AI quality scoring today runs every call through one English scorecard. When a call is in Spanish, the system translates it to English and scores the translation. Literal translation removes the courtesy, tone, and phrasing that our criteria measure. Formal address (usted), culturally standard greetings, and idioms don’t map word-for-word to our English script standards. As a result, Spanish calls score lower even when the agent performed to standard.

This happens on both of our lines:

Spanish line: Calls are scored against criteria written for English conversations.
English line: When a Spanish-speaking caller reaches the English queue and a bilingual agent helps them, that call is scored on the English scorecard and penalized for serving the caller in their language.

Business impact

Inaccurate QA data: Those scores are unreliable, which skews team and individual performance reporting.

Unfair agent evaluations: Bilingual agents, who are the hardest to hire and keep, get lower scores for doing the right thing. This hurts coaching, morale, and retention.

Manual workaround cost: Our leads each spend about 5 hours per week re-reviewing Spanish calls by hand, which cancels out the efficiency AI scoring is supposed to give us.

Client experience risk: Spanish-speaking clients are a core population for our firm. We can’t measure or improve their experience accurately if the scoring doesn’t reflect what happened on the call.

Requested capability

Detect the spoken language of each call automatically, whatever line or queue it came in on.
Route each call to a scorecard assigned to that language. For example, Spanish calls go to a Spanish scorecard and English calls go to the English one.

Evaluate Spanish calls in Spanish, against Spanish-language criteria, without translating to English first.

Success criteria
Any call spoken mainly in Spanish is scored on the Spanish scorecard with no manual reassignment, and its scores line up with human reviewer scores at the same rate English calls do.