WhatsApp Chatbot for Singapore Businesses (2026): How to Build One, What It Costs, and When You Need the API

First, the disclosure, then the one confusion that costs people money

We build WhatsApp chatbots. Specifically, the third path in this guide: custom AI agents on Meta's WhatsApp Business Platform for Singapore businesses with ten to two hundred staff. Read on knowing where our commercial interest sits. Most businesses that email us should start on one of the first two paths.

Now the confusion. "WhatsApp Business" is two products that share a name.

• The WhatsApp Business app is the free green app on a phone. One number, one phone, a handful of linked devices, and a few automated messages. No API, no real chatbot for WhatsApp, no integration beyond a catalogue and labels.
• The WhatsApp Business Platform (usually called "the API" or "the Cloud API") is what every actual chatbot runs on. Meta hosts it, you or a provider connect software to it, and you pay per message delivered. This is where AI agents, broadcast campaigns and CRM syncs live.

Every "how much does a WhatsApp chatbot cost?" thread goes in circles because half the people mean the app and half the API. The answer is three numbers.

How much does a WhatsApp chatbot cost in Singapore?

WhatsApp chatbot pricing splits three ways. Platform fees below come from the builders' public pricing pages. We checked Wati and respond.io in October 2026 and both fit the bands, but tiers move often, so check the current page before you budget. The build band is from our AI automation cost guide, not our pricing page; more on that under path 3.

Two things the table hides. Meta's per-message pricing is a moving target, charged by category (marketing, utility and authentication templates, with separate rules for the free-form service replies inside the 24-hour window) and by the customer's country, and the rules on which replies are free have changed more than once, most recently in October 2026; the Singapore rate card and worked monthly totals are in our separate WhatsApp Business API pricing guide. And the platform fee is the small line; over three years, the number that matters is your team's time when the bot cannot answer.

Path 1: the free WhatsApp Business app, and exactly where it stops

If you are a sole proprietor, a hawker stall taking pre-orders, or a two-person tuition centre, start here and pay nobody. Install the WhatsApp Business app, register a company number (a prepaid SIM or a landline via voice verification), and switch on three things:

1. Greeting message, sent on first contact or after fourteen days of silence. Set expectations: "We reply within an hour, 9am to 7pm. Reply 1 for prices, 2 for opening hours."
2. Away message, sent outside the hours you set. Say when you will reply.
3. Quick replies: canned answers behind a slash command, so "/price" pastes your price list. For you, not the customer.

The hard limits

• Nothing reads what the customer typed and decides what to do. "Reply 1 for prices" still needs a human to see the "1" and paste the list.
• One number, one primary phone, a few linked devices. When the phone goes to a site visit, the shop's WhatsApp goes with it.
• Broadcast lists only reach people who saved your number and cap out in the low hundreds. This is not a marketing channel.
• No integration. Nothing lands in a Sheet, a CRM or a calendar unless a person types it.

Is there a free WhatsApp chatbot? Only in this sense: free automated messages, not a free bot. Anyone offering more for free is running a builder's free tier (see path 2) or an unofficial WhatsApp Web automation, which gets numbers banned. We have seen Singapore businesses lose numbers with years of customer history that way.

Path 2: a no-code chatbot for WhatsApp on a BSP builder

The right path for most Singapore SMEs with a support or sales inbox shared by two to five people. A Business Solution Provider (BSP) is a company Meta has approved to onboard businesses onto the API; most sell a flow builder on top. Drag boxes, write each button's text, connect a form, and the bot runs on your official number.

The names you will meet in Singapore are SleekFlow, Wati, respond.io, Omnichat, Twilio if you have developers, and 360dialog for cheap raw API access when you will build the bot elsewhere. Several appear in our guide to AI chatbot companies in Singapore, with the caveat that we are on that list too.

What it costs

Entry tiers on the mainstream builders cluster around USD 30–100 a month (Wati's and respond.io's public pages sat at USD 29 and USD 79 when we checked in October 2026), mid tiers at USD 100–300, "enterprise" above that; some price in SGD and some bill per seat, so the same team can land anywhere from roughly SGD 50 to 500 a month. Meta's per-message charges are passed through on top, sometimes with a markup; if the pricing page does not say so, ask. Free tiers are usually too limited to run a business on.

Setup, in order

1. Create a Meta Business Portfolio (the old "Business Manager") if you do not have one from running ads.
2. Sign up with the BSP, connect through their embedded signup, and register a number. If that number is on the WhatsApp Business app today, you will be asked to delete the app account first, so back up the chats. Some providers now offer Meta's "coexistence" mode that keeps app and API on one number; ask before you plan around it.
3. Submit a display name that matches your registered business name or a brand you can prove you own.
4. Build the flow: a welcome message with buttons (Book, Prices, Talk to a human), one branch per button, a lead form that writes to a Sheet or the built-in CRM, and a catch-all that hands off to the inbox.
5. Set the handoff rule. The bot should give up early and loudly, not loop.

Where it stops

Flow builders answer the question you anticipated, not "can I change my 3pm to 5pm and add my son". Most now sell an "AI reply" add-on pointed at your FAQ page; in our experience it is fine for simple questions and unpredictable once the answer needs a lookup. And you do not own it: switch providers and you rebuild the flows and re-apply for templates.

Path 3: a custom WhatsApp AI chatbot on the Cloud API

This is what we build, so discount accordingly. A custom WhatsApp AI chatbot is software you own between Meta's Cloud API and your systems: a language model, a retrieval layer over your documents and records, a few tools (check a booking, look up an order, create a lead), a human handoff, and an evaluation set that tells you when it gets worse. The WhatsApp part is easy; the grounding and the tools are the work.

What it costs

Our single-channel chatbot band is roughly SGD 8k–35k as a one-off build, the band from our AI automation cost guide. Our pricing page quotes SGD 20k–80k, but that is for a full automation system; a scoped, single-channel chatbot sits below it. The 35k ceiling is for one channel; the multi-channel builds in our chatbot companies guide run to SGD 60k. The bottom is a bot grounded on a clean FAQ with one integration, say a Calendly or a Google Sheet. The top is a bot that speaks to a clinic management system, a payment gateway and a CRM, needs PDPA-grade logging, and must clear an accuracy bar before it goes near patients. Run costs after launch are the model's token bill, a small server and Meta's per-message charges; at typical SME volume we see low hundreds of SGD a month, an observation rather than a promise. A retainer to keep extending the bot is optional and priced separately.

When path 3 is the wrong answer

Under a few hundred conversations a month, the maths rarely works; a builder and a part-time human are cheaper. If no document or database holds the right answers, the bot will improvise. And if nobody on your team will own it after launch, read why most B2B AI agents fail in production before you sign anything, including with us.

One Meta rule that affects AI bots

Meta's Business API terms now prohibit general-purpose AI assistants (a ChatGPT clone on WhatsApp). A bot that serves your own business, answering about your products, bookings and orders, is what the platform is for. Keep the scope narrow.

How to create a WhatsApp chatbot for business: the six steps that trip people up

1. The number

Use a number the company controls, not a staff member's personal SIM. A number lives on the app or the API, not both, unless your provider offers coexistence.

2. Meta business verification

You can send and receive before verification, but with a cap on business-initiated conversations: 250 unique customers in a rolling 24 hours at the time of writing. Verification needs documents matching your business name, typically your ACRA BizFile profile, and a company email on your own domain. An ACRA name that does not match the Meta profile name is the most common hold-up we see; fix the profile first.

3. Message templates

Any message your business starts (a reminder, a status update, a promotion) must use a template Meta approved in advance, categorised as marketing, utility or authentication; the category sets the price. Submit templates a week before you need them. Common rejection reasons:

• A placeholder at the very start or end, or placeholders with no example values.
• Promotional language in a utility template ("Your order has shipped. Use code SAVE10 next time" is marketing; Meta will recategorise or reject it).
• Shortened URLs, or links to a domain that is not yours.
• Vague content that could say anything ("Hi {{1}}, {{2}}").
• Typos, or mixed languages in one template. Submit one per language if you serve English, Chinese and Malay speakers.

4. The 24-hour customer service window

When a customer messages you, a 24-hour window opens in which you or the bot can reply with free-form text. Outside it you can only send approved templates. This breaks most first bots: a customer asks on Friday night, the human answers Monday morning, and the message cannot be sent. Resolve or hand off inside the window; anything later goes out as a utility template that invites a reply.

5. Handoff to a human

Every bot needs an exit. Decide the triggers (customer asks for a person, confidence drops, a complaint, a refund), where the chat lands (shared inbox, Slack, CRM task), and what the customer is told. Test it with an angry message.

6. Connect it to something

A bot that writes nowhere is a message you still retype. Every lead and booking should land in a Sheet or your CRM with the number, timestamp and question.

WhatsApp chatbot examples by Singapore industry

These are the patterns we see work, described generically. None is a named client; each is a starting scope.

Clinics and dental practices: appointment reminders and rescheduling

A utility template goes out 48 hours before the appointment with two buttons, Confirm and Reschedule. Confirm writes back to the clinic system; Reschedule offers the next three open slots, books one, and confirms. Handoff for anything clinical. The win is the no-show rate, measurable before and after. The bot never states a result or diagnosis unless the practice has deliberately decided it will.

F&B: pre-orders, catering enquiries and the Friday rush

Buttons for Today's specials, Pre-order and Catering. Pre-order collects items, pickup time and name, totals the order, sends a PayNow QR, and drops it into a Sheet the kitchen watches on a tablet. Catering collects headcount, date and budget and hands off to the owner. Path 2 handles this; the moment you want the bot to answer "anything without peanuts for a school of 200?", you are in path 3.

Property agents and developers: viewing bookings and qualification

A click-to-WhatsApp ad lands the lead in chat. The bot asks the three qualifying questions the agent would ask anyway (buying or renting, budget, timeline), offers viewing slots from the agent's calendar, books one, and pushes the lead to the CRM tagged by source.

Logistics and e-commerce: order status without the "where is my parcel" queue

The bot takes an order number or the customer's registered number, looks up the status from the courier or OMS, and answers. Late beyond a threshold, it hands off with the tracking attached. Outside the 24-hour window, status changes go out as utility templates. This flow often clears the largest share of a support inbox, so it is usually the first thing we scope.

The four failure modes, and how to design around them

These decide whether the bot works, not the provider.

One more, a habit rather than a rule: letting the bot pretend to be a person. Say it is a bot in the first message. Customers do not mind an honest assistant; they mind being fooled.

WhatsApp chatbot integration: CRM, Google Sheet, and what to leave out

"Can it connect to X?" Usually yes. The better question is whether it should.

Start with a Sheet

For the first month, a Google Sheet is the right CRM. Every lead, booking or order becomes a row: number, name, what they asked, what the bot did, whether a human took over. Two hundred rows teach you which fields your real CRM needs.

Then the CRM

Most builders have native connectors for HubSpot, Zoho and Pipedrive; custom bots use the CRM's API directly. Our rule: the bot creates and updates the contact and logs the conversation, but does not move deal stages on its own. Choosing the CRM at the same time? Our separate WhatsApp CRM comparison covers that; the CRM should follow your sales process, not the bot vendor's integration list.

Glue tools for everything else

The rest of the WhatsApp automation, meaning calendar bookings, payment confirmations, courier lookups and Slack alerts, usually runs through an automation layer rather than hand-written code. The trade-offs are in n8n vs Zapier vs Make; for WhatsApp bots, self-hosted n8n is our default when data residency matters, Make when the team will maintain it without us.

What to leave out of version one

Payments inside the chat (use a PayNow QR or a payment link), anything that changes a financial or medical record, and marketing broadcasts. Broadcasts are the fastest way to get a new number throttled if your opt-in records are thin. Get the service bot right, earn a clean messaging tier, then add campaigns; if campaigns are the main goal, that is a marketing automation project with WhatsApp as one channel.

Which path is right for you: a two-minute decision

Stop at the first yes.

1. Under fifty conversations a month, one person answering? Path 1.
2. A shared inbox, repeatable questions, leads to capture, and a team that will maintain button flows? Path 2. Take two free trials, build the same five-button flow in each, and keep the one your team finds easier to edit.
3. Questions that need a lookup, data you cannot hand to a third party's inbox, or a volume where every point of deflection is a salary? Path 3. Scope one measurable flow, budget in the SGD 8k–35k band, and own the code, the prompts and the eval set.

Most Singapore SMEs land on path 2 and are well served there. The only bad outcomes are paying path 3 money for a path 2 problem, or running an unofficial hack on a number your business depends on.