Finance will always have a better model than a localisation lead with a spreadsheet. That is not a reason to have no model. Cost per accepted string is a blunt unit: supplier fee plus estimated internal review hours, divided by strings that passed the gate — not strings that left the translator.
Rewinds belong in the numerator. A cheap rate card that returns 12% of a job is not cheap. Machine translation that looks inexpensive until post-edit hours are counted is the same story in a different font.
We do not ask cohorts to build a full activity-based costing system. We ask them to stop being surprised. If Arabic costs three times German per accepted string, the weekly review should say so before the quarterly vendor meeting becomes a morality play.
The limitation is real: internal time is often untracked. Guess it, label it as a guess, and improve the guess later. A labelled guess is still Localization Performance Analytics. An invoice opened in panic is not.
Module seven of the Locale Scorecard Lab is where this column appears. People argue with it. That is the point of putting it on the same page as time-to-locale rather than in a finance folder nobody opens on Monday.