the whole company
Follow-up. Deals are lost to silence more than to objections, and the follow-up is exactly what falls off when you are delivering. An automated sequence that drafts and schedules the next touch, with you approving the send, protects revenue you already earned the right to. Delivery admin. The scheduling, the recap notes, the status update, the file naming, the handover document. None of it is what the client buys, all of it takes evenings. Bookkeeping intake. Receipts, invoices, categorization. It is the classic January problem, and it compounds because a year of unsorted records costs more than an accountant charges to fix. What stays with you is the work customers actually pay for, plus anything where being wrong damages a relationship you cannot afford to lose.
Section 3
A build-or-buy rule for a company of one to five
Buy when the workflow is generic and the vendor's version is boring. Build when the workflow is your actual differentiation, or when the priced version scales with volume in a way that punishes growth. The model below makes that comparison concrete, including the maintenance cost people forget. On the team-side consequences of either choice, see [The Future of Work: Preparing Your Team for AI Automation](/blog/the-future-of-work-preparing-your-team-for-ai-automation).
Section 4
Building it in the evenings without breaking the day job
Two hours a week, one workflow at a time, and nothing that touches a live customer until it has produced correct output for a fortnight on real inputs. Start from your own logs rather than a blank page. Pull the last thirty follow-up emails you sent, the last twenty recap notes, the last month of receipts. Those are your examples, your test set, and your style guide. An automation built from your own past work needs less correction than one built from a prompt describing what you wish you did. Keep the whole thing recoverable. Credentials in one place. A written note on what each automation reads and writes. A manual fallback you have performed at least once. When you are the only operator, an outage on a Tuesday is a business problem, not a support ticket.
Section 5
The dependencies a bootstrapped company cannot afford
Two risks matter more here than anywhere else. The first is data you cannot get out. Before a workflow becomes load-bearing, export everything once and confirm the export is usable rather than a formality. The second is a price rise you have no negotiating position against, which arrives most often at the renewal after you have integrated deeply. There is also a quieter one. When you automate customer communication with your own name on it, you have delegated your reputation to a system that occasionally gets things wrong. Keep a review step on anything that goes out under your signature, be plain when AI is involved, and never let an automated decision touch money, contracts or a customer's record without you seeing it. The cost of being wrong is asymmetric for a small company. One bad automated message to a major client can outweigh a year of saved hours.
Section 6
Hours returned to revenue
That is the metric. Not tasks automated, not tools deployed. Hours that came back and then went into selling, delivering or improving the offer. If the hours came back and dispersed into a longer inbox habit, the automation broke even at best. Review quarterly with three lines: total stack cost, hours returned, and revenue per working hour. Cancel anything that cannot show up in that table. Bootstrapping is a discipline about what you decline to spend, and a tool stack is the easiest place to stop being disciplined without noticing.