How to Automate Customer Enquiries: A Singapore SME Guide

To automate customer enquiries well, you automate the boring half of every conversation: acknowledging the message, asking the questions you always ask, sorting the enquiry to the right person and logging where it came from. You do not automate the part where someone decides on a price, a date or a promise.

For most Singapore service businesses, enquiries arrive in five places at once: WhatsApp, Instagram DMs, email, the website form and the phone. The real problem is rarely "we need a chatbot". It is that nobody knows which enquiries are waiting, who is handling them, and which ones quietly died.

Below is how to fix that, in the order we would do it.

Where enquiries actually leak

Before buying anything, spend one week counting enquiries, first-reply times and outcomes per channel. The common leaks:

• ▸The personal phone. The boss's own WhatsApp is still the main line. Enquiries pile up during site visits, and nobody else can see them.
• ▸Instagram DMs nobody owns. They sit in the requests folder for days because the person who "does social" is part-time.
• ▸The web form email. It lands in a shared inbox, gets read, and everyone assumes someone else replied.
• ▸Missed calls. No callback list, so a missed call after 6pm is simply lost.
• ▸The half-answered chat. The customer asked about price, got "depends ah, can send photo?", sent the photo, then heard nothing.

Automation fixes the last four far better than it fixes the first. If one person's phone is the bottleneck, you need a shared inbox before you need a bot.

Step one: decide what a "complete" enquiry looks like

Every automated flow depends on knowing which details you need before a human can act. Too few, and staff spend three messages chasing basics. Too many, and customers abandon the form.

A good rule: ask for what you need to route and roughly price the job, nothing more. Here is what that looks like across common Singapore service businesses.

Two notes. First, a "budget band" (for example, a few preset ranges) gets far more honest answers than an open "what's your budget?". Second, treat anything health-related as sensitive: collect it at consultation, not in a WhatsApp flow.

Step two: sort and route with rules first

Most sorting does not need AI. Simple rules handle the bulk of it and are easy to check when something goes wrong.

Rules that work well:

• ▸By entry point. An enquiry from the "commercial cleaning" landing page goes to the commercial team. A click-to-WhatsApp link from a specific ad carries a pre-filled first line, so you know which campaign it came from.
• ▸By form answers. Landed property renovations go to the senior designer; BTO enquiries go to the package team.
• ▸By time. After-hours enquiries get an honest auto-reply ("we reply from 9am") and land at the top of the morning queue.
• ▸By keyword, carefully. "Leaking", "urgent" or "no cold air" flag an enquiry as priority. Keep the list short and review it monthly, because keyword rules misfire on sentences like "not urgent".

Routing should end in a named owner, not a shared pile. An enquiry assigned to "the team" is an enquiry assigned to nobody.

Step three: use AI where free text gets messy

WhatsApp and Instagram customers do not fill in forms. They send one long message, often in Singlish, sometimes with a voice note or photo: "Hi can come Tampines 4rm this Sat? 2 aircon not cold, one dripping water."

This is where AI earns its place. Used well, it can:

• ▸Classify the enquiry into your service types (aircon repair, not servicing).
• ▸Extract details into your fields: area Tampines, flat type 4-room HDB, 2 units, preferred date Saturday, symptom "not cold, leaking".
• ▸Spot what is missing and draft the follow-up question ("Wall-mounted or ceiling units?").
• ▸Summarise long threads so the person picking it up does not scroll through 40 messages.

The trade-offs to accept honestly:

• ▸It will sometimes be wrong, especially on ambiguous messages and mixed languages. Plan for a person to glance at its sorting at first.
• ▸It should only fill fields, not invent them. If the area is not mentioned, the field stays empty, and the follow-up asks for it.
• ▸It must not answer questions about price, availability or policy unless it is looking those answers up from your own, current information.

That last point is the difference between a useful AI and a liability. We cover it more in our 12 grounded AI automation examples for Singapore SMEs.

WhatsApp: know which one you are using

"WhatsApp automation" means very different things depending on which version you are on. At a buyer level:

The features of the Business app, including automated greeting and away messages, quick replies and labels, are listed on WhatsApp's own small-business page. For many businesses under a few dozen enquiries a day, setting these up properly is the right first step and costs nothing.

The 24-hour customer service window

On the Business Platform, Meta's documentation is clear: when a customer messages or calls you, a 24-hour customer service window opens, and it resets each time they message again. Inside the window you can reply freely. Once it closes, you can only send pre-approved template messages.

What that means in practice:

• ▸If a customer asks for a quote on Monday and you reply on Wednesday, you cannot just send a normal message. You need an approved template, such as a quote-ready notice.
• ▸Under Meta's current pricing, charges apply per delivered template message (since 1 July 2025), while non-template replies inside an open window are free. Fast replies are not just better service; they are cheaper.
• ▸Plan your templates for follow-ups, appointment reminders and "we're ready with your quote" before you go live.

One more policy point. Meta now restricts general-purpose AI assistants on the Business Platform, as noted on its AI providers page. A business using AI to handle its own customer enquiries is a different use case, but it is another reason to keep your AI focused on your services, not chatting about anything.

Speed to first reply

Customers in Singapore message three or four providers at once for the same aircon job or cleaning slot. The first useful reply often wins, not the cheapest.

"Useful" is the key word. An auto-reply that says "thanks, we'll get back to you" buys you a little patience. One that says "thanks, to quote you we need: number of units and whether they're wall-mounted or ceiling" moves the job forward while you are on site.

A practical target set per channel:

• ▸Acknowledgement: instant, automated, every channel, including after hours.
• ▸First human reply in office hours: a target your team can actually hit, measured weekly.
• ▸After hours: say honestly when you will reply. Do not promise "we'll call you back shortly" at 11pm.

Missed calls deserve their own flow: log them, and send a WhatsApp or SMS within minutes saying you missed their call and asking what they need.

Handing off to a human

A bad handoff undoes all the good automation. The customer answers three bot questions, then a staff member asks the same three questions again.

A good handoff has four parts:

1. ▸Clear triggers. The customer asks for a person, gets upset, mentions a complaint, or the enquiry is high-value (a full condo renovation, a 300-pax event).
2. ▸Full context. The person receives the summary, the extracted fields and the source, not just a notification.
3. ▸An honest message to the customer. "Passing you to Wei Ling, our designer. She'll reply by 2pm." Name and time, not "an agent will assist you".
4. ▸The bot steps back. Once a person takes over, automation stops replying in that thread.

Track source and outcome, or you are guessing

Every enquiry should end with two things recorded: where it came from and what happened. Without both, you cannot tell whether your Google Ads, Instagram posts or Carousell listing actually produce jobs.

Keep the outcome list short: booked, quoted, not a fit, no response, lost to competitor, plus a simple "why lost". After a month, you will know which channel sends enquiries that convert. Our lead generation service is built around this loop: we only call a campaign good when it produces enquiries that turn into bookings.

What not to automate

• ▸Quotes and prices, unless they are pulled from a fixed, current price list, and even then a person should confirm anything non-standard.
• ▸Complaints and refunds. Route them to a senior person immediately.
• ▸Anything medical, safety-related or legal.
• ▸Promises about dates and availability unless the system reads your real calendar.
• ▸Chasing customers who went quiet with repeated automated nudges. One polite follow-up is service; four is spam, and on WhatsApp it can get your number reported.

PDPA: what to get right

Automated or not, the PDPA applies to the names, numbers and details you collect. The PDPC's Advisory Guidelines on Key Concepts set out the obligations. The ones that matter most for enquiry handling:

• ▸Tell people why you collect their details and use them only for that. A short line in the auto-reply and form, linking to your privacy notice, covers most of it.
• ▸You stay responsible for your vendors. If an inbox tool or AI provider processes enquiries for you, the guidelines say you have the same obligations as if you did it yourself. Check where data is stored and put it in the contract.
• ▸Do not keep enquiries forever. Stop retaining personal data once it no longer serves its purpose or a legal or business need. Set a clean-up schedule for old, unconverted enquiries.
• ▸Do not collect NRIC numbers at enquiry stage. The PDPC's NRIC guidelines restrict collection to when it is required by law or needed to verify identity to a high degree of accuracy.
• ▸Be open about AI. The PDPC's guidelines on personal data in AI systems encourage being transparent about how AI uses personal data. Saying "our assistant sorts your enquiry for our team" is honest and costs nothing.
• ▸Marketing is separate. Replying to an enquiry is not permission to add the customer to promotional broadcasts. Get that consent separately.

A start-with-one-channel plan

Do not automate five channels at once. Pick the one with the most enquiries (for most Singapore service SMEs, WhatsApp) and do it properly.

Weeks 1–2: measure and define.

• ▸Count enquiries, first-reply times and outcomes for your busiest channel.
• ▸Write down your must-have fields and your service categories.
• ▸Decide who owns which type of enquiry.

Weeks 3–4: the basics.

• ▸Move off personal phones to a business number everyone can see.
• ▸Set up instant acknowledgement with your intake questions.
• ▸Add rules-based routing and an after-hours message.
• ▸Start logging source and outcome for every enquiry.

Weeks 5–8: add intelligence carefully.

• ▸Test AI extraction on 50 past enquiries before using it live.
• ▸Run it in "suggest" mode: AI fills fields, a person confirms.
• ▸Define handoff triggers and write your follow-up templates.

Month 3: review, then expand.

• ▸Compare first-reply time and booking rate with week one.
• ▸Only then bring in the next channel, usually the web form or Instagram.

What "grounded" looks like in practice

Our clearest example is our own product, Onz, a live events site for Singapore. Its "My Day" planner turns a one-line request into an itinerary of real events, meals and travel times, using live NEA weather and OneMap travel times. When halal mode is on, it only suggests MUIS-certified options: 739 halal restaurants plus certified stalls in 29 hawker centres.

Onz is not an enquiry inbox. But the principle carries straight over: the AI only works from real, current data, and it is restricted to what that data says. An enquiry assistant that quotes from your actual price list and books from your actual calendar is useful. One that improvises is a risk.

Where to start

If enquiries are slipping between WhatsApp, Instagram and your inbox, the fix is usually smaller than you think: clear intake, simple routing, and one channel done properly before the next.

Our AI agents and automation page explains how we approach this, including where we think AI is not worth it. If you would like a second opinion, send us a WhatsApp, tell us how your enquiries come in today, and we will say honestly what is worth automating and what is not.