WhatsApp chatbot cost in India: the short answer
A WhatsApp chatbot is three separate bills, not one, and most quotes you receive show you only the first. The software that runs it costs anywhere from nothing to about Rs. 2,499 a month for a small business, Meta charges your business separately for the messages themselves, and the flows have to be designed and connected once, which typically runs Rs. 15,000 to Rs. 2,50,000 depending on whether it only answers questions or reads live data from your own systems.
Comparing this with a bot that sits on your website? Those costs work differently — we break them down in AI chatbot cost in India. This page is only about WhatsApp.
Mixing them up is how businesses end up angry six weeks in. They compare two subscriptions, pick the cheaper one, then receive a Meta invoice nobody budgeted for. Or they budget for messages and forget that somebody has to decide what the bot actually says.
| Cost layer | Who bills you | How often | Typical India range in 2026 |
|---|---|---|---|
| Platform subscription | Your chatbot software vendor | Monthly or yearly | Free to Rs. 2,499 a month on ZupiChat |
| WhatsApp message charges | Meta, to your own WhatsApp Business Account | Monthly, after the fact | Set by Meta, by message category and country |
| Build and flow design | Your agency or your own team | One-time, then on every change | Rs. 15,000 to Rs. 2,50,000 |
Only the first is a price you can shop around on. The second follows from your own habits. The third decides whether the bot removes work from your team or quietly adds to it.
Bill one: the platform subscription
This is the software layer: shared team inbox, flow builder, contacts and tags, template management, broadcast scheduling, reports and the API. It does not include the messages. Think of it as the dashboard, not the postage.
On ZupiChat, our own WhatsApp CRM, Starter is free forever, Growth is Rs. 999 a month with three WhatsApp numbers and three team members, and Enterprise is Rs. 2,499 a month with unlimited agents. There is no free trial because the free plan is the trial. You run real conversations on it and upgrade when you outgrow the seats. API access is on every plan, including the free one, which matters the day you want the bot to read from your own software.
Four questions to ask any platform before you pay
- What happens to the price when you add a second WhatsApp number or a fourth agent? Per-seat and per-number limits are where a cheap plan turns expensive.
- Is API access locked to the top tier? If it is, every future integration becomes a forced upgrade.
- Does the vendor resell Meta’s messages with a markup, or does Meta bill you directly? ZupiChat adds no markup: message charges are set by Meta and billed by Meta to your own WhatsApp Business Account.
- Can you export your contacts and chat history if you leave? If not, that is not a subscription, it is a lock-in.
Ask one more thing: which API the platform actually runs on. ZupiChat runs on the official WhatsApp Business Platform, the Cloud API. It is not a Meta Business Solution Provider and carries no Meta-granted badge, and much of the market is in the same position whatever the homepage implies. Avoid the cheap unofficial route, where automation is bolted on to the consumer app: it puts your number at risk of being blocked, and the number is the asset. For a full sales CRM rather than a messaging tool, see our guide to CRM software cost in India for small business.
Bill two: Meta’s per-message charges
Meta bills your business directly for WhatsApp messaging, against the payment method attached to your own WhatsApp Business Account. Your chatbot vendor is not in the middle of that transaction. That is why a Rs. 999 subscription can sit next to a message bill several times its size in a campaign-heavy month.
Rates are set by Meta, they differ by message category and by the country of the number you are messaging, and Meta revises them. Do not take a per-message figure from a sales deck as fixed, and do not accept a blended “cost per message” that bundles a vendor’s fee with Meta’s rate. Check the current India numbers on Meta’s own WhatsApp pricing documentation before you build a budget on them.
The four categories, and why your mix matters more than your volume
| Category | What it covers | Who opens it | How it behaves on the bill |
|---|---|---|---|
| Marketing | Offers, launches, re-engagement, abandoned cart nudges | Your business | The dearest category in most markets |
| Utility | Order, dispatch and delivery updates, receipts, transaction reminders | Your business | Priced below marketing |
| Authentication | One-time passwords and login codes | Your business | Priced on its own scale, usually high in volume |
| Service | Your replies inside the customer service window | The customer | Not charged under Meta’s current model |
The pattern to internalise is simple: messages you start cost money, marketing costs the most, and conversations the customer starts are the cheap ones. When a customer messages you, a customer service window opens and you can reply freely inside it for a limited time. Under Meta’s current model those replies are not charged, though Meta revises this, so verify rather than assume. Let the window lapse and your next reply has to be a paid template.
What actually moves the monthly number
- Category mix. Ten thousand marketing messages and ten thousand order updates are the same effort and very different invoices.
- Who starts the conversation. A click-to-WhatsApp ad, a QR code on the invoice or a wa.me link makes the customer message first. That is cleaner consent and a cheaper conversation.
- List hygiene. Dead and wrong numbers cost the same as live ones, and they quietly damage your quality rating as well.
- Duplicates. When the website and the order system both send “your order has shipped”, you pay twice and look careless.
- Response speed. Every reply that misses the service window turns a free answer into a paid template.
Bill three: the one-time build and flow design
This is the layer that never appears on a pricing page, and it usually decides whether the bot deflects work or creates it. Someone has to define what the bot asks, in what order, what it does with each answer, what it says when it does not understand, and when it stops guessing and fetches a human.
A simple FAQ bot
Greeting, a menu of six to eight buttons, prepared answers for prices, timings, location and catalogue, and a permanent exit to a human. No integration, no live data. Typically Rs. 15,000 to Rs. 40,000 and about a week, most of it spent writing copy rather than configuring software. A capable in-house person can build this themselves; the cost is their time, not a licence.
An order-status bot wired into your system
Now the bot has to identify the customer from a phone number or order ID, query a live record in your order system or ERP, phrase every possible state including the awkward ones, and cope when that system is down. Typically Rs. 60,000 to Rs. 2,50,000, over three to six weeks. If the source system has no API, you are buying an integration layer first, and that is a custom software project with its own scope. Our breakdown of custom software development cost in India covers that layer.
What inflates a build quote
- No API on the source system, or an “API” that is really a nightly CSV export.
- Multiple languages, because each language version of each template is submitted and approved separately.
- Payments or document collection inside the chat.
- Handover rules that depend on business hours, agent skills or response timers.
- A bot expected to understand free text instead of offering buttons. That is an AI automation scope, not a menu.
Opt-in rules, and why a bought list destroys your quality rating
You may only message people who have given your business permission, on a surface where it was obvious they were opting in to WhatsApp messages from you, by name. Consent collected for email, or for your sister brand, does not transfer.
Meta does not police this with an approval queue; it polices it with quality rating. Blocks and reports pull your number’s rating down, a poor rating cuts your daily messaging limit, and a number can end up restricted. You get a colour on a dashboard as your warning.
This is why a purchased database is the most expensive cheap thing in WhatsApp marketing. You pay Meta’s marketing rate on every message, you get almost no replies, and you spend the one asset that cannot be repurchased: the standing of your number. Recovery is slow.
Opt-ins that hold up
- A checkbox at checkout or on the enquiry form that names WhatsApp explicitly, with date, source and page stored against the contact.
- Click-to-WhatsApp ads and wa.me links, where the customer sends the first message. Paid campaigns pointed at WhatsApp build a list that is both consented and cheap to talk to.
- A QR code on the invoice, the packaging or the counter.
- An IVR or missed-call opt-in that you actually log.
Give people a way out as clearly as you gave them a way in. Honour a stop request instantly and automatically, and test that it works: an unhonoured opt-out generates reports, and reports are what cost you the number.
Template approval, and why campaigns get rejected
Anything you send outside the customer service window has to use a template Meta approved in advance. You submit fixed text with variable placeholders, Meta reviews it, often within minutes but sometimes far longer, and then approves, rejects or recategorises it. Write festive campaign templates the week before, not on the morning of the sale.
The usual reasons for rejection
- Category gaming. A discount dressed up as an order update gets recategorised as marketing, so you pay the marketing rate anyway and lose credibility with the reviewer.
- Variables doing all the work. “Hi {{1}}, {{2}}” tells a reviewer nothing about what will actually be sent. Templates whose placeholders could contain anything get refused.
- Placeholder placement. Variables at the very start or very end of a message, or two sitting next to each other, are a standard rejection.
- Duplicates. The same content submitted twice in the same language.
- Sloppiness. Broken links, spelling errors, sample values that do not match the placeholder.
- Restricted content. Meta’s commerce and business messaging policies rule out entire categories, and what is permitted varies by country.
Keep a small library of approved templates and re-use them with different variable values instead of writing a fresh one for every campaign. Re-used templates are an asset; a queue of one-off submissions is a delay you will feel on launch day.
What the bot should answer, and what it must hand to a human
The most expensive chatbot is the one trying to be clever. Bots pay for themselves on the handful of questions your team answers all day, every day, and lose money the moment they are asked to improvise.
Give the bot
- Hours, address, directions, price list, catalogue, payment options.
- Order and delivery status, appointment booking, rescheduling and cancellation.
- Payment link resend, invoice copy, warranty registration.
- Qualifying questions before a human call: requirement, budget band, city, timeline.
Do not give the bot
- Complaints and refunds. Escalate at the first sign of annoyance, not after it.
- Anything involving money it cannot verify: negotiated discounts, custom quotes, credit terms.
- Medical, legal or financial advice, whatever industry you are in.
- The edge cases in your own policy that your own staff argue about.
Handover rules that work
- A “talk to a person” option visible on every screen, not buried three levels down.
- Automatic escalation after two failed attempts to understand. Never loop the same apology three times.
- The full transcript handed to the agent, so the customer is not asked to repeat themselves.
- Outside business hours, say so and give a return time instead of pretending someone is coming.
A bot that says “let me get someone” in ten seconds beats a clever one that wastes two minutes first.
Ongoing cost: who maintains the flows
A chatbot is not a build-and-forget asset. Prices change, offers expire, templates get rejected on resubmission, policy shifts, and the questions customers ask drift with your product range.
Budget for someone to do this every month:
- Read the conversations the bot failed or escalated, and add those questions to the flow. This single habit is what makes a bot deflect more work each month, not less.
- Submit and maintain campaign templates.
- Update prices, offers and menu options inside the flows.
- Watch quality rating and messaging limits before either becomes a problem.
- Repair the integration when your order system changes shape.
Realistically that is a few hours a month. Agencies price flow and campaign upkeep for a small setup in the Rs. 3,000 to Rs. 15,000 a month band, on the same logic as website maintenance cost in India. The cheaper route is to own it internally: one person who knows the flow builder and reads the escalation log weekly. What never works is nobody owning it, which is how a bot ends up quoting last year’s prices.
A worked monthly example, with the bills kept apart
Take a retailer with one WhatsApp number and three agents: roughly 4,000 utility messages a month, two marketing broadcasts to an opted-in list of 5,000, and about 1,200 conversations a month that customers start themselves.
| Line item | Billed by | Month one | From month two |
|---|---|---|---|
| ZupiChat Growth, 3 numbers and 3 team members | ZupiChat | Rs. 999 | Rs. 999 |
| Build: FAQ flow plus order-status integration | Your agency, one-time | Rs. 85,000 | Nil |
| Flow and campaign upkeep | Agency retainer or internal time | Covered by the build | Rs. 3,000 to Rs. 15,000 |
| 10,000 marketing messages | Meta, to your own WABA | 10,000 x Meta’s current India marketing rate | Same |
| 4,000 utility messages | Meta, to your own WABA | 4,000 x Meta’s current India utility rate | Same |
| 1,200 customer-initiated conversations | Meta, to your own WABA | Not charged inside the service window | Same |
Notice the shape rather than the totals. The subscription is a rounding error next to the message bill in any month with broadcasts in it, and the build, spread across a year, is roughly Rs. 7,100 a month. The lever that controls your spend is not the platform you chose: it is your category mix, your list quality, and how many conversations the customer starts.
Run the same sheet for your own business: count outbound messages by category, multiply each count by Meta’s current published India rate, add the platform plan you genuinely need, and divide the build by twelve. If the first line dwarfs the rest, stop shopping for software and fix the list.
What to do before you sign anything
- Write down the five questions your team answers most often in a week. That list is version one of your bot.
- Pull last month’s real volumes, split into marketing, utility, authentication and customer-initiated.
- Price those against Meta’s current India rates, not a vendor’s blended figure.
- Run a free plan with real conversations for a fortnight before paying anyone for anything.
- Commission the integration only once the simple flow has proved it removes work.
If you want that arithmetic run against your own numbers, or an order-status integration scoped honestly, talk to the team at Codelith Lab. We are based in Pune and work with businesses across India, and we will tell you when a menu-driven flow on a free plan is all you need.
