October 24, 2026
Cutting the Cost of a Venture Experiment

Our goal for the next two months is to run 100 venture experiments and to know, before we start, what each one will cost. With a hundred experiments, one real business is far more likely to come through than with ten. That only works if each experiment is cheap, and if the cost never surprises us.
This week was about cost: where the money goes, how to see it, and how to stop it when it should stop.
1. Low-Cost Models by Default
Most of the work in an experiment happens when nobody is watching: research runs, quality checks, judges and audits. In mid-October those background calls alone used up our entire balance with one AI provider. We moved every one of them to a lower-cost model by default:
Low Effort by Default — New ventures start on the lowest-cost level. Higher levels are a choice, not the default.
One Model per Venture — A venture keeps the model it started with from plan to launch, so its cost is predictable.
No Silent Upgrades — A failed call on the low-cost model no longer falls back to the expensive one on its own.
2. Credit You Can See
Every screen that starts work now shows a small credit light for each model provider: green above five dollars, yellow below, red at zero. When a provider is red, the buttons that would start work on it are disabled, so a run can no longer die halfway through for lack of credit.
3. Paying Only for What Runs
The Stop all switch pauses every venture, its ads and its email sequences in one click. Ventures that do not need ads right now do not run them. The expensive product builder is held back for ventures that have proven demand.
4. Pages for Cents
Studio, our own page builder, replaced per-page product builds with pages that cost cents. That changes the cost of an experiment more than any other single step this month.
Weekly Learnings
Background work is where money leaks — The calls nobody sees cost the most. Every one of them now has an owner and a default.
Cost has to be visible — A balance you can see on the screen prevents more failures than any alert email.
Cheap only matters with good ideas — Lower costs let us run more experiments, but the ideas still decide how many of them work.
Plan for Next Week
One Switch for Every Model Call — Find the last places that call the expensive model directly and route them through one setting.
A Price Before Every Experiment — Show what an experiment will cost before it starts, ads included.
Better Video Ads — Raise the quality of video ads, so a small ad budget gives a reliable answer.
Time to First Revenue — Once the hundred are running, shorten the road from the first email signups to the first payment.