The most expensive hour in any rollout is the one spent arguing about whether it worked. Someone puts up a chart of output. Someone else says the work does not feel any different. Both are right, because output was never the thing being bought.
Volume is the easiest number to move. It is also the weakest evidence you have. A system that produces four times as much undifferentiated work has multiplied a liability, not created leverage. The chart goes up. The brand stays exactly where it was.
We publish our own operating numbers because they are checkable. Thirteen agents. More than 44,000 runs in 63 days. Under $50 in model spend. A 12-operator roster behind it. Those are inputs and costs. They prove the machine runs, and that it runs cheaply. They do not prove it deserves a line in next year’s budget. That is a different question, and it needs different measures.
Three measures that survive a board meeting.
Time to first usable draft. Not time to a draft. Time to the version a senior person will attach their name to without rewriting it. This is the number that decides whether a system is help or homework.
First-pass approval rate, tracked as a trend. Every decision at the human approval gate is a recorded judgment about your standards. Flat means the system has stopped learning. Rising means your taste is compounding into the machine.
The cost of the second brief. The first piece of work in a new territory is always expensive, because someone has to decide what the brand thinks. What matters commercially is the marginal cost of the ninth piece in that same territory. That curve is the entire argument for building a system instead of hiring more hands. The mechanics sit on /system.
The approval gate is the instrument.
Most teams treat the human gate as friction to be minimized. It is closer to the opposite. The gate is where the system finds out what good means inside your company, and it is the only place that judgment gets captured in a form the system can reuse. Audit-grade logs turn each approval and rejection into evidence you can replay six months later, when someone asks why the brand said what it said.
That is also the honest answer to the governance question. Read-only connectors and a human gate are not a compliance concession. They are what makes the throughput defensible once it arrives. Our /posture page sets out where those boundaries sit.
What to hold the system to at 90 days.
We tell clients to expect three to five times the throughput inside 90 days. On its own that number means very little, and we say so. Paired with the three measures above it becomes a claim you can test. More work, approved faster, at a falling marginal cost, on a system you can walk away from, because the routing is model-agnostic and nothing is locked in.
If volume rises and approval rates fall, you have bought a factory for work nobody wanted. If volume rises and approval rates rise with it, you have bought operating leverage, and it will show up as market clarity long before it shows up in headcount. The sequence is mapped on /journey.
Choose your three measures before the first agent runs. Baseline them the week before. Then the quarterly review is a reading, not a debate.
If you want help choosing the measures that will hold up in front of your board, book the strategy blueprint call.
