How Replyk Lowers the Number of WhatsApp Messages Meta Bills You For

Your invoice is two numbers multiplied together

Meta's rate card sets one of them and you cannot negotiate it. The other one is how many messages your business delivers, and that is entirely yours. From 1 October 2026 the second number matters more than it ever has, because service messages, the ordinary replies you send while a chat is open, stop being free.

Most software treats message count as a side effect of whatever the model happens to produce. Replyk treats it as a design constraint. What follows is the list of mechanisms, each with the actual behaviour rather than the marketing summary, so you can check them against your own inbox.

One: finish the sale inside the 72 hour free window

This is worth more than the other seven combined, so it goes first. A conversation that starts from a Click-to-WhatsApp ad and gets a reply inside the first 24 hours opens a 72 hour window in which every message, of every category, costs nothing. The rate card simply does not apply.

The expensive pattern is not an expensive message, it is a slow sale. A lead that sits unanswered for six hours, gets a reply, goes quiet, gets a follow up on day two, and finally decides on day four has spent most of its life outside the free window. Every message after hour 73 is billed at the full rate, and a COD confirmation flow can easily be six or seven of them.

The Agent answers in seconds, at three in the morning and on Eid, works the objection, and asks for the order while the window is wide open. It is handed the exact remaining hours on the window as a number in its context, not a vague instruction to hurry, so a follow up it schedules for tomorrow is scheduled against a real deadline.

A conversation that ends in a confirmed order on day one costs $0.00, whatever Morocco's new rates say.

Two: wait while the customer is still typing

Real customers do not write paragraphs. They write "salam", then "chhal taman", then "kayna f noir?", in three messages over twenty seconds. Answer each one as it lands and you have delivered three messages where one would do, and you have also answered the first question without knowing the second.

Replyk collects a customer's burst and treats it as one turn. The Agent waits for 30 seconds of silence before it answers, so the three fragments become one question with one answer. There is a hard ceiling of 5 minutes: a customer who keeps typing without pausing gets answered anyway, because a perfect batch that arrives too late has already cost the sale.

The claim is checked every five seconds across the whole system, so the wait is a real 30 seconds rather than a polling artefact. On a burst of four fragments this alone removes three billed messages, and it makes the answer better, because the Agent reads the whole question before replying to any of it.

Three: one bubble is the default, four is a hard ceiling

An AI that greets you in one message, answers in a second, and asks a follow up question in a third has just tripled the cost of one turn for no benefit. This is the single most common way automated WhatsApp bills get out of hand, and it is invisible until the invoice arrives.

Replyk's Agent is instructed that one bubble is the default and that a second one has to be earned by genuine conversational rhythm, never by splitting a thought or padding. Below that instruction sits a delivery layer that caps any single turn at four bubbles and merges anything beyond that into the last one, so a model having an off day cannot run up the bill regardless of what it generates.

Nothing the Agent wrote is discarded when the cap applies. The content arrives, in fewer delivered messages.

Four: one carousel instead of a stack of images

Showing a customer six products as six image messages is six billed messages, and it is also a worse experience: the photos push the conversation off the screen and the customer scrolls back looking for the one they liked.

Replyk sends them as a single interactive carousel. One delivered message carries up to ten product cards, each with its photo, name, price and a button, swiped sideways like a catalogue. One charge instead of six, and the customer taps the product they want instead of describing it.

The same logic applies to the order form. Collecting a name, phone, city and address as four separate questions is four delivered messages and four chances to drop out. Sent as one structured form, it is one.

Five: junk never reaches a billed reply

A wrong number costs exactly as much to answer as a buyer. So does a competitor fishing for your prices, a bot, and the person who has asked the same question four times and never ordered. In a market where Click-to-WhatsApp ads bring volume, this is not a rounding error.

Replyk scores inbound conversations for spam and flags them before a reply is generated. You set how much evidence is required before the flag fires, and what happens next: silence the Agent, notify a human, tag the conversation, or block it outright. A flagged chat consumes neither a Meta message nor an AI turn.

The same subsystem handles the quieter version of the problem: conversations that should stop. A chat that has gone nowhere over many exchanges can be closed automatically rather than left to accumulate polite follow ups that are billed one by one.

Six: handover without double answers

The most expensive pattern in a shared inbox is two answers to one question. A human replies, the automation replies as well, the customer gets contradictory information and asks a clarifying question, and the store has paid for three messages to create confusion.

When someone on your team takes over a conversation, the Agent stands down for that conversation and stays down until the chat is released back. When it is released, the Agent resumes with the full history, so the customer is not asked anything twice.

The switches are per channel and per number, not global. Turning the Agent off on one WhatsApp number does not silence another, and does not touch your Facebook Page or Instagram. There is a business wide pause when you need one, but it is a deliberate act rather than a side effect of a setting three menus away.

Seven: follow ups that check the window before they send

Automated follow ups are where WhatsApp bills quietly grow. A reminder scheduled for tomorrow at ten is trivial to configure and, if tomorrow at ten is past hour 73, it is a paid template that nobody decided to buy.

Every scheduled message in Replyk is evaluated against the real window state stored on the conversation before it goes out. If it would land inside the free window, it goes. If it would land outside, it is a paid send, and paid sends after the free window require a switch that is off until an owner turns it on.

This is deliberately stricter than it needs to be. The alternative, silently sending and letting the merchant find out on the invoice, is how trust in automation gets lost.

Eight: no markup on what Meta charges

Some platforms resell WhatsApp messaging with a margin on top, or a percentage on your messaging spend. Replyk does not. Meta bills the payment method on your own WhatsApp Business Account, directly, at Meta's published country rate.

That means the savings described above are real savings to you rather than a smaller number on a bill that has a margin folded into it. You pay Replyk for the software and for AI usage, and nothing at all on your message volume.

What it adds up to on one conversation

Take one customer who sends three quick messages asking about two products, and who ends up ordering. Outside any free window, at Morocco's rate of $0.0230 per delivered message, here is what each mechanism removes.

Nine delivered messages becomes three. The customer is told exactly the same things. And if that conversation had arrived from a Click-to-WhatsApp ad and closed inside 72 hours, both the nine and the three would have cost nothing at all, which is why mechanism one comes first.

Frequently Asked Questions

Do I have to configure any of this? No. The batching window, the bubble defaults, the carousel format and the window aware follow ups are on by default. The spam threshold and the paid message switch are yours to set, and both ship on conservative values.

Does waiting 30 seconds make the Agent feel slow? In practice it feels more human, not less. People expect a beat before a reply, and answering a half finished question is what actually feels robotic. The wait only applies while the customer is still sending; once they stop, the answer comes within seconds.

Does this work on Messenger and Instagram? The message shaping does. Batching, one message per turn, spam filtering and clean handover apply on every channel. The 72 hour free entry point and the per message rate card are WhatsApp concepts and have no equivalent on Messenger or Instagram, which use a 24 hour window with no template path.

Will fewer messages hurt my conversion rate? The opposite, in our experience. A single clear reply outperforms a wall of bubbles, and a carousel of six products converts better than six photos the customer has to scroll back through. The cost saving is a consequence of writing better messages, not a trade against it.

Can I see what a conversation would look like before going live? Yes. The sandbox renders exactly what a real customer would receive, including carousels and forms, using the same delivery path, so the bubble count you see there is the bubble count you would be billed for.

Does Replyk charge per WhatsApp message? No. Meta bills message fees directly to your own WhatsApp Business Account at its published rates. Replyk charges for the platform and for AI usage, with no percentage on your messaging.