Word counts survive because they are easy to extract and easy to put on a slide. They also survive because they feel fair: more words, more work. Fairness is not readiness.
A 4,000-word legal string can block a store listing while 40,000 words of help centre copy sit harmlessly unpublished. If your scorecard weights them equally, you will celebrate the help centre and miss the listing. Localization Performance Analytics starts by naming the unit you actually manage — string, job, screenshot set, or binary asset — and refusing to let “words” pretend to be that unit.
Volume still belongs on the page, as remaining inventory. Pair it with time-to-locale and with a quality sample that is not volunteered by the supplier. Then write a gate sentence a release manager can read without a glossary of TMS statuses.
In GB product organisations we still see weekly packs titled “98% complete” where complete means a job left the translator queue. Review, screenshots, and legal have not started. That is not dishonesty so much as a metric that was never asked a better question.
The Locale Scorecard Lab spends a full module deleting vanity columns. People are often relieved. The number they feared losing was not helping them argue.