Unit economics before scale: the model that kills bad ideas early

Published 31 March 2026 · LoxiLabs

Cost per outcome

Take the decision the system makes and price it: inference or API cost, licence share, data pipeline cost, and the people needed to run and monitor it, divided by the volume production will see. Then compare it to the value per decision from the Prove sprint.

What the model usually reveals

Three patterns. Use cases that are attractive at pilot volume and marginal at production volume, usually because inference cost scales linearly with volume and value does not. Use cases where the people cost dominates and automation of the review step matters more than model quality. And use cases where the economics are so favourable that the only question is how fast to scale.

Where the model lives

In the Find phase as a rough estimate to rank candidates. In Build as a working spreadsheet the owner can change. In Scale as a monitored metric alongside the outcome, because costs drift as models and volumes change.

Why it belongs before the demo

A prototype flatters a use case because it runs at demo volume on curated data. The economics are what separates a demo from a business, and they are cheap to model early.