A text lands at 8:40 on a Tuesday night. You can tell it's a question, because there's a question mark sitting at the end, but the rest is in a script you don't read. Might be Bengali. Might be asking your hours, your prices, whether you're open Sunday.
So you do the dance everyone does. Copy the message, paste it into a translator, read the English, type your reply, paste that back in, copy the result, send it. By the time you've gone around twice, the customer has already texted the next shop on the block, and that shop answered in ten seconds.
Now picture a normal week here, where your walk-ins speak Spanish, Mandarin, Russian, and Bengali in the same afternoon. The language gap isn't some rare event on a NYC block. It's the front desk on an ordinary day, and every message you can't answer fast is a booking that quietly walks.
What a text assistant actually does with a message you can't read
A text assistant is software that reads your incoming texts and drafts the replies. Point it at your business line, and when a message shows up in a language you don't speak, it handles three steps in about a second: it reads the message, shows you the English, and sends your answer back in the customer's language once you okay it.
You stay in English the entire time. The customer stays in theirs. There's no live translator on the phone, and no waiting until morning for the one bilingual person on your crew to come in.
For the everyday questions, that's the whole thing. "Are you open Sunday?" turns into a quick "Yes, 10 to 6, walk-ins welcome," and it lands in Bengali the way you'd have said it.
Where it's genuinely good
Translation software earns its keep on short, factual, low-stakes messages. That happens to describe most of what hits your front desk.
- Booking and rescheduling. A date, a time, a service. "Can you do Thursday at 3?" is easy to carry across any language.
- Hours and holidays. Open Sunday, closed Labor Day, late on Fridays.
- Address and directions. Which train, which side of the street, where to park.
- Standard prices. What a men's cut runs, what a basic cleaning costs, your deposit for a big job.
- Walk-in and wait questions. "Do you take walk-ins?" "What's the wait right now?"
Picture that Bengali text from 8:40. The assistant shows you "Are you open tomorrow, and how much for a beard trim?" You tap back "Open 10 to 7, beard trim is $15," and it goes out in Bengali before the customer has set the phone down. That's a booking you'd have lost to the copy-paste dance.
These work because the reply only has to carry facts, not feeling. A time and a price mean the same thing in every language. The machine doesn't need to be a poet, it just needs to get "3 o'clock" right.
Where it gets shaky, and you want to know before it bites
Here's the honest part. Machine translation is strong on plain sentences and genuinely weak on nuance, and that gap matters most exactly where the stakes are highest.
- Legal and medical detail. Contract terms, liability, an allergy, a medication, aftercare instructions. A small slip here isn't a typo, it's a problem. Get a human.
- Tone and formality. Several languages have formal and casual ways to say "you," and the software picks one without really knowing your relationship. It can land warmer or colder than you meant.
- Slang, idioms, and sarcasm. "Hit me up" and "it cost an arm and a leg" come out literal and strange. Keep your own replies plain and they translate clean.
- Dialects. The Spanish on your block isn't the Spanish in a textbook. Usually fine for booking, occasionally off on a single word.
None of this makes the tool useless. It makes it a sharp receptionist for scheduling and FAQs, and the wrong tool for a fight about money or a medical question. Know which kind of message you're looking at, and hand the hard ones to a person.
Spot-check three things before anything goes out
Set the assistant to show you the draft before it sends, at least while you're learning to trust it. The review takes a few seconds, and it catches the stuff translation engines fumble: the pieces of a message that aren't really words.
- Names. A customer's name and your own shop name can get "translated" or garbled into something odd. They should pass through untouched.
- Addresses. Street names and cross streets sometimes get mangled. If you're sending directions, read them in English and make sure they still point to your door.
- Prices and times. Confirm the number and the time say what you meant. "$40" and "4 o'clock" should survive the round trip exactly.
Once you've watched it handle a few dozen of these and the drafts come back clean, you can let the simple confirmations send on their own and keep the review step for anything carrying a dollar figure or a date.
Worth doing this week
You don't have to rewire the whole front desk to start. A few small moves:
- Think back over the last month and name the two or three languages your texts actually come in. That's what you're solving for, not every language on earth.
- Turn on a text assistant for your business line with translation on and the "show me the draft first" setting on.
- Write three canned answers in plain English: your hours, your address with the nearest train, and your standard prices. Plain sentences translate best.
- Run it in draft-only mode for a week and read every translated reply before it goes, watching the names, addresses, and prices.
- Flag the message types that should always go to a human: anything legal, medical, or about a refund.
Do that, and the next time a question comes in at 8:40 in a script you can't read, you answer it in a minute, in their language, and the booking is yours instead of the shop down the block's. If you'd rather not wire it up alone, that's the kind of thing we set up and tune for NYC shops.