What a WhatsApp AI bot really costs per month in 2026

Updated 2026-09-24 · 5 min read · By the Convo360 team

The three bills

Confusion about chatbot pricing usually comes from platforms that bundle everything into one per-message number. Unbundled, there are exactly three costs and each behaves differently.

• Platform fee. Flat, monthly, independent of volume. This is what you pay for the software.
• Meta conversation charges. Paid to Meta, per conversation window, varying by category and country.
• AI usage. Paid to your model provider, driven by how much text is processed.

On a bring-your-own-keys platform like Convo360 you pay the second and third directly to Meta and to your AI provider, at their published rates, with no margin added. That makes the bill longer and the total lower, and it means you can verify every line against the provider's own reporting.

Keeping them separate also makes each one controllable. If your bill rises, you can tell immediately whether it was volume, campaigns or model choice, and each has a different remedy. A single bundled number tells you only that things got more expensive.

What Meta charges after October 2026

From 1 October 2026 each number receives 1,000 free service conversations a month. Service conversations are ones the customer started. Beyond the allowance, the Indian rate is roughly ₹0.115 each.

Template messages you send are charged separately by category. Utility templates, such as order updates, are inexpensive. Marketing templates cost more. Authentication templates sit in between.

Rates change, so treat these as the shape of the bill rather than a quotation, and check Meta's current pricing for your market before budgeting.

What the AI actually costs

This is the part people most overestimate. Modern small models are extremely cheap for chat-length text, and a support reply is short.

A realistic support exchange, including the knowledge base context supplied to the model, runs to a few thousand tokens. At current rates for a small model such as GPT-4o mini or Gemini Flash, that is a fraction of a rupee per conversation.

Two things push it up. Long knowledge base context sent on every message, and conversations that run to twenty turns because the bot is not resolving anything. The second is a quality problem that happens to also be a cost problem, which is a useful way to notice it.

Verify current pricing with your provider before quoting numbers to anybody, since model prices have moved several times in the last year and always downward.

One saving worth knowing: keep the knowledge base focused. Sending your entire policy document with every message costs tokens on every turn and usually produces worse answers, because the relevant paragraph is buried. Retrieving three or four relevant passages is both cheaper and better.

A worked example

An Indian D2C brand on a ₹2,999 plan, handling 2,000 customer conversations a month and sending 3,000 utility templates for order updates.

• Platform fee: ₹2,999
• Service conversations: 1,000 free, 1,000 charged, about ₹115
• Utility templates: 3,000 at a few paise each, roughly ₹300 to ₹500
• AI usage on a small model: roughly ₹400 to ₹900 depending on conversation length

Total somewhere between ₹3,800 and ₹4,500. For comparison, one support executive costs around ₹22,000 a month before overhead, and cannot answer at 11 pm.

The same business on a bundled per-message platform at ₹0.35 a message would pay considerably more for the identical traffic, with the difference being margin on Meta's rate.

Where costs actually run away

Three patterns account for most unpleasant surprises, and none of them is the per-unit rate.

The first is marketing broadcasts. Marketing templates are the most expensive category, and a monthly blast to fifty thousand contacts costs far more than a year of support conversations. Segment properly.

The second is a bot that does not resolve. A conversation that runs twenty turns and then reaches a human costs AI tokens and produces nothing. Reopen rate is a better early warning here than any billing alert.

The third is using an expensive model for simple work. Reserve the larger models for complex reasoning and run routine support on a small one. On Convo360 the model is chosen per agent, so this is a dropdown rather than a rebuild.

A fourth pattern catches growing teams: running everything on one number. Meta's allowances and limits are per number, so a business with distinct brands or regions often benefits from separate numbers, each with its own free allowance and its own quality rating.

How to budget sensibly

Start with your conversation count, not your customer count. Pull three months of message data and count distinct conversation windows, because that is the unit Meta bills.

Then decide your template volume deliberately. Utility messages that prevent support conversations usually pay for themselves; marketing broadcasts need a return calculation of their own.

Finally, set a spending alert with your AI provider. It costs nothing, it catches a misconfigured flow that loops, and it means the first time you learn about an unusual month is from an email rather than from a bill.

Review the whole thing quarterly. Model prices fall, your volume rises, and the mix that was right in January is rarely right in July.

And compare against the right alternative. The question is rarely whether a bot is cheaper than nothing; it is whether it is cheaper than the hours currently being spent, plus the enquiries currently going unanswered after 7 pm. Measured that way, the arithmetic is usually not close.

Frequently asked questions

Related reading

Available across India — see WhatsApp & AI calling automation by state and city.