Can it handle Spanish?

Language support in a voice product is really three separate capabilities. Recognizing what was said, understanding what was meant, and saying something back naturally.

A vendor can be strong at one and weak at another. Recognition on a clear line is very good across major languages; recognition on a noisy mobile with a regional accent is a different test entirely.

Then there is switching. Plenty of real calls start in English and move into Spanish halfway, or mix both in one sentence.

Test with real callers, not with your own careful pronunciation. The accents and the background noise are the point.

Test the hard parts specifically: a street name, a surname that has to be spelled, a phone number read back. Those are where recognition errors cost you an appointment.

Check what the transcript looks like. A call that went fine but produced an unreadable transcript is only half working, because you still need to act on it.

Ask what happens when the language is not supported. A clean handover to a person is a perfectly good answer; pretending to understand is not.

Say you run a roofing company where about one caller in six prefers Spanish, and Marisol in your office is fluent. Before trusting any vendor's language claim, have her place three test calls from her cell while standing outside near traffic.

On the first, she speaks only Spanish from the opening word and asks for a repair estimate. For the second, she starts in English and switches to Spanish halfway through giving her address, which is 1420 Calle Ocho. The third is in English with her natural accent, and she spells her surname, Echeverría, letter by letter.

Then read the three transcripts together. Did the first call reach a person or get muddled through? Is the address from the second call usable by someone driving to it? How did the surname come out? Marisol will spot errors in seconds that you'd never see, and her verdict is worth more than a feature list.

Have whoever answers your phone keep a tally for a week: total calls, and calls where Spanish was needed. The ratio decides your approach.

At two in a hundred, a transfer rule to a bilingual colleague during office hours is a fair answer, and your main exposure is after hours. Work out what happens to a Spanish-speaking caller at 8pm. Caller ID comes through with the call, so at the very least Marisol can have a list of numbers to ring back at nine the next morning. GreetKeeper answers in English and routes by what the caller needs, so during office hours a caller who asks for Spanish can go straight to Marisol.

At thirty in a hundred, Spanish calls deserve a direct route. Write the transfer rule so a caller who asks for Spanish reaches Marisol at once, and nominate a second bilingual number for when she's busy. Test that path on a busy morning before you rely on it.

Watch for a quieter failure: a call that seemed to go fine but left a garbled transcript. If your team can't act on the record, the call only half worked.

GreetKeeper answers in English, and routes by what the caller needs. A caller who asks for Spanish is a routing case like any other: name the person, and the transfer goes to them.

Nominate a second number for that rule. If your bilingual colleague is on another call, the transfer tries the backup before the assistant takes a message.

Transcripts record what the assistant heard, which is how you find out whether recognition is really working on your callers rather than in a demo.