The ManyChat account has been dormant for about a year, and the fortnight it was meant to run in 2025 mostly never ran. This plan rebuilds the flows around two angles, sets up a real year-on-year baseline, and says what has changed on both platforms since.
The whole thing in about two minutes.
Eighteen posts were scheduled between 15 July and 1 August 2025. Two show as posted. Any conclusion drawn about whether the approach works was drawn from a fortnight that mostly did not run.
The subscriber column was left empty for every row. The year-on-year read has to come from the flow node statistics still sitting in ManyChat, and from re-running two of the original posts against today's audience.
5,110 contacts entered the main flow. The larger branch, 3,063 of them, dead-ends at a tag and is never asked to opt in. On the branch that does ask, 1,559 passed the delay and only 295 were sent the ask.
The gated-URL flow asked in message two and converted at roughly 10% per entrant against under 5%. Decide before posting whether a post is a list post or a traffic post, and build the flow to do that one thing.
Sends run at 78.2%, 83.7%, 59.7%, 72.3% and 70.7%, where healthy Messenger should be well above 90%. A 23-hour delay sits on the edge of Meta's 24-hour window and delivers at 12.5% with zero clicks.
It is the Recurring Notifications opt-in. Meta kept it running for the UK when it switched it off elsewhere, and the paid replacement is not available to UK businesses. Before anything is built on that list, send one broadcast to about twenty recent contacts and see whether it arrives.
Engagement bait is demoted harder at Page level than post level. The mechanic still works; the cadence has to change, with comment posts interleaved with filmed content, winner content and draw content.
Engagement bait costs reach. The spam policy, which is enforcement rather than ranking, prohibits "offering giveaways ... of cash or cash equivalents in exchange for engagement". A £100 cash freebie entered by commenting is squarely that. Part seven says what to do about it.
It can send promotional template messages outside the 24-hour window, which is exactly what Messenger is losing, and it should be cheaper than the SMS the client sends now. Weeks 5 and 6, but the consent wording has to be right from the first opt-in in week one.
The last scheduled run covers 15 July to 1 August 2025: eighteen planned posts, daily, going live around 19:05, each a "this or that" prompt attached to a £100 giveaway and pointing at the complimentary competitions category on the site.
Execution stopped almost immediately. Only two posts show as Posted. One workflow shows as Complete. The rest never started.
The subscriber column was never filled in. "Number of Subscribers today" is empty for every row, so there is no recorded baseline.
Minor: the 31/07 row reads "£1000 freebie" where every other row reads £100. Confirm which was correct before that copy is reused.
5,110 contacts in, roughly 239 opt-ins out
| Step | People | Note |
|---|---|---|
| Entered the flow | 5,110 | |
| Non-match branch | 3,063 | Dead-ends at an "Add Tag: Website click" action and is never asked to opt in |
| Condition branch | 2,047 | |
| Passed the 25-minute delay | 1,559 | Only 295 were then sent the opt-in message, with no condition between the two to explain the gap |
| Sent the opt-in ask | 295 | 81% clicked through |
| Opt-ins | about 239 |
238 in, about 23 opt-ins, roughly double the rate per entrant
| Step | People | Note |
|---|---|---|
| Entered the flow | 238 | |
| Reached the opt-in ask | 167 | 14% converted |
| Opt-ins | about 23 | |
| After the 23-hour delay | 12.5% delivered | Zero clicks. The delay sits on the edge of Meta's 24-hour window, so that follow-up has effectively never reached anyone |
Per person entering, Flow B converted better, roughly 10% against under 5%, because it asks immediately rather than behind a delay and a link click.
Inference from two flows of very different volumes, not a clean test.
78.2%, 83.7%, 59.7%, 72.3%, 70.7%, then 12.5%. Healthy Messenger sends should run well above 90%.
The trigger set contains a dozen overlapping copies (Facebook Comments #7 copy copy copy…). Where a trigger with no keyword was created before one with a keyword, the keyword-less one fires regardless.
One branch, a 45-minute delay into "just a reminder", has never run at all and has an unconnected next step.
Check this before anything else, because it stops everything dead: ManyChat's comment trigger will not work at all if the Facebook Page carries a country or age restriction. This is an 18+ product, so an age restriction may well have been set at some point in good faith. Ten seconds in Page settings, and nothing below runs without it.
The mechanic: a free giveaway post on the Page. Someone comments. The comment trigger drops them into a flow, and the flow has one of two jobs depending on the post: add them to the broadcast list, or send them to a specific page on the site.
One goal per post, not both. The old Flow A tried to do the website click and the opt-in in the same sequence, with the opt-in behind a 25-minute delay, and lost most of the audience between the two. Decide before posting whether this is a list post or a traffic post.
Ask earlier. Flow B asked in message two, before any link click, and converted at roughly double the rate per entrant. That is the default shape.
| Step | What it does |
|---|---|
| 1 | Comment trigger: one keyword, one post |
| 2 | Greeting plus the ask (list or link, never both) |
| 3 | Tags: source post, action taken, first-time or returning contact |
No delay over 18 hours. Everything beyond that falls outside the messaging window.
In the old build, people already on the list routed down the non-match branch and hit a dead end. That branch should carry the sales message, tonight's draw, closing soon, check your entries, because those are the warmest people in the account.
The mechanic: the same this-or-that comment post, but the flow delivers a discount code rather than an opt-in ask. The post is then boosted through Meta ads to push reach past the existing follower base. New engagers land in a flow built to convert rather than to collect.
Why it is stronger. Angle 1 harvests an audience the Page already has. Angle 2 buys a new one, and the discount code gives the flow a reason to exist beyond "join our list": it creates a measurable purchase event rather than a soft opt-in.
| Step | What it does |
|---|---|
| 1 | Comment trigger, which fires on boosted posts as well as organic |
| 2 | Greeting, unique discount code, link |
| 3 | Tags: paid or organic, code issued, code redeemed |
Unique or trackable discount codes on the site. Without them there is no attribution and this collapses into Angle 1 with a budget attached.
Tracking that connects a Messenger click to a completed purchase.
Meta gambling and gaming authorisation on the ad account, if it is needed. Meta's current policy: "Ads that promote online gambling and gaming are only allowed once an ad account has obtained authorisation", requested through the Authorisations and verifications tab in Meta Business Suite with evidence of licensing. It is granted to specific ad account IDs, and the UK is not on the unsupported list.
Whether Palladium needs it is worth establishing rather than assuming. Meta exempts "prize promotions that involve purchasing a product at the usual retail price as long as online gambling isn't the primary business model", but it defines the restricted category as anything where "anything of monetary value is included as part of a method of entry and prize", and names sweepstakes among the types caught. Paid entry with a cash or car prize sits close to that line. Get it confirmed before any spend, because ads are rejected without it.
"Real Money Gaming" and "prior written permission" are retired Meta wording for ads. The organic Pages policy does still use "prior, written permission", which is a separate matter from advertising.
Ads, dark posts and boosted posts all support ManyChat's comment trigger, so the flow fires on paid reach without modification.
But boosting optimises for engagement, which finds people who comment rather than people who buy. Run boosted posts first to prove the plumbing holds at volume, then move budget to Click-to-Messenger, which optimises for conversations started and skips the comment step entirely.
Sequence audiences retargeting first, warm second, cold last. Cold prospecting is the most expensive layer and the slowest to produce data to optimise against.
Paid reach on a free giveaway finds the UK comping community, who enter everything and buy nothing. This is judgement, not a sourced claim, but it is why the paid and organic tag split matters. Judge the cohorts on what they spent, not how many of them there were.
Re-run the same posts against the current audience. Same copy, same format, same 19:05 slot, same complimentary-comps link. Two of the original this-or-that prompts, run as written. The only thing that changes is the flow underneath, which has to be rebuilt.
Comment volume per post, opt-in rate per entrant, delivery rate and drop-off point, all against the equivalent numbers sitting in the old flow nodes. That is a real year-on-year read on audience health.
A ranking of formats. Two posts at dormant-Page volume, with prize appeal, timing and Meta's ranking changes all varying underneath, is a read rather than a verdict. Anything that looks dramatically better is a hypothesis to run three or four more times.
Screenshot every existing node's statistics, and pause every legacy trigger. With that many keyword-less copies in the account, the first new post will otherwise fire something unpredictable.
| Date | Change |
|---|---|
| 29 Dec 2025 | Marketing Messages on Messenger (beta). Meta's paid replacement for Recurring Notifications, which Meta began switching off from January 2026. Billed to the card on the linked ad account, and can reuse Messenger Lists already collected. |
| 6 Feb 2026 | Message Tags removed, deprecated by Meta. |
| 13 Mar 2026 | Utility Message Templates on Messenger (beta), replacing Message Tags. Essential, non-promotional messages only: confirmations and account alerts. |
The purple "Subscribe to our updates" widget in both existing flows is the Recurring Notifications opt-in, the feature being withdrawn.
Meta's own documentation says: "On February 10th, 2026, Recurring Notifications on the Messenger API will be discontinued for partners and businesses globally, except for the AU, EU, JP, KR, and UK geos." So a UK Page keeps it.
The paid replacement does not fill the gap here: Meta's supported country list for Marketing Messages on Messenger does not include the UK. That is consistent with there being no Marketing campaigns tab in the account.
Action: send one Recurring Notifications broadcast to about twenty recent contacts before building anything on it. Ten minutes, and it is the only thing that will tell you whether the list is still reachable.
| Date | Change |
|---|---|
| 22 Oct 2025 | AI hub (beta): Knowledge, AI Replies, AI Comments, AI Goals, overview dashboard |
| 24 Feb 2026 | Knowledge v2: import website pages, edit extracted text, refresh when the site changes |
| Mar 2026 | AI Playground: simulates DMs, detects knowledge gaps, shows which source fed each answer |
| 8 Apr 2026 | AI Behavior: set role, brand voice and guardrails on what it must never say |
ManyChat's help centre carries the same line across AI Knowledge, AI Behavior, AI Replies and AI Comments: "This feature is currently available only for the Instagram channel and requires an active paid plan with access to the Manychat AI feature." AI Playground is documented inside AI Replies, so it inherits the same limit.
The exception is the older AI Step, which "currently supports Facebook Messenger, Instagram, WhatsApp, TikTok, and Telegram". So an AI answer layer on Messenger is possible, but it is built on AI Step, not on Knowledge v2, and it does not read the website the way Knowledge v2 does.
| Date | Change |
|---|---|
| Nov 2025 and Feb 2026 | Inbox upgrades: labels, then the rename from Livechat, plus unread, tag and custom-field filters and snooze reminders |
| 7 Apr 2026 | Randomizer expanded to 12 paths, up from 6 |
| 7 to 8 Apr 2026 | Home tab redesign, smarter mobile filters |
One correction to carry: ManyChat's own blog suggests using the Randomizer to pick a giveaway winner. Do not. The Randomizer is a percentage traffic splitter built for A/B tests. It assigns each contact to a fixed group the first time it sees them, distributes in batches, and ManyChat acknowledges "discrepancies, especially with large numbers of contacts". It would not stand up as a verifiably random draw. The twelve paths are useful for varying content, not for deciding who wins.
ManyChat moved to Active Contact billing on 2 March 2026 across Free, Essential, Pro, Business and Advanced.
ManyChat's own wording: "By default, these plans are available to accounts created on or after March 2, 2026. We're gradually making the new pricing model available to accounts created before this date, with availability rolling out country by country ... Once the new pricing model becomes available for your account, we'll send you an email with details about the migration."
This account was created before that date, so the old plan probably still applies. There is no published migration date, and ManyChat gives no commitment beyond that email.
Check the billing screen before worrying about overage, and check whether the client has been paying for a dormant contact list for a year.
Meta announced that "all videos on Facebook will be shared as reels, eliminating the need to choose between uploading a video or reel", and renamed the Video tab to the Reels tab. It was announced as coming "in the coming months" rather than switched on that day.
Live is a different matter. Meta's post says only that Facebook "remains a home for all types of video, short, long, and Live". That Live stays its own category and can still surface in the Reels feed through recommendations is trade reporting, not a Meta statement, and Meta's own Reels ranking documentation does not mention Live at all.
Source: Meta Newsroom, "Making it Easier to Create Videos on Facebook", 17 June 2025. The Live detail: Social Media Today, same date.
There is no longer such a thing as a feed video on Facebook. Anything uploaded as video competes in the Reels pool and is scored like a reel: retention, watch time, completion.
Meta's own words: "Our recommendation engine is also now surfacing 50% more reels from creators published that day, so you see the newest content faster." The same announcement rolled out friend bubbles, "so you can spot which reels and posts your friends have liked", across both Reels and Feed. Meta's engineering write-up sets out the model: it estimates relationship strength between users and ranks content friends have interacted with.
Sources: Meta Newsroom, "Finding and sharing Reels on Facebook just got easier and more fun", 7 October 2025; Engineering at Meta, "Friend Bubbles", 18 March 2026. A separate 50% figure appears in the same announcement about reels over a minute and watch time: the two are not the same number.
Recency carries more weight than it did. Draw-night content should go up on the night, not the next morning. Inference from the freshness change, not something Meta stated about scheduled content.
The change with the most direct consequence for this Page.
Accounts continuing to post primarily unoriginal content may be deemed non-recommendable and demonetised, with an appeals route available. Meta also states views and time spent on original reels approximately doubled in the second half of 2025 against the second half of 2024, which is Meta's own figure and unverified.
Source: Meta Newsroom, "Rewarding Original Creators on Facebook", 13 March 2026.
The penalty is account-level, not post-level. For a Page whose non-follower reach depends on being recommendable, losing that status removes the discovery engine entirely.
Meta rebuilt the retrieval stage, narrowing millions of candidates to thousands in under 100ms, into a single unified neural network, allowing far more candidates to be evaluated without breaking the latency budget.
Source: Engineering at Meta, "SilverTorch: Index as Model", 26 May 2026, paper accepted at SIGIR 2026.
Meta frames this as efficiency and recommendation quality. The marketing read, that a much bigger pool now competes for each slot and new content surfaces faster, is third-party interpretation rather than a Meta statement. Treat as inference.
Still live in Meta's Content Distribution Guidelines. The current wording covers "posts that explicitly request engagement (such as votes, shares, comments, tags, likes or other reactions) for purposes other than a specific call to action". Content in the guidelines has its distribution reduced, which Meta calls a demotion.
Two details matter more than the headline:
Sources: Meta Transparency Center, Content Distribution Guidelines, Engagement Bait, and "Types of content we demote" (2 July 2025); Meta Transparency Center, "Repeatedly posting low quality content" (19 January 2022). The "more heavily" comparison traces to a Meta Newsroom post of 18 December 2017 and an archived Business Help Centre page last updated 12 February 2020.
The previous schedule was a daily comment-prompt post for eighteen consecutive days. That is the repeat pattern Meta names, and the penalty attaches to the account, not just the post.
The mechanic still works, but the cadence has to change: comment-CTA posts interleaved with filmed content, winner content and draw content, rather than run daily as the entire content strategy.
It is also a reason to split the reach metric two ways: per format, and by whether the post carried a comment CTA. That is the only way to separate demotion from ordinary format variance.
The whole system depends on someone doing something that gives you a route to their inbox. There are only three of those on Facebook, and the rest of what looks like engagement is worthless here.
| What they do | Can you message them? | What it costs them |
|---|---|---|
| Comment on a post | One private reply, within 7 days | Low effort, public, familiar |
| Comment during a Live | Same, and they are already watching | Low effort, high intent |
| Message the Page (keyword, m.me link, QR, ice breaker) | Yes, and it opens the 24-hour window immediately | Higher effort, far higher intent |
| Like or react | No | |
| Share | No, and rewarding it breaks Meta's promotion rules | |
| Follow the Page | No |
A private reply to a comment does not open the messaging window. It buys you one message. The window opens only when they reply or tap a button in it, and an "open website" button does not count: ManyChat's own wording is that those people "will not appear in your contact list".
So the post's job is not "get comments". It is get comments from people who will then tap. A post that harvests 400 comments from people who never tap is worth less than one that gets 80 who do.
"Comment CASH or CAR" is the natural way to build this, and it is the worst of both worlds. It fails on policy and it fails on plumbing.
| Problem | Detail |
|---|---|
| It is Meta's own definition of bait | Meta names five types and demotes them: react, comment, share, tag and vote baiting. Comment baiting is asking people to comment with specific predetermined answers: words, numbers, phrases or emojis. Vote baiting is asking people to vote by comment or reaction. A fixed-option prompt is both. |
| You cannot read what they typed | The comment text never reaches the flow. There is no variable, no custom field and no condition that can see it. So you cannot have one trigger that branches on the answer. |
| Every option needs its own trigger | Two options means two triggers and two duplicated flows, each hardcoding its own tag. Four options means four. |
| One automation per person per post | Someone who comments both words gets one outcome, not two. There is no second bite. |
| Matching is "contains", not exact | There is no exact-match option on the Facebook comments trigger. "I'd go cash all day" fires CASH. So would "cashmere". Every keyword needs misspellings added and an exclusion list built. |
Engagement bait is a ranking guideline: the penalty is reduced distribution. Meta's Community Standards on spam are enforcement, and they prohibit something narrower and closer to home:
"Offering giveaways (i.e., offering others a chance to win) of cash or cash equivalents in exchange for engagement. (e.g., 'Anyone that likes my page will be entered to win $500')."
A £100 cash freebie entered by commenting is a cash giveaway in exchange for engagement. The same policy also bars "like/share-gating", requiring engagement before someone can access something.
This does not make the freebies unrunnable, but it changes what they should look like: the entry should run through a route that is not "engage with this post to be entered", which is what the message-us and Live patterns below do. Worth putting to the client before the next freebie, because this is the policy behind account restrictions rather than a quiet loss of reach.
Ask an open question in public. Trigger on any comment. Put the fixed options as buttons inside the private reply.
The post never tells anyone which word to type, so it is not a fixed-answer prompt. The tag comes from a button, so it is exact rather than guessed from text. And the tap is what opens the 24-hour window, which is the thing you actually need.
The trade is that some people comment and never open the reply. That loss happens today too, silently, and at least this way it is measurable.
The default. Most posts should be this.
Post: "Honest answer, if it landed tomorrow: what's the first thing you'd do with it?" Trigger: any comment on that post. Private reply: "Go on then, which is closest?" with buttons.
Buttons give you a clean tag, the window, and an opt-in in one tap. Keep button labels short: 20 characters, three regular buttons or up to eleven quick replies.
There is a ranking argument for the open question too. Meta's own feed documentation lists among its modelled signals "the number of comments in the post", "average length of comments on a post" and "a measure of how relevant the comments on a post are to the topic of the post". A wall of one-word answers scores worse on two of those three than a thread of real replies.
The strongest post they can run
"Next month's headline prize: camper or truck? Whichever wins is the one we run." Meta's bait policy exempts genuine requests for help, advice or recommendations, and a vote that actually changes what happens is the closest thing to that in this business.
It also produces the single most commercially useful tag there is: someone telling you what they would buy a ticket for. Only run it when the vote really does decide, because the first time it does not, it becomes a bait post with extra steps.
Fewest people, best people
An inbound message opens the window immediately, with no private-reply step and no tap needed. It also sits outside the comment-baiting wording, which is about comments and reactions rather than messaging a Page. That reading is ours, not a Meta statement.
Far fewer people will do it. The ones who do are worth several of the ones who will not.
Keep it, but as the minority
It still produces the most volume, and volume has its place. Cap it at two or three a week, never two days running, and make the keyword a real answer ("CASH") rather than a code ("BLUE"), because a real answer is both less bait-shaped and useful as a tag.
Build rules: one trigger per keyword, set to that specific post only, first-level comments only, misspellings added as extra keywords, an exclusion list for words that appear inside ordinary comments, and a catch-all any-comment trigger created after the keyword triggers so it does not swallow them.
Highest intent, and nothing has to be asked for
People comment through a live draw anyway. The standard comments trigger works on a Live post, so the trigger has to exist before you go live, because it cannot pick up comments made before it was switched on.
Meta's own live guidance: stream for at least 20 to 30 minutes, schedule it in advance and promote it in separate posts beforehand, reintroduce yourself as new people join, welcome commenters by name, pin a comment to seed the thread, and close with one explicit ask.
The account holds a dozen duplicated triggers. A keyword-less trigger created before a keyword one fires regardless of keyword, which is exactly how the old account behaved. Screenshot the node statistics, then pause them all.
The automation will not publish otherwise. That first message can hold one content block with buttons, nothing more.
Two separate steps and the second one is easy to miss.
Each branch hardcodes its own tag. Name them so they group and age out: pref_prize_cash_202609. Only collect a dimension you would actually use to decide who gets which message.
It is the only way to message someone after the 24 hours, it needs their explicit tap, and it cannot be added retrospectively. Miss it and you cannot tell them about the draw they entered.
Meta caps automated replies, ManyChat throttles when a post spikes, and 2026 reports describe large comment volumes getting a fraction of DMs delivered. The private reply window is 7 days, not 24 hours, so a post can be swept again later in the week for people who were missed.
UK promotion rules are stricter than most social posts assume. CAP's own guidance: "For online promotions, such as banner ads or social media posts, significant conditions must be in the ad itself." A Facebook caption is rarely accepted as too space-limited to carry them.
| Line | Content |
|---|---|
| 1 | The hook or the question, in their own words |
| 2 | The prize line, on freebie posts only |
| 3 | The ask, exactly as it must be done |
| 4 | The significant conditions: closing date and time, how to enter, the free entry route, 18+, UK only, the number and nature of the prize, and the promoter's full name and correspondence address |
| 5 | Link to the full terms |
Two specifics worth holding onto. The free entry route must be "promoted at the same level as the paid for route", not buried in small print. And announcing a winner once, in a post or a comment, is "unlikely to be sufficient" on its own.
| Metric | Why |
|---|---|
| Opt-ins per 100 comments | The primary number. Comment volume on its own tells you nothing |
| Button tap rate on the private reply | Whether those comments came from people willing to open a conversation |
| Unique commenters, first-time against returning | The same core comments on everything. Without the split, later posts look artificially weak |
| Delivery rate | The early warning on window, throttling and policy problems |
| Reach, split by format and by whether the post carried a comment CTA | The only way to separate a demotion from ordinary format variance |
| Hides, spam reports, blocks and unsubscribes after the reply | The signals that precede a Page-level problem |
Nothing here carries a benchmark, deliberately. There is no trustworthy baseline for this Page yet, and the published engagement tables are cross-industry and mostly Instagram. Weeks one to four set the baseline.
Tag a friend and share to enter: banned by Meta's promotion rules, which say a promotion "must not require or incentivise participants to share, repost, tag others or in any other way publicise your promotion".
Like to enter: a like gives no route to message anyone, so it produces nothing this system can use.
The same prompt every day: the repeat pattern that carries account-level risk.
Emoji as the keyword: Meta names emoji answers inside comment baiting, and they are the least reliable thing to match.
Vertical clips and live draws. Live is reported to sit outside the Reels classification while still being pullable into the Reels feed by recommendations, though Meta does not say so itself (Part six). Either way it doubles as winner evidence, which is the reason to do it.
The actual prize in the actual unit, winner handovers, draw-night screenshots. Since June 2025 a stills-and-text slideshow is a reel whether intended or not, scored on retention and watch time against the whole Reels pool, and since March 2026 on originality too. Stitched supplier or stock prize photos with captions over the top sits close to the pattern Meta describes as low-value editing, and the exposure is account-level. My read, not Meta's. A bad trade for a format that is cheap to produce.
Prize reveals, how it works. Still a standard feed post, no Reels treatment, no added originality exposure. They will not generate discovery reach: carousels do not enter the recommendation pool the way reels do.
Recency carries more weight since the October 2025 refresh.
Meta publishes that as an ads spec; there is no separate organic spec page, and the same range is what organic multi-image posts use.
Caption truncation. The "about 125 characters before see more" figure is reverse-engineered by agencies, not stated by Meta. Write so the hook, the ask and the closing time sit in the first two lines and the exact cut-off stops mattering.
Reel safe zones. Blogs cite three different sets of pixel figures for the same spec. Meta's own page is behind a login wall, so nothing here quotes a number. Keep text well inside the frame and check it in preview.
Meta publishes nothing supporting it, in either direction. Put the link where it serves the reader, and do not build the plan around a belief nobody can source.
The published engagement-rate tables (Socialinsider, Metricool, Later) are cross-industry, predominantly Instagram, and not applicable to a UK comps Page. Reach per format comes from the client's own Page Insights over a four-week window.
The opportunity. ManyChat launched Marketing Messages for WhatsApp on 9 April 2026: promotional messages outside the 24-hour window using approved templates. That is the capability Messenger is losing.
For a competition operator the send pattern is fixed and predictable: draw closing tonight, tickets running low, live draw starting, you have won. Those are exactly the sends currently going out by SMS, and WhatsApp template messages should be materially cheaper per send than UK SMS.
WhatsApp template pricing varies by category and changes, so the saving has to be calculated against the client's actual SMS invoice and their actual monthly send volume. Ask for both. It is the single strongest commercial argument available in this engagement and it should be made with their numbers, not estimates.
WhatsApp needs a verified business number, template approval, and opt-ins collected under WhatsApp's own consent rules. Messenger opt-ins do not transfer. Weeks 1 to 4 build the audience and prove the flows; weeks 5 and 6 are when there is something worth migrating and evidence to justify the setup work.
Get the consent wording right now, covering WhatsApp specifically, on every opt-in the flows collect. Retrofitting consent later is the thing that kills these migrations.
Messenger is the acquisition channel: it is where the comment triggers and the boosted posts land people. WhatsApp is the retention channel. With no email capture in the flows and Recurring Notifications being withdrawn, WhatsApp is the only owned channel this plan builds towards, which raises the stakes on weeks 5 and 6 rather than lowering them.