Your automation saved money. Audit that.
Enter the savings figure you'd put on a slide — hours saved times rate. Then step through the audit: review time, error rework and tooling spend come off the headline one deduction at a time, and the two numbers rarely match.
The deduction ledger shows exactly which silent cost — review time, error rework, or tool spend — eats the claimed win. Nearly every inflated ROI number is hiding the same three culprits.
Why do claimed and measured automation savings differ?
The headline usually counts every hour the old process took. The audit subtracts what nobody put on the slide: the review time humans still spend, the rework when the automation gets it wrong, and the tooling bill itself.
What is error-rate accounting?
Multiplying how often the automation gets a task wrong by the time a human needs to notice and fix it. A modest error rate with slow rework can quietly eat a third of a claimed win — it scales with volume, just like the savings do.
What belongs in true cost of ownership?
Everything the automation costs to keep running: subscriptions and API usage, the human review loop, error rework, and maintenance when the process it automates changes. Savings minus all of that is the number worth reporting.
This playable is part of the registry: its tool page
Next: will that model fit on your GPU?
Can You Actually Run This? shows the VRAM bill line by line — weights, KV cache and overhead against real GPU memory sizes.