Multi-channel outbound without getting banned (LinkedIn, WhatsApp, SMS)

Getting banned on an outbound channel is usually a detection problem rather than a rules problem. LinkedIn, Meta and the US carriers each adjudicate your sending independently, and only one of the four systems you touch answers in plain language. LinkedIn restricts an account quietly. Meta pauses a template for three hours. A carrier drops a message and hands your provider a code that may never reach your screen. Email alone replies with a readable SMTP error. So the architecture that survives is not a longer rules list — it is per-channel instrumentation plus one suppression key every channel reads.

Four adjudicators, and only one of them writes back

Start by naming who is actually deciding, because teams routinely aim a fix at the wrong party.

LinkedIn adjudicates the account. Its User Agreement, at section 8.2, tells members not to "use bots or other unauthorized automated methods to access the Services, add or download contacts, send or redirect messages, create, comment on, like, share, or re-share posts". Enforcement is contractual, not statutory, and it lands on the profile rather than on a message. LinkedIn's help page on the types of invitation restriction names three triggers in its own words: "You've sent many invitations within a short amount of time", "Many of your invitations have been ignored, left pending, or marked as spam by the recipients", and "If you send an excessive number of invitations and we suspect the use of an automation tool, we may suspend or restrict your account."

Meta adjudicates at three levels at once. The phone number carries a quality rating and a messaging tier. Each template carries its own quality rating. And the recipient carries a per-user ceiling on how many marketing templates they will receive from anybody. A message can be refused by any of the three while the other two look healthy.

US SMS has two adjudicators in series. The Campaign Registry verifies the brand and the campaign before a message is allowed onto the 10DLC channel at all, and then the mobile carrier filters at send time against content, volume and complaint signals. The Federal Communications Commission sits above both on consent and revocation.

Email is adjudicated by the receiving mailbox provider, and it is the one that tells you. Google's bulk sender guidance asks senders to "Keep spam rates reported in Postmaster Tools below 0.30%". Microsoft began rejecting non-compliant high-volume mail on 5 May 2025 with 550; 5.7.515 Access denied, sending domain [SendingDomain] does not meet the required authentication level. You get a string you can grep.

That asymmetry is the whole problem. Three of your four channels fail silently, so the reflex of "we will notice if something breaks" is only true on one of them.

What each system measures about you, in its own units

Volume is almost never the metric, which is why volume-based advice keeps failing.

• LinkedIn measures acceptance and cadence. Its own restriction triggers are about invitations left pending or marked as spam, and about invitations sent in a short amount of time. An account sending forty invites a week at a 12% acceptance rate is in more danger than one sending eighty at 45%.
• Meta measures read rate and negative feedback, per template. Meta's documentation says a template is paused "due to recurring negative feedback from customers, or low read-rates" — which means a template that nobody blocks but also nobody opens is on the same path as one people report.
• US carriers measure registration status first, then content shape. An unregistered or mismatched campaign is a routing problem before it is a content problem, and throughput is assigned by registration: The Campaign Registry describes AT&T as providing throughput per campaign with its own message class, and T-Mobile as allocating a daily allowance per brand, tied to the brand's EIN and shared across every campaign under it.
• Mailbox providers measure authentication, complaint rate and engagement. Authentication is binary and cheap to fix; complaint rate is the one that moves slowly in both directions.

One shared property is worth stating plainly: every one of these systems scores the sending identity, not the campaign. Burn a LinkedIn profile, a WhatsApp number, a 10DLC brand or a sending domain, and the damage travels with that identity into every campaign that uses it.

LinkedIn's refusal is silent, so you have to build the alarm yourself

LinkedIn will not tell you that you are approaching a restriction. It tells you afterwards, and the help page's own guidance on how long it lasts is vague by design: a few hours for a first restriction, several days where restrictions repeat, and up to a month where an account is holding an excessive number of pending invitations.

Because the ceiling is unpublished and adaptive, the only defensible approach is to pace against the signals LinkedIn says it watches. Autocloz's LinkedIn pacing does this with five gates, and the defaults are worth reading as a description of the mechanism rather than as magic numbers:

• A warmup ramp that starts a newly connected sender at a floor of five actions on its first day and grows on a smoothstep curve to the configured daily target over 21 days by default.
• A weekend damper that cuts the day's ceiling to 30% of the weekday allowance, evaluated in the sender account's own timezone.
• A trailing seven-day invite cap, defaulting to 80 and measured as a rolling window rather than Monday-to-Sunday, so a Sunday-night burst and a Monday-morning burst cannot both be full.
• Burst refusal, which defers the next action for twenty minutes when five actions have already landed inside ten minutes.
• An accept-rate throttle that halves the effective daily ceiling for a dispatch when the trailing accept rate sits below 30%, and does not throttle at all when the sample is under ten invites, because a small sample cannot justify slowing a healthy sender.

There is also a hard stop at 1,000 pending invitations per sender. That number is a cushion rather than a limit LinkedIn publishes, and it exists because the pending pile is one of the triggers LinkedIn names.

The alarm you build on top should be three trends, checked weekly per sender: acceptance rate, pending-invite count, and median time from invite to acceptance. All three degrade before a restriction, and none of them is visible in a send counter. The invite, message and soft-ban limits people actually observe on LinkedIn are worth reading as observation, clearly labelled as such.

WhatsApp tells you in numbers, and those numbers have names

Meta is the opposite case. Almost everything is published, coded and queryable, and most teams simply never look.

Messaging tiers cap how many unique WhatsApp users you may message in a rolling 24 hours. Meta's messaging-limits documentation lists 250, 2,000, 10,000, 100,000 and unlimited, with new numbers starting at 250 and moving up either by completing a scaling path — business verification, or 2,000 delivered messages outside customer service windows to unique numbers within a 30-day moving period using high-quality templates — or automatically, when you maintain quality and use at least half your current limit within seven days.

Template quality is separate and moves faster. Meta's template-pausing documentation sets the ladder explicitly: a template that reaches the lowest quality band is paused for 3 hours on the first instance, 6 hours on the second, and disabled on the third. That ladder is the single most useful operational fact about WhatsApp marketing, because it means a bad template costs you the template, not the number — provided you notice at instance one.

Per-user limits are the layer nobody plans for. Meta states that WhatsApp "may limit the number of marketing template messages a WhatsApp user receives from any business in a given period of time when they are less likely to be receptive", publishes no number for it, and returns error code 131049 when it bites. And the fact that reshapes most 2026 plans: Meta's documentation states that WhatsApp does not currently deliver marketing template messages to WhatsApp users with United States phone numbers. Utility and authentication templates are a different category and are not covered by that sentence.

Autocloz refuses to dispatch a WhatsApp step whose linked template is not in approved status, and refuses free-form text unless the 24-hour customer service window is genuinely open — resolved from the conversation's stored window expiry, or from a reply recorded against that handle in the last 24 hours. The WhatsApp Business API architecture, tiers and error codes go deeper on what each refusal means on the wire.

SMS is the only channel where somebody vouches for you before you send

10DLC — application-to-person messaging over an ordinary ten-digit US number — inverts the usual order. On every other channel you send first and get judged afterwards. Here the judging happens at registration.

The Campaign Registry states that a Verified brand status "is a requirement to register campaigns and send messages on 10DLC", and that an EIN or tax ID is required when a brand is added so that the brand's identity can be confirmed. Optional standard vetting returns a 0-100 score that determines throughput levels for some carriers. India runs the same idea through a different mechanism: the Telecom Commercial Communications Customer Preference Regulations, 2018, issued by TRAI on 19 July 2018 and in force from 28 February 2019, mandate distributed ledger technology for registration and define a header as "an alphanumeric string of maximum eleven characters or numbers assigned to an individual, business or legal entity".

Consent is the other half, and it is stricter than email's. CTIA's Messaging Principles and Best Practices says a consumer opt-in "should not be transferable or assignable" and "should apply only to the campaign(s) and specific Message Sender for which it was intended or obtained". That single sentence invalidates every purchased SMS list, whatever the seller's paperwork claims. The full picture of what TCPA, 10DLC and carrier policy each demand is the reference to work from before a first send.

The one architectural decision that protects all four channels

If you take a single structural decision from this, take this one: suppression must be shared, and it must be evaluated at dispatch, not at list build.

A contact who opts out of one channel and receives a touch on another three days later is the single most reliable way to convert a mild irritation into a complaint, and a complaint is the input every one of the four adjudicators weighs most heavily.

Autocloz implements this by writing an opt-out against the wildcard channel. When an inbound SMS or WhatsApp message matches a carrier-recognised stop keyword — the set is STOP, STOPALL, UNSUBSCRIBE, CANCEL, QUIT, END, OPTOUT and OPT-OUT, matched as the whole message or as the first or last token so that "Please stop" and "STOP." both count — the workspace records a suppression row on channel *. Every subsequent dispatch on every channel checks that row, matching both the literal handle and its E.164 form, so a number stored as +1 (415) 555-1212 on one list and 4155551212 on another resolves to the same suppression. Email additionally matches an @domain row, so a single company-wide opt-out blocks every address on that domain.

The honest boundary, and it matters: the suppression key is the identifier, not the person. A STOP from a phone number stops SMS, WhatsApp and calls to that number. It does not stop email to the same human, because nothing in the system asserts that the two identifiers belong to one person. Linking them is a policy decision you make deliberately, and in most jurisdictions it is the safer one.

Pacing that survives contact with a real account

Three pacing rules do most of the work, and they are boring on purpose.

Cap per sending identity, not per campaign. Six campaigns sharing one LinkedIn profile is one profile's risk. Autocloz's per-channel daily allowances are properties of the plan and the sender — on the free plan that is 1,000 emails, 250 SMS, 250 WhatsApp, 100 calls and 50 LinkedIn actions a day — and the LinkedIn ramp then applies on top of the plan number, not instead of it.

Randomise the gap. A perfectly periodic send histogram is the easiest bot signature there is. Autocloz adds a random 60 to 300 seconds on top of each LinkedIn sender's configured minimum gap after every action, so the interval distribution never collapses to a constant.

Stop the sequence when a human answers, everywhere. A reply recorded on any channel in the trailing fourteen days skips the LinkedIn step for that lead. This is the cheapest complaint reduction available and it costs you nothing you wanted.

Autocloz's free plan carries all five channels, the compliance gate and the shared suppression list on five users and ten mailboxes, which is enough to watch every refusal code above on your own senders — start free and instrument one channel properly before you add the second.

How to find out in a day rather than a quarter

Build the dashboard around leading indicators, one per channel, and look at it weekly.

• LinkedIn: trailing seven-day acceptance rate per sender, plus pending-invite count. A falling acceptance rate is the restriction forming.
• WhatsApp: template quality rating per template, checked before you scale a send, plus the number's messaging tier and its recent movement. Error 131049 in your logs means per-user limits, not a bad number.
• SMS: the ratio of provider-accepted to carrier-delivered, split by carrier. A gap that opens on one carrier and not the others is a registration or content-classification problem, not a list problem.
• Email: Google Postmaster Tools spam rate against the 0.30% line, plus authentication pass rate. Both are published thresholds rather than folklore.
• All four: replies per thousand touches, tracked per sending identity. It is the only number that degrades on every channel when something is wrong, and the only one your buyers actually generate.

Two of those need a place to live. Running five channels from five tools is how the acceptance-rate trend ends up in nobody's report, which is the argument for putting every channel's touches and refusals on one timeline and for keeping the suppression list and the gate itself in one place. If you are weighing a LinkedIn-only tool for this, the honest trade-offs are set out in the comparison with HeyReach.

What this cannot prevent, and what Autocloz does not do

Pacing lowers risk. It does not remove it, and anyone selling it as removal is selling you something that does not exist.

LinkedIn's User Agreement prohibits automated access outright, so LinkedIn automation is a contractual exposure rather than a technical one. Every tool in this category — including this one — operates in that grey zone, and the honest position is that good pacing makes a restriction less likely and never impossible. LinkedIn can change its detection at any time without notice, and it does.

Autocloz does not verify your 10DLC or DLT registration before dispatching an SMS. It surfaces the requirement as guidance in the channel-health checks and in Autopilot's pre-flight, but the send path itself will hand an unregistered message to your provider and let the carrier make the decision. Treat registration as a prerequisite you confirm yourself, not as something the software blocks on.

Opt-in enforcement is off by default and switched on per workspace, and today only the WhatsApp dispatch path reads it. The SMS path checks the do-not-contact list but not the consent ledger, so a workspace that turns on opt-in enforcement expecting it to cover texts is expecting something the code does not do. Consent evidence still gets recorded for both channels; the gate is what differs.

Quiet hours on SMS, WhatsApp, calling and LinkedIn are evaluated in the campaign's configured timezone if one is set, otherwise in the sending account's timezone — never in the recipient's. Only the email path consults a lead's own timezone, and only where the campaign carries a sending window. If you are dialling or texting across time zones, set a per-campaign timezone deliberately; the default will otherwise compute a legally significant window from the wrong clock.

And LinkedIn's stale-invite withdrawal marks your ledger so the accept-rate and pending-count arithmetic stays honest. It does not rescind the invitation on LinkedIn itself — the provider-side withdrawal is not wired up. If your pending pile is the problem, withdrawing inside LinkedIn is still a manual job. Nothing here is legal advice either: the gate enforces the rules you configure, in the jurisdictions you tell it about, and knowing which rules apply to a given recipient stays yours.