Overview
Sending optimization changes how support replies go out on WhatsApp, without changing what the agent writes. It has four independent parts, each with its own switch:
- Reply consolidation: when an agent confirms several short messages in a row, the platform waits a few seconds and sends them together as a single message. Each confirmation becomes a part of the pending reply.
- Review by Maestro: before sending, Maestro can fix spelling, adjust formatting or propose an interactive format. You choose whether it only suggests or may improve automatically within rules.
- Interactive formats: Maestro can propose buttons, a list, a link with a button or a WhatsApp Flow that is already published and authorized for the inbox.
- Automatic distribution per contact: in a unified hybrid inbox, a share of the eligible support sends goes out through WhatsApp Web and the rest through Cloud API.
None of this creates hidden messages or resends on assumption. While parts are waiting, no message has been sent: the contact receives nothing, and the conversation does not count a first reply or SLA. The final message is created only once, and it is the one that triggers automations, notifications and the cost record.
Sending optimization does not guarantee exact savings, and the savings summary is not an invoice. Since 2026-10-01 a rate applies to the service message (the support reply over the Cloud API inside the 24-hour window). The platform uses an estimated monthly allowance of 1,000 service messages per number. Estimated monthly allowance; Meta does not publish this limit in the documentation tracked here. Merging parts avoids messages. Replying over WhatsApp Web avoids service messages over the Cloud API, and the reply over Web is not billed by Meta, because it does not go through the Cloud API. After the month's 1,000 in the estimate, each message avoided represents one projected charge less (estimated savings in money); before that, it represents estimated allowance saved. The real amount also depends on the market, the price table in force and the Meta invoice.
Prerequisites
- A WhatsApp Cloud inbox or the logical inbox of a unified hybrid inbox. In a hybrid inbox, the WhatsApp Web inbox shows the owner inbox's settings, read-only.
- The matching features turned on for the account: reply consolidation; review by Maestro; interactive formats; and distribution per contact. All of them start turned off.
- To configure the inbox: permission to manage the inbox. For distribution per contact: the administrator role.
- For review by Maestro: an account bot linked to a Maestro agent, authorized in the inbox's profile list.
- For WhatsApp Flow: a published Flow in the same inbox, with a Cloud API connection.
- To see the summary values in the Cost Center: access to the WhatsApp Cost Center.
Step by step
Turn on reply consolidation
- Go to Settings → Inboxes → your inbox → Configuration and scroll to Reply consolidation.
- Turn on Combine parts before sending.
- In Wait after the last part, choose 5 s, 10 s or 15 s, or enter another value from 1 to 30 seconds.
- In Total wait limit, enter how long the reply may wait after the first part, up to 60 seconds.
- Click Save consolidation. Nothing is sent when you save; the rule applies to the next replies.
Reply with consolidation turned on
- In the conversation, write and confirm the text as usual. The button is now called Add to reply.
- The part shows up in the Your reply strip, above the editor, with the countdown and the expected sending time.
- Keep adding parts if you need to. Each new part restarts the wait, without going past the total limit.
- To send earlier, use Send now. To review, use Pause and review. To give up, use Cancel reply.
- When the wait ends, the platform creates one message with the parts in the order they were confirmed, separated by line breaks.
Turn on review by Maestro
- In the same section, open Review by Maestro.
- Choose the Review mode: Keep the original text, Suggest and ask for approval or Improve automatically.
- Authorize the Allowed Maestro profiles and select what Maestro can adjust.
- Choose what happens If Maestro fails and, if you want, the Daily review limit.
- Save. When you choose Improve automatically, the platform asks for a confirmation that lists operations, formats and the failure rule.
Authorize a WhatsApp Flow
- In Allowed formats, select WhatsApp Flow.
- In Allowed WhatsApp Flows, click Add WhatsApp Flow.
- Choose the published Flow, the Entry screen, the Label the agent will see, When to suggest and the Button text.
- Save. The platform reads from the chosen screen the fields the Flow needs to receive; you do not edit any data schema.
Turn on distribution per contact
- Go to Settings → Inboxes → your Cloud inbox → Hybrid.
- With the unified inbox active, turn on Automatic distribution per contact.
- Enter the Percentage through WhatsApp Web. The percentage through Cloud API is calculated automatically.
- Click Save configuration.
Settings & options
Waiting time
- Wait after the last part (1 to 30 s, default 10 s): the reply goes out when nobody adds parts for that long.
- Total wait limit (up to 60 s, default 30 s): the reply never waits longer than that after the first part. Editing or removing a part does not extend this limit.
- A pending reply accepts up to 32 parts and 16 KiB of text. Beyond that, send it now or review it before adding more.
- The clock belongs to the server. Closing the tab, switching conversations or reloading the page does not restart the wait or duplicate the reply.
Review modes
- Keep the original text: parts are combined exactly as written. Maestro is not called.
- Suggest and ask for approval: Maestro proposes a version and shows the differences. Nothing goes out until you approve, edit, send the original or cancel. The suggestion never turns into an automatic send when time passes.
- Improve automatically: Maestro applies the allowed operations without asking for approval on each message. It only applies without review the operations and languages that passed the platform's quality evaluation. While an operation has no approved evaluation — which is the case for all of them at launch — the result goes to your review, as in the Suggest mode. Cases that Maestro flags for review also go to you.
- Each agent can restrict their own mode in Preferences, within what the inbox allows. The preference never widens what the inbox authorizes.
If Maestro fails
- Pause for review (recommended): the reply stays paused until the agent decides.
- Send the original text: when Maestro does not answer in time, the parts go out as written, as long as the original text is valid.
- The daily review limit counts requested reviews or generated tokens. When the limit is reached, the failure rule applies until the next day.
- If a review has no complete usage measurement, that attempt's reservation continues to occupy the daily limit. Even when some tokens are known, the platform does not release the unconfirmed part; this may prevent another review when the ceiling is reached. When tokens are complete and only their price is unknown, the limit uses the measured quantity normally. The reservation controls reviews or tokens without creating a monetary charge.
Formats
- Text is always allowed.
- Buttons, List and Link with a button use formats WhatsApp already accepts. A link with a button only uses an address that is already in the text of the reply or that the team (an agent or a bot) already sent in this conversation; in the second case, only if its domain is on the allowed list. Links sent by the contact are never used, and Maestro never makes up a link.
- WhatsApp Flow: Maestro only chooses among Flows that are published and authorized in the inbox, and never creates or publishes a new Flow. A WhatsApp Flow always goes out through Cloud API and does not count in the distribution per contact. At launch, a Flow proposal is marked Not verified and available for tests only, until the platform confirms the send on a real account.
Distribution per contact
- The proportion applies to support sends from the composer that can go out on both channels, counted per contact. A consolidated reply or a group of attachments counts as one send, and all its parts go out on the same channel.
- The choice is deterministic and reproducible: with 30% through WhatsApp Web, for every 10 eligible sends to the same contact, 3 go out through WhatsApp Web.
- Capability and safety come first: a template, a WhatsApp Flow, a quote that only exists on one channel, a closed 24-hour window or an allowed manual choice set the channel without using up the proportion.
- A contact without a known phone number (shown only by their WhatsApp username) is left out of the percentage split, and replies to them go out through the Cloud API.
- Saving a new percentage restarts the count for all contacts. Merging contacts keeps the count of the contact that remains.
- Campaigns, bots, the API and scheduled messages are not part of this proportion.
Privacy and retention
- The text of parts, suggestions and drafts is deleted 7 days after the reply is sent, canceled or expires. You can change this period in Privacy.
- A reply left untouched for 24 hours without being sent expires without sending and follows the same retention rule.
- The reply's audit records keep only identifiers and codes, never the text, and are kept for 90 days.
- The sent message follows the conversation's normal retention. Data subject erasure requests also delete the pending text before these periods.
Summary in the Cost Center
The Overview tab of the WhatsApp Cost Center shows, per inbox and period, the Support sending optimization: confirmed parts, messages created and messages avoided; sends reserved, attempted, accepted, delivered and read per channel; and the estimated savings, split by source: consolidation (messages avoided by merging parts) and sending over WhatsApp Web (replies that did not go out over the Cloud API).
The savings take into account the number's estimated monthly allowance of 1,000 service messages. While estimated usage has not gone past 1,000 in the month, what was avoided shows as estimated allowance saved, in number of messages. The money amount projects only messages that would have been billed above the estimated allowance; the limit does not confirm a charge by Meta. Canceled, expired or rejected replies, replies with unconfirmed sending and replies that were not delivered do not count as savings. Only service messages Meta delivered, billed or treated as inside the estimated allowance, count toward the money projection. Messages avoided inside an ad's free window would have been free: they show as volume only. A message that has no confirmed price yet is left out of the savings and counted apart, without hiding the rest. Projection unavailable appears only when no avoided message of the period has a confirmed price; in that case the summary shows the volume. Maestro usage shows up separately and is not part of the savings. The savings remain an estimate, with no guarantee of an exact amount, and are not an invoice.
Use cases
- An agent who writes in several short lines: the contact receives one complete message instead of five notifications in a row.
- A team with a writing standard: Maestro fixes spelling and formatting, and the agent approves before sending.
- Offering a form: Maestro suggests the authorized scheduling WhatsApp Flow when the contact asks for a time slot.
- A number in a hybrid inbox: part of the support sends goes out through WhatsApp Web, in the proportion set by the administrator.
Tips, limits & best practices
- Start with Keep the original text and a 10 s wait. Adjust after looking at the summary in the Cost Center.
- Prefer Pause for review as the failure rule. Sending the original is useful only when review is optional in your operation.
- What goes to the AI: when you use Maestro, the reply draft and the recent text of the conversation (up to 20 text messages and at most 16 KiB, most recent first) are sent to the AI model configured in Maestro to produce the suggestion. Attachments and files are never sent. Maestro never makes up a link, and a link written by the contact never becomes a link with a button.
- Do not use consolidation to get around the size limit: the platform does not split a long reply into several messages. Above the WhatsApp limit, the reply pauses for you to shorten it.
- Automatic outgoing translation, when turned on in the inbox, happens only once, on the final message. The review preview shows the text before translation.
- The agent's signature is applied only once, at the end of the consolidated message, not to each part.
- The switch to Cloud API only happens BEFORE sending: if WhatsApp Web is unavailable or at its safe sending limit, the whole reply goes out through Cloud API, when allowed, and the count moves on normally. Once WhatsApp has received the attempt, the platform does not change channels, and the distribution does not try to make up for it later.
- WhatsApp Web's safe sending limits still apply. The distribution never raises those limits.
Troubleshooting
- The reply was paused with "The contact replied": the contact wrote before the send. Review it and use Resume sending, Send the original text or Cancel reply. Once the reply has gone out for sending, the platform does not promise to undo it.
- "You no longer have access to send in this conversation": the conversation changed agent or team, or you lost access. In this version, resolving or transferring the conversation does not stop the wait; the send is checked at the last moment and, without access, the reply is paused. Someone with access can cancel it.
- "Sending on hold": there are two cases. The send was stopped before going out, for example by WhatsApp Web's safe sending limit, or WhatsApp rejected the message definitively. In both cases, the platform does not try the other channel on its own. Use Release the same message, which tries the same message again without creating a copy; this new attempt counts again toward the WhatsApp Web sending limit when it goes out through it. Or use Give up on sending. The next replies in the conversation wait for this decision to keep the order.
- "Sending not confirmed": we do not know whether WhatsApp received the message. The platform never resends it on its own, nor through the other channel. Use Update status to check. If you give up, the message will not be resent; if WhatsApp confirms it later, the conversation will show it.
- The message's "try again" button became "Resolve sending": messages from this optimization are resolved in the pending reply strip, with the actions above.
- "The suggestion is out of date": the reply changed after the suggestion. Generate a new suggestion or send the original text.
- The suggested WhatsApp Flow is unavailable: the Flow was unpublished, revoked or changed, or the conversation is pinned to WhatsApp Web. Send the original text or cancel.
- "The settings changed in another session": someone else saved first. Click Load current version, which discards your unsaved changes, and make your change again.
- The feature was turned off with pending replies: nothing is sent or deleted because of that. Pending replies and held sends stay visible so you can cancel them or give up; releasing requires the current settings to still allow the send.