Playbooks

Your bank balance is a lagging indicator in a current indicator's costume. The five-step close a founder can run in 90 minutes: the statement read, the receivables sweep, the subscription delta, the three numbers that predict trouble, and the one-page note that compounds.

The invoice is real, the logo is right, and only the bank account is a lie. Why spotting fakes is the wrong plan, the thirty-second callback rule that beats even the perfect email, and the two supports that make it survive a busy office.

An ops hire runs $60–80k a year plus the management overhead nobody budgets. Before you write the JD, sort the work that’s drowning you: judgment earns the hire, repetition earns a system — and doing that sort in the right order often changes the answer.

Sprawl accretes one $29 Tuesday fix at a time — every tool had a buyer, but the portfolio has no owner. The one-hour audit: name a human for every charge, map the overlapping jobs, cancel the zombies, and put every renewal on one calendar so it stops regrowing.

Admin time never arrives as blocks — it arrives as crumbs, too small to count and big enough to eat a working day per week. The one-week tally method that makes it visible, and the four-bucket call (automate, design out, delegate, keep) that follows.

You already know the back office is chaos — HR, support, training, knowledge all improvised. "Fix it" never happens because you're treating it as one big project. It isn't. Pick the one function that's bleeding the most time, systematize it, write down what you learn, roll to the next. Ninety days, one function at a time, no new hires.

Add up your subscriptions and you'll get the wrong number. The real bill is the integration tax — per-tool admin, duplicate data entry, context-switching, onboarding overhead — that no invoice shows and that quietly becomes the owner's second job. How to total it in an afternoon, and when consolidation actually beats best-of-breed.

Between five and twenty employees, a small company acquires the org chart of a big one without the people to fill it. You need four specialist functions and can't afford four hires. The way out isn't a generalist — it's seeing that each function is mostly repeatable machinery with a thin layer of human judgment on top, and carrying the machinery for you.

A field report: a non-engineer deployed servers, fixed a payment bug, and wired analytics in one evening with AI. The honest framework for when a generalist plus structured AI can do specialist work — three conditions — and the line where you still need the human. Bias on the table, written by someone who builds exactly this.

Between employee five and fifty, you acquire four departments and hire none of them. They all run the same loop — intake, triage, respond, document, improve — and it breaks at the same step every time. The operating pattern, the stack principles, and where each function's deep-dive lives.

The default answer is buy — and "we could build that ourselves" is not a reason to override it. The maintenance iceberg that sinks most build decisions, the four questions that actually settle it, and why AI changed this math less than the hype claims. From a solo founder who built four products and still tells you to buy.

Every SaaS company added "AI-powered" to their landing page in the last 18 months. Most of them are lying. The five-question framework that exposes whether the AI is structural or decorative — in five minutes of a demo, before you commit a quarter of budget.