Does it work where your meetings actually happen?
This is the question that decides everything else. A bot that joins video calls is excellent for a remote team and useless in a car, a corridor or a site visit. A worn device is the opposite.
Be honest about the split. If ninety percent of what matters happens on scheduled video calls, buy a meeting bot and ignore the rest of this list.
What does the price actually cover?
The headline is usually the hardware. The recurring cost is transcription, and it is frequently metered in minutes. Work out your real monthly usage — three hours of meetings a day is roughly sixty hours a month — and price that, not the device.
A free tier of 300 minutes a month is five hours. That is two days of a normal week.
Does it handle your languages, in the way you use them?
"Supports Arabic" and "handles a sentence that starts in Arabic and ends in English" are different claims, and only one of them matters in a Gulf meeting. Test it with a real recording of a real meeting before you commit.
What happens to your recordings?
Ask three specific things: where the data is stored, whether it is used to train models, and what deleting means. Vague answers to any of them are an answer.
For an always-worn device this is not a compliance checkbox. It is the whole question, because the thing being stored is every conversation you have.
How long does the battery last in your day, not in the specification?
Continuous-recording figures are measured in ideal conditions. What matters is whether it survives from your first meeting to your last with the syncing you actually do.
Standby matters more than people expect. A device that is flat when you need it has no features at all.
Can you get a person on the other end?
Hardware fails, pairing breaks, a transcript comes back wrong. What you want to know is whether someone answers, how fast, and in what timezone.
This is the least glamorous question on the list and the one you will care about most in month two.