An offline Python CLI and editable experiment workbook by Ahmadreza Samadi for measuring full workflow cost per accepted result. The CLI has no dependencies; the workbook can be used without Python.

The three workbook sheets — Compare, Results and Attempts — pair manual and AI outcomes for the same batch. Record every retry and failed attempt, labor effort, elapsed time and non-labor cost. Capacity: 10 paired results and 30 attempts. Replace all fictional rows before using it for your own experiment.

The worked example is fictional: manual review costs 24 for three accepted results (8 each); the AI workflow costs 18.40 for two accepted results (9.20 each). Lower total spend alone does not establish a better workflow. The gallery shows these fictional workbook inputs and outputs.

Acceptance is caller-defined, not independent verification of correctness. Unknown costs must not become zero; zero accepted results leave unit cost undefined. This project does not measure ROI, establish a company outcome or recommend a model.

MIT licensed and developed with AI assistance and synthetic tests. Excel desktop and Google Sheets native recalculation have not been tested.

Download the v0.1.1 workbook: https://github.com/ahmadreza-smdi/ai-workflow-cost/releases/download/v0.1.1/AI-Workflow-Experiment.xlsx

Source, CLI instructions and release: https://github.com/ahmadreza-smdi/ai-workflow-cost

Method: https://a-samadi.com/writing/ai-cost-per-verified-result.html

Maintainer: https://a-samadi.com/about.html