What a WhatsApp chatbot costs: the parts nobody quotes

Any specific figure for what a WhatsApp chatbot costs is wrong for somebody, because the variable that dominates is not the build. It is the per-message charge multiplied by volume, and that depends on which countries your customers are in, which template categories you send, and how many of your conversations are started by you rather than by them.

What is stable is the structure. Knowing which components exist, and which of them scale with usage, lets you estimate your own situation and notice when a quote has left something out.

What we'll cover

• Why a single figure is meaningless
• The components
• Per-message charges
• Building it with CodeWords
• Provider margin
• The build itself
• The parts nobody quotes
• Estimating your own
• Whether it is worth it at your volume
• Making it survive
• Limits worth knowing about
• What to work out first
• Frequently asked questions

Why a single figure is meaningless

Country rates vary considerably, and a customer base in one market costs differently from the same volume in another.

Template category matters, since marketing, utility and authentication are priced differently.

Who starts the conversation matters, because a customer-initiated exchange and a business-initiated one are not the same cost.

Volume changes the rate through discounting at higher tiers.

The model has changed twice in recent years, so figures in older articles describe a scheme that no longer exists.

Estimate the structure, then apply current rates from Meta's documentation.

The components

Meta's per-message charges, which is the variable cost.

Provider fees, if you use a Business Solution Provider — a subscription, a per-message margin, or both.

Platform or builder fees for whatever you build it in.

Model costs if the bot generates responses, which scale with conversation volume.

Build effort, once, whether internal time or an agency.

Ongoing human effort: the handover queue, the review of what it could not answer, and maintenance.

Per-message charges

Meta moved to per-message pricing in July 2025, replacing the previous conversation-based model where a single fee covered a 24-hour window.

Template messages are the charged ones, by category.

Service messages have been treated differently, and Meta has announced changes taking effect in October 2026 affecting service messages and utility templates inside the service window.

Rates are per country, which for an international customer base means a weighted average rather than a single figure.

Volume discounts exist at higher tiers for some categories.

Check Meta's current pricing page rather than any article, including this one, because this specific area has changed repeatedly.

Building it with CodeWords

CodeWords connects to more than 3,000 integrations, and the connection is made once and reused.

1. Open CodeWords and start a new automation.
2. Describe what should happen in plain language to Cody, the automation builder: what the bot handles and when it hands to a person.
3. Authorize the connection to your WhatsApp Business Platform account and the systems it needs.
4. Describe the exceptions: a template send, a message outside the window, a handover, a failed send.
5. Run it at low volume for a month and use the actual message counts to estimate at scale.

You describe the outcome; Cody builds it, connects it, and deploys it. The free plan covers light use, with Pro at $39 per month and Business at $100 per month as usage grows; details are on the pricing page.

Provider margin

A Business Solution Provider adds a fee, and how it is structured varies more than the amount.

Per-message margin dominates at volume, even when the subscription looks larger at the point of purchase.

Some bundle the platform, which is convenient and raises the switching cost.

Ask for the effective rate at your projected volume rather than the headline.

Direct Cloud API has no margin and more operational work, which is a real trade rather than an obvious saving.

Compare over a year at projected volume, not at current volume, since the structures cross over.

The build itself

The engineering is the smaller part, which surprises people.

Approvals take calendar time — verification, display name, templates — which is elapsed rather than effort.

Conversation design is the real work, and it is the part most often underestimated because it looks like writing.

Integration with your systems is where the effort concentrates, particularly with an older booking or order system.

A first version is days, a version you would put in front of customers is longer.

Agency quotes vary enormously, and the spread is mostly about how much conversation design and integration is included.

The parts nobody quotes

The handover queue. Somebody answers the conversations the bot does not, and that is a staffing cost proportional to volume.

Reviewing what it could not answer, weekly, which is the work that keeps it improving.

Template maintenance as policy and product change.

Monitoring and the response to a quality rating decline.

Model cost growth, which scales with conversations rather than with your revenue.

The second version, because the first one will need changing once real conversations exist.

Estimating your own

Count your current contact volume by type, which most businesses have approximately.

Split it into customer-initiated and business-initiated, since they cost differently.

Estimate how many the bot handles without a person, conservatively, and treat the rest as handover.

Apply current rates for your countries and categories.

Add the provider margin at that volume.

Add a person's time for the handover queue, which is the line most often missing and frequently the largest.

Whether it is worth it at your volume

Cost is only half the question and the other half is rarely asked properly, because the comparison is usually against doing nothing rather than against the alternative.

Below a certain volume a person is cheaper, straightforwardly, and that threshold is higher than most vendors suggest.

The saving is fragmented where contacts are spread thinly across a day, since an hour saved in six-minute pieces is not an hour recovered.

Response time may be the real benefit rather than cost, particularly where a fast reply wins work that a slow one loses.

Compare against the actual alternative — a shared inbox with two people, or an answering service — not against an imagined worst case.

Measure the baseline first, since the saving claimed afterwards is otherwise an estimate made by whoever wanted the project.

Deciding not to build it is a legitimate outcome, and considerably cheaper than a bot nobody maintains.

Making it survive

Alert on cost anomalies, since a loop or a misconfigured template can multiply spend quickly.

Report cost per resolved conversation, which is the number that means something.

Watch model spend separately, because it grows with conversation length as well as count.

Review template usage quarterly and retire the ones that are not earning.

Report the outcome: messages by category, cost, handovers, and cost per resolution.

Limits worth knowing about

Pricing changes, and has done repeatedly, including a change scheduled for October 2026.

Country mix drives the average more than anything you control.

Quotes that exclude the handover are describing part of the system.

Model costs are volatile in a way platform fees are not.

Savings claims should be checked against a measured baseline rather than an estimate made before building.

What to work out first

Your split between customer-initiated and business-initiated conversations, by country. That single breakdown determines most of the variable cost, it is available from your existing contact volume, and it turns a question with no answer into an estimate you can defend.

Two habits make the difference. Include the handover queue in any cost estimate, since it is proportional to volume and is the line most often omitted from both internal plans and agency quotes. And report cost per resolved conversation rather than cost per message, because the second number always looks better and tells you nothing about whether the thing is worth running.

Frequently asked questions

Why can nobody quote a figure?

Because the dominant variable is per-message charges multiplied by volume, and rates differ by country and by template category. The same bot costs very differently depending on where your customers are.

What are the actual components?

Meta's per-message charges, provider fees, platform or builder fees, model costs if it generates responses, the one-off build, and the ongoing human effort of the handover queue and weekly review.

How does the pricing work now?

Per message for template messages since July 2025, replacing conversation-based charging, with rates by country and category and volume discounts at higher tiers. Changes take effect in October 2026, so use Meta's current documentation.

Is a provider worth the margin?

It depends on your volume and appetite for operational work. Per-message margin dominates at scale even when the subscription looks larger at purchase, so compare the effective rate over a year at projected volume rather than the headline.

What is usually underestimated in the build?

Conversation design, because it looks like writing, and integration with existing systems — particularly older booking or order systems. The engineering is the smaller part and approvals take calendar time rather than effort.

What gets left out of quotes?

The handover queue, which is a staffing cost proportional to volume and is frequently the largest ongoing line. Then template maintenance, weekly review of unanswered questions, monitoring, and the second version.

How do I estimate my own cost?

Split your current contact volume into customer-initiated and business-initiated by country, estimate conservatively how many the bot handles without a person, apply current rates and provider margin, and add the staffing for everything handed over.

What number should I report?

Cost per resolved conversation. Cost per message always looks better and says nothing about whether the system is doing its job.

When is a bot not worth the cost?

Below the volume where a person is simply cheaper, which is a higher threshold than vendors suggest — and wherever the saving is fragmented across a day, since an hour recovered in six-minute pieces is not an hour.

What is often the real benefit?

Response time rather than cost, particularly where a fast reply wins work a slow one loses. That is worth measuring separately, since it justifies the spend on different grounds from headcount.

Related reading

• How to build a WhatsApp chatbot
• Choosing a WhatsApp bot building approach
• WhatsApp messaging limits explained