Here is a message a wet-market fishball supplier actually got at 9.47pm: "Boss tmr got fresh one anot ah, need 5kg for my stall." And here is one a physio clinic got the same week, from an older customer: a WhatsApp voice note, entirely in Mandarin, asking whether her Tuesday appointment could move because her daughter cannot fetch her. Neither of these is a neat, form-filled enquiry. Both are completely normal in Singapore.
We tell ourselves "everyone here speaks English," and on paper that's mostly true. But the language people *buy* in, complain in, and change their minds in is not the language on the government form. It's Singlish, it's Mandarin or Hokkien with an auntie, it's Malay or Tamil for a whole slice of your customers, and very often it's three of those in one sentence. If your WhatsApp replies only work in tidy textbook English, you're quietly losing the messages that don't look tidy, which is most of them.
Singapore doesn't speak one language. It speaks four, plus Singlish
Officially we have four languages, and the split is real, not ceremonial. A meaningful share of your customers are more comfortable reading and replying in Mandarin than in English, especially older folks and heartland businesses. Malay and Tamil aren't rounding errors either, they're daily languages for hundreds of thousands of people who will happily spend money with you if you meet them where they are.
Then there's the thing no textbook covers: Singlish and code-switching. People don't politely finish one language before starting the next. They ride both at once, like:
- "eh bro still got slot ah, tonight can anot"
- "你们 open 到几点? need to collect one thing"
- "boss the price got GST or already included ah, don't play play"
- "can lah but my friend also want, confirm plus chop or not?"
A person reads all of that in half a second. The question is whether whatever answers your WhatsApp does too.
How Singaporeans actually type (and it's not in full sentences)
If you've ever watched your own WhatsApp Business inbox, you already know. Nobody writes "Good evening, I would like to enquire about your operating hours." They write "still open?" at 10pm. They shorten everything: "tmr," "anot," "liao," "pls," "ur." They drop punctuation and subjects entirely. They send a voice note because typing in Chinese is a pain on a phone. They send a photo of the thing they want with the caption "this one how much." They fire off three separate bubbles instead of one message, then add "sry" and a fourth.
This is the real texture of enquiries here, and it's exactly where old-school automation falls over. A customer who gets a robotic "Sorry, I didn't understand that" after typing "got slot anot" doesn't rephrase politely. They just close the chat and message your competitor down the road who happened to reply like a human.
Why a keyword bot dies at Singlish (and a modern AI assistant doesn't)
The older WhatsApp "chatbots" most SMEs tried were basically keyword matchers. If the message contains "price," send the price list. If it contains "open," send the hours. That works right up until someone writes "wah your thing ex sia" (that's a price question with zero price keywords) or "open till late anot" or a Mandarin voice note. The bot sees no matching keyword and either stays silent or dumps the wrong canned reply. Worse, it can't handle a sentence that switches language halfway, because it was never built to.
A modern language-model assistant works differently. It reads for *meaning*, not for trigger words. It can tell that "ex sia" is a complaint about price, that "anot" is a yes/no question, and that a message opening in Mandarin and ending in English is still one coherent request. It handles voice notes by transcribing them first. It doesn't need you to predict every phrasing in advance, which is good, because in Singapore you can't. This is the whole reason we built microcrew to reply naturally in the customer's own language, understand Singlish and code-switching, and do it within seconds, at 2am included.
Sounding like the same shop in every language
Here's the trap people fall into once they solve the language problem: the replies start sounding like four different businesses. Warm and chatty in English, stiff and formal in Mandarin, oddly stilted in Malay because it's a bad machine translation. Customers feel that. It reads as "we outsourced this," which is the opposite of what a small SG business wants to project.
The goal isn't translation, it's *the same you* in another language. If your English replies are friendly and end with a cheerful "see you then!", your Mandarin replies should carry the same warmth, not switch to bureaucratic officialese. That's a matter of setting the voice once and having it hold across every language, rather than bolting Google Translate onto a formal script. A good assistant lets you define the tone, the do's and don'ts, the way you'd greet a regular, and keeps that consistent whether the customer opened in Tamil or in Singlish.
Know when to stop replying and pass it to a human
Handling language well does not mean automating everything. Some messages should always reach a person, and knowing where that line sits is what separates a helpful setup from an annoying one. Sensible handoff points:
- An upset customer ("very disappointed, third time already") — never let automation try to smooth over real anger.
- Anything involving money disputes, refunds, or a special discount the owner has to approve.
- A medical, legal, or safety question where a wrong answer causes harm — a clinic should loop in staff, not guess.
- When the customer explicitly asks to speak to a human — respect it immediately, don't loop them.
- Genuinely novel requests the assistant isn't confident about — better to say "let me check with my colleague and get back to you" than to bluff.
The right model is: the assistant covers the 80% of repetitive, after-hours, "still got slot ah" traffic instantly, and quietly hands you the 20% that actually needs a human touch. If you run a clinic, our clinics playbook goes deeper on where that line should sit; property agents will find the same logic in the property-agent guide.
Getting names and details right — a quiet PDPA point
One underrated benefit of handling language properly: you capture details correctly. A customer who says "my name is Nurul, N-U-R-U-L" or "chinese name Tan Wei Ming" should end up in your records spelled right, not mangled. Same for NRIC-style references, delivery addresses in Malay, or a Tamil name a rushed staffer might typo at 6pm.
Under Singapore's PDPA, you're responsible for the personal data you collect, so accuracy and careful handling aren't just nice-to-haves. Capturing a customer's name and contact correctly in their own language, and not scattering it across screenshots and personal phones, is both better service and better compliance. Treat the language layer as part of your data hygiene, not separate from it.
You don't need to hire a four-language front desk to serve Singapore properly. You need replies that meet customers in the language they actually message in, Singlish and all, sound like *your* shop every time, and know when to tap you on the shoulder. If you'd like to see what that looks like on your own WhatsApp, microcrew does exactly this, and new users get 50% off the first month. Message us on WhatsApp at +65 8466 3236 and try it on a real enquiry, no hard sell. Or read more playbooks first, that's fine too.