A WhatsApp chatbot can create two separate monthly bills. The chatbot provider charges for the agent and its usage allowance. Meta may charge for messages the business delivers through the WhatsApp Business Platform, according to category and recipient market. Agentkit counts visitor messages against its plan allowance; Meta charges, where applicable, attach to delivered outbound messages. One inbound message is not one billable Meta message.
Use this worksheet for a normal month and a busy month. Leave headroom for the busy one. Fill the latest rates from Meta's pricing FAQ for the countries you actually message, and fill the plan price and allowance from your provider's current terms. A blank rate means unknown, not free.
| Input | Normal month | Busy month | Where to check |
|---|---|---|---|
| Incoming visitor messages across all connected channels | ___ | ___ | Chatbot usage records |
| Plan allowance needed for incoming messages | ___ | ___ | Provider usage definition |
| Monthly plan price | ___ | ___ | Provider pricing and billing cycle |
| Delivered marketing messages × applicable Meta rate | ___ × ___ | ___ × ___ | Meta category and recipient market |
| Delivered utility templates × applicable Meta rate | ___ × ___ | ___ × ___ | Meta category and market |
| Service messages beyond the monthly free allowance × Meta rate | ___ × ___ | ___ × ___ | Meta rate card and business phone number |
| Delivered authentication messages × applicable Meta rate | ___ × ___ | ___ × ___ | Meta category and market |
| One-time setup and staff work, shown separately | ___ | ___ | Your implementation scope |
Monthly estimate = provider subscription + applicable Meta message charges + any documented provider extras. Keep one-time setup outside that total. Add a Meta row for each destination market and category. Add any provider markup or inbox seat charge explicitly. Meta prices delivered messages, so a conversation count is not the right multiplier.
First bill: the chatbot plan and its allowance
Providers use different meters. Some count conversations, some count AI replies or credits, and some count incoming messages. Ask for the unit definition before turning a traffic forecast into a plan choice. The chatbot pricing models guide has a broader vendor comparison worksheet.
For Agentkit, the usage guide says each visitor message counts once; the agent's answer does not count separately. Its allowance is shared by all chatbots in a workspace. WhatsApp is available on paid plans, and the current Agentkit pricing page lists the plan's message allowance, chatbot count and seats. Budget conservatively from incoming visitor messages. The public usage guide does not specify whether messages during human takeover count, so check your Usage panel before relying on a lower estimate. A business already using the website widget must include that channel in the same capacity estimate.
The WhatsApp connection checks whether the agent can respond before generating an AI answer. If quota is exhausted, the webhook records the inbound as missed and sends no AI answer. Assign a person to follow up on that message; the code path does not schedule an automatic reply after a reset or upgrade. The WhatsApp setup guide explains the paid-plan and Business app requirements.
Second bill: Meta's delivered messages
Meta describes four Business Platform categories: marketing, utility, authentication and service. Its pricing page says charges depend on delivery, recipient market and category. Record delivered marketing, utility and authentication messages separately, and check how Meta counts service messages against your free allowance.
A customer message opens or resets a 24-hour service window for ordinary replies. After that window closes, Agentkit lets a teammate send an approved template. Meta also describes a separate 72-hour pricing exemption after certain ad and Facebook Page entry points; verify that exemption against your current account terms before using it in a quote. A delayed inbound webhook can arrive after the window has closed, in which case Agentkit records the message without generating an AI reply.
The 24-hour window is not a free-reply allowance. Meta's pricing FAQ says that, effective October 1, 2026, it charges for service messages from the 1,001st service message per business phone number each month. The first 1,000 are free. Meta's general pricing landing page still says service messages and in-window utility replies are free, which conflicts with the newer dated FAQ. Check the current rates linked from the FAQ and the connected account's actual bill for a quote. Check utility-template charges too; do not apply the service free allowance to another category. Message classification, delivery and recipient market all matter.
Meta says in its Business Platform FAQ that the WhatsApp Business app is free to use. Eligible traffic on the connected Business Platform can incur Meta charges. If you keep your existing Business app number while adding AI, Meta's Platform rules still apply to the connected automated and template traffic. The number-retention guide covers human takeover and the connection check; use this page to price the traffic.
Work through a normal and a busy month
Here is an illustrative workload, not observed Agentkit customer usage or a market price. One Agentkit bot serves both a website widget and WhatsApp. Suppose it receives 300 incoming WhatsApp messages in a normal month and 1,800 in a busy one. The website channel receives another 200 and 500 visitor messages respectively. On Agentkit's documented counting rule, the shared plan needs room for 500 visitor messages normally and 2,300 in the busy month. On October 20, 2026, the monthly Hobby plan costs $29.99 for 2,000 messages, one bot and one seat. It fits the first estimate but not the second. The monthly Standard plan costs $119.99 for 12,000 messages, two bots and three seats. It covers both estimates, if those other limits fit. An upgrade rather than an assumed fixed overage belongs in the busy-month budget; check the live plan page before purchase.
Now suppose one connected business phone number sends 240 service replies in the normal month and 1,500 in the busy month. The October 2026 rule leaves zero service replies to price in the normal month and 500 in the busy month, assuming no other service traffic on that number uses the allowance. These service replies are separate from the 300 and 1,800 incoming WhatsApp messages in the plan estimate.
The business also expects 40 delivered marketing templates in a normal month and 150 in a busy month, plus 15 and 50 delivered utility templates. It sends to one market. Let S, M and U be Meta's applicable service, marketing and utility rates in the chosen currency. The normal-month Meta estimate for these categories is 40 × M + 15 × U. The busy-month estimate is 500 × S + 150 × M + 50 × U. Fetch rates from Meta for that market on the day you quote. Utility and authentication rates can change across volume tiers, so check the applicable tier for a busy month. Add other categories the business sends, and confirm any discount against the connected account's bill. This calculation is illustrative; use your own traffic and Meta account rates for a quote.
Before paying for a larger AI tier, check your own incoming traffic and human takeover activity. After connecting, compare the first full billing period against the Agentkit Usage panel and the connected WhatsApp Business account's Meta charges. The Agentkit connection guide explains how approved templates work. The Meta FAQ says a solution provider may receive Meta's bill, so confirm who invoices you during setup.
Quote checks that prevent an expensive surprise
Ask for the price of the specific configuration you intend to run: number of agents and seats, monthly billing versus annual prepayment, included messages, the behavior at the limit, and any provider markup on Meta traffic. For Meta, record destination markets, template categories, delivered counts and the rate-card date. Budget setup and staff work separately, especially if the team must write approved answers, test takeover or create templates.
Test with an opted-in number before moving a busy support line. Confirm the AI answers an approved question, a person can pause it, and a teammate can send an approved template after the window closes. Compare your estimate with the first complete bill.
