Sales cadence best practices (timing, channels, and touches)

A sales cadence is three numbers — how many touches, across how many days, over how many channels — and the difficult part is that you do not get to choose all three. Every channel you add carries its own ceiling: a per-mailbox daily cap, a rolling weekly invite budget, a statutory calling window, a carrier registration. The cadence you can actually run is the intersection of those ceilings. This is how to compute that intersection gap by gap, and how to read which ceiling stopped you when it stalls.

The three numbers that define a cadence, and the fourth nobody counts

A cadence is the schedule; a sequence is the machinery that executes it. The schedule is described by touch count, span in days, and channel count. Most advice stops there.

The fourth number is contacts per account, and it changes everything about how the first three behave. Five contacts at one company, each enrolled in an eight-touch email cadence, is forty messages from your domain into one recipient gateway over three weeks. That gateway does not see five independent conversations. It sees one sender with a burst pattern, and it is the entity that decides whether your mail lands. Two contacts per account at eight touches has the same account coverage and half the concentration.

So write the cadence down as four numbers before you build anything: touches, days, channels, contacts per account. If you cannot state all four, you have a template rather than a plan. The mechanics underneath — states, step kinds, the worker tick — are covered in detail in what a sales sequence actually is; this post is about the schedule you put on top of it.

The five ceilings a multichannel cadence runs under

Each channel has a different authority setting its limit, and each uses a different clock. Confusing them is the most common reason a cadence that looks fine on paper delivers half its touches.

Email — the receiving provider sets it, and it is measured in reputation, not volume. Google's bulk sender requirements, in force since 1 February 2024, apply to senders of more than 5,000 messages a day to Gmail accounts and require SPF, DKIM and a DMARC record on the From domain, alignment of that domain with either SPF or DKIM, one-click unsubscribe conforming to RFC 8058 on marketing mail, and spam rates in Postmaster Tools kept below 0.3%. Microsoft began applying an equivalent authentication bar to senders of more than 5,000 messages a day to Outlook.com, Hotmail.com and Live.com on 5 May 2025, routing non-compliant mail to Junk. Neither rule is a send-rate cap. Both are conditions on whether your sends count for anything.

LinkedIn — the platform sets it, and the window rolls. LinkedIn caps connection invitations per account and does not publish the number; the practical ceiling is widely reported at roughly a hundred a week, applied over a rolling seven days rather than resetting on Monday. A cadence that fires forty invites on Monday and forty on Thursday has spent its week by Thursday afternoon.

Calling — a statute sets it, and it runs on the recipient's clock. The FCC rule at 47 CFR 64.1200(c)(1) restricts telephone solicitations to residential subscribers to 8 a.m. through 9 p.m. at the called party's local time. Local to them. A single national calling window is wrong at both ends of a wide country.

SMS — carriers set it, through registration. In the United States, application-to-person traffic on ten-digit long codes requires brand and campaign registration with The Campaign Registry, and carriers apply throughput classes based on that registration. The CTIA Messaging Principles and Best Practices is the industry document the registration process implements. An unregistered campaign does not send slowly; it does not send.

WhatsApp — Meta sets it, through templates and windows. Business-initiated messages must use a template Meta has approved in advance, and free-form replies are only possible inside the customer service window that a customer's own message opens. The WhatsApp Business API guide covers that mechanism properly.

Five authorities, five clocks: reputation windows for email, a rolling seven days for LinkedIn, the recipient's local hours for calls, a registration class for SMS, a customer-initiated window for WhatsApp. Nothing synchronises them for you.

How to choose the gap between two touches

There is one decision rule worth internalising, and it is about the shape of the gaps rather than their size: widen them as the cadence runs.

The reasoning is a hazard argument. Take everyone who has not replied by day n. The probability that this group replies on day n plus one falls as n grows, because the people most likely to respond have already responded and what remains is a progressively less responsive residue. Equal gaps therefore allocate the same effort to the tail as to the opening, and the tail returns less per touch. A geometric shape — 2, 3, 4, 6, 9 days — concentrates effort where the yield is.

Three constraints override that shape:

• The channel's own memory. A LinkedIn invite and a LinkedIn DM two days apart read as one person being persistent. The same two touches nine days apart read as two unrelated approaches. Keep same-channel touches inside the window where the earlier one is still remembered.
• Reply-detection latency. Your next touch must not fire between a prospect replying and your system noticing. Autocloz's autopilot rulebook ships a reply cooldown of 24 hours for exactly this reason, and the same idea applies whatever tool you use: the minimum gap is bounded below by how fast you detect a reply.
• The bounce feedback loop. Your second email should not go out before you know whether the first bounced. Bounce classification is not instant, and a cadence that fires step two an hour after step one is enrolling addresses you already have evidence against.

Cap the whole thing at a span you will actually maintain. The count question is treated on its own terms in how many follow-ups you should send, which derives the number from your complaint budget rather than from convention.

The channel order the ceilings force on you

Most cadence templates open with the LinkedIn connection request because it feels warmer. That is backwards, and the reason is scarcity.

Rank your channels by how expensive one unit of the resource is. A LinkedIn invite from a single account is the scarcest thing you own — on the order of a hundred a week, non-transferable, and spending one on a bad-fit prospect costs you a good-fit prospect later that week. An email from a pool of three mailboxes is the least scarce. Spend the abundant resource first to qualify, and the scarce one second on what survived.

That produces this order, and each step earns its place:

1. Email first. It is your highest-throughput channel and it doubles as a validity test. An address that hard-bounces has told you the record is wrong before you spend anything else on it.
2. Suppress the bounces. A hard bounce should remove that contact from every downstream channel in the cadence, not just from email. A wrong email address is evidence about the record, not about the channel.
3. A passive LinkedIn touch before an active one. A profile view or a post like produces a notification without consuming invite budget. Autocloz models these as their own step kinds — linkedin_view and linkedin_like — precisely so they can precede an invite.
4. The invite, on the shortlist only. Spend the weekly budget on prospects that opened, clicked or otherwise showed a signal. Single-channel LinkedIn tools model this pacing differently because it is their only resource; the side-by-side against HeyReach sets out where the two approaches diverge.
5. The call, inside the recipient's legal window. The call is where a human's time is spent, so it goes last among the cold touches and only on accounts with a reason behind them.
6. SMS and WhatsApp only where a consent basis exists. These are not cold channels in most jurisdictions. Treat them as follow-up to a conversation that has already started.

Autocloz's free plan covers 5 users and 10 mailboxes with mailbox warmup and SPF, DKIM and DMARC monitoring included, which is enough to run the email and LinkedIn halves of this properly — start free and build the schedule before you scale the volume.

A worked cadence, with the arithmetic behind every gap

Numbers below are illustrative, computed from Autocloz's shipped defaults rather than measured from any customer. Substitute your own.

Setup: 600 leads, two contacts per account, three connected mailboxes, one LinkedIn account, one rep making calls.

Email capacity. A new mailbox in Autocloz ships with a daily limit of 40, an hourly limit of 10, and a minimum 90 seconds between sends. Three mailboxes give 120 sends a day. Step one across 600 leads therefore takes five working days to complete, not one. That single fact reshapes everything downstream: your day-2 follow-up is day 2 relative to each lead's own step one, spread across a five-day enrolment window, not a date on your calendar.

LinkedIn capacity. Autocloz's autopilot rulebook defaults to 20 invites a day and 80 over a trailing seven days — deliberately under the platform's own ceiling to leave room for the account holder's organic activity. At 80 a week, inviting all 600 leads would take seven and a half weeks. You are not going to do that. You will invite the roughly 15% that engaged with email, which is about 90 people, which is a week and a bit of budget. The arithmetic is what tells you the invite has to be a shortlist step rather than a step-one step.

Call capacity. One rep making 50 real dials a day covers 600 leads in twelve working days, and only inside 8 a.m. to 9 p.m. at each recipient's location.

The cadence that falls out:

• Day 0 — email one. Completes across the list over five days at 120 a day.
• Day 2 — LinkedIn profile view. Costs no invite budget.
• Day 3 — email two, on a different thread, with a different angle.
• Day 5 — LinkedIn invite, but only for leads that opened or clicked. Roughly 90 people; the rest skip this step.
• Day 7 — call, for connected profiles and email engagers, inside the recipient's window.
• Day 11 — email three, the value-add touch.
• Day 16 — the breakup email, then exit.

Seven steps, sixteen days, three channels, two contacts per account. Total email volume: three sends times 600 leads equals 1,800 messages over sixteen days, which is 112 a day against a 120 ceiling. It fits, with almost nothing to spare — which is the point of doing the arithmetic. A fourth email step would not fit, and would silently queue rather than send.

What Autocloz enforces at dispatch, and what it only stores

Worth knowing precisely, because the difference between an enforced setting and a stored one is the difference between a cadence and a wish.

Enforced. Each channel account carries daily_limit, hourly_limit and min_gap_sec, and the effective daily cap is the minimum of the account's own limit, the provider's tier ceiling where one exists, and the workspace channel default — every one of the three a ceiling, never a floor. Non-email channel accounts default to 50 a day, 10 an hour and a 60-second gap; email mailboxes to 40, 10 and 90 seconds. LinkedIn additionally runs its own pacing layer: a warmup ramp with a day-one floor of 5 actions over a default 21-day horizon, weekend allowance cut to 30% of the weekday figure, a burst check that refuses when the last 5 actions all landed inside 10 minutes, and a random 60 to 300 second jitter on top of the configured gap. The daily ceiling is then scaled down when the trailing seven-day acceptance rate falls below a 30% threshold, with a floor at 5 actions.

Enforced, but on a clock you may not expect. The quiet-hours gate resolves its timezone in this order: the campaign's per-channel override, then the sending account's own timezone, then UTC. Not the recipient's. A fresh workspace ships SMS and call quiet hours at 21:00 to 08:00 with the timezone set to UTC, which for a team operating from India means the gate is closed from 02:30 to 13:30 local — the opposite of the intent. Set the timezone on the channel account, or set a per-campaign override, before you trust the default. The campaign *sending window* is a separate control that does read Lead.timezone first, then the campaign window's timezone, then the workspace timezone, then UTC — but nothing in the product infers a lead's timezone from a phone number or a company address. That field is populated only if you populate it.

Stored but not read. The LinkedIn weekly_cap value under the workspace channel defaults has no reader anywhere in the codebase. The weekly invite ceiling that actually fires comes from the autopilot rulebook's linkedin_invites_per_week, default 80. The dead field is documented as a known bug in the source rather than papered over with a toggle, and it is the kind of thing worth checking in any tool before you rely on a number in a settings page.

Diagnosing a cadence that has stopped part-way

Cadences rarely fail loudly. They defer. Read the reason rather than guessing at it.

• "Nothing sent today at all." Check whether the campaign's selected senders are active. Autocloz refuses to fall back to mailboxes you did not select, and stamps the enrolment's last_skip_reason with a message saying exactly that on every pass.
• "It sends, then stops around the same time each day." That is a daily cap, and the effective one is the minimum of three layers. Raising the account limit does nothing if the workspace default is lower.
• "LinkedIn steps defer with no obvious reason." The pacing layer returns a defer with a reason in the touch's error message. Burst detection and the accept-rate throttle both look like an unexplained pause from the campaign view.
• "Sends stopped and the campaign says bounces." Autocloz's bounce gate resolves per campaign, then workspace default, then the autopilot rulebook. Manual campaigns with no override sit behind a workspace safety floor of 15% over the trailing 24 hours, and the autopilot rulebook's own default is 3%. Those are different numbers for different campaign types, so check which one applies before concluding your list is fine.
• "Some leads never reach step two." An out-of-office reply deliberately keeps the enrolment running; a human reply and a hard bounce stop it. If a lead is stopped, the enrolment carries the reason.

For the deliverability half of a stall — the part where messages send but do not arrive — testing deliverability properly is the diagnostic, and the SPF, DKIM and DMARC generator covers the record side.

What a cadence cannot fix, and what Autocloz does not do

A cadence is a delivery schedule. It has no view on whether the message is worth delivering. If the list is wrong or the offer is wrong, a better schedule increases the rate at which the wrong message reaches the wrong people. Build the ideal customer profile first.

Autocloz's auto-stop is per enrolment, which means per lead. A reply from one contact at an account does not stop the sequences running against their four colleagues. If you are multi-threading an account, you need an operator to intervene, or you need to enrol one contact per account and expand manually.

Quiet hours are not evaluated in the recipient's timezone, as described above. For a call cadence under 47 CFR 64.1200(c)(1) that distinction is a compliance question rather than a preference, and the TCPA guide for cold calling treats it in full.

The product does not tell you your cadence is too long. There is no gate on touch count, and nothing warns that step seven is producing complaints rather than replies. That judgement stays with you, and the honest way to make it is to watch complaint and unsubscribe rates by step number rather than in aggregate.

And nothing about a cadence changes what happens after a reply. The schedule ends where the conversation starts, and the conversation is where deals are actually won or lost — handling the objections that arrive in the reply is a separate skill the cadence exists to create opportunities for. For the sequence-building surface itself, the multi-channel sequence builder is where the schedule above gets expressed as steps.