The omni-channel AI outbound machine

Cold email on its own is running out of runway. Reply rates are in the low single digits even when the copy is sharp, and the reason is that the inbox is now everyone's arena and nobody's living room. A better template does not fix that.

The way past it is not more email. It is the same message, arriving in three places, each written specifically for the person on the other end. That is what an outbound machine actually looks like, and there is less AI magic involved than the pitch decks suggest.

The single-channel treadmill

A dozen years ago, a cold email with a reasonable subject line got a reply most days. That is no longer true. Filters have gotten better. Inboxes are louder. The prospect has already received two cold emails this morning from people who sound exactly like you.

The usual answer to that is volume. Send more. Send to more people. A/B test the subject line until one of them breaks through. It works for about three weeks, and then the inbox providers figure out what you are doing and your domain reputation takes a hit. The treadmill speeds up and you fall off.

The other common answer is a "better template": one that sounds more human, more specific, more like it was written by somebody who actually cared. That also works for about three weeks, until the template has been copied by every agency in the category and is now its own kind of generic. The problem with templates is that they are templates.

Why three channels compound

The thing that actually cuts through is touching the same person in three different places inside the same fortnight. The message is coordinated; the channel is not the same.

A realistic rhythm looks like this:

• Monday: a short email that names something specific about the prospect's company.

• Wednesday: a LinkedIn connection request with a one-line note that picks up the same thread.

• Friday: a WhatsApp message, where available and appropriate, that lands on the phone rather than the screen.

The first touch is in the inbox, which is where they expect to be sold to. They ignore it. The second touch is on LinkedIn, where they half-recognise the name from Monday. The third touch is on WhatsApp, where the name is now familiar enough that they read it. The individual channels have not changed their reply rates. The combined probability has.

This is how people already sell to each other in real life. The warm lead you eventually meet for coffee usually bumped into you twice before the third attempt stuck. Outbound is not weird for needing the same shape.

The compounding bit matters because it changes the calculation on each individual channel. Nobody replies to the LinkedIn note on its own. Plenty of people reply to the LinkedIn note when the email three days ago already put the name in their head. You are not paying for each channel independently; you are paying for the combined weight of them.

Personalisation that isn't lipstick

Calling a prospect by their first name is not personalisation. Most mail-merge systems figured that out by the late 2000s. Real personalisation is writing a message that could not have been sent to the person in the next row of the spreadsheet, because the context would be wrong.

The bit the AI is actually doing is exactly that: reading enough of the prospect's public context to write a message that is only sendable to them. Not a template with their company name stitched in.

What gets read, in practice:

• The role, and what that role typically cares about this quarter.

• Recent public signals: a new hire, a funding round, a product launch, a podcast they went on.

• What the company sells, and to whom, so the pitch is in their language rather than yours.

• The channel itself, so the LinkedIn note sounds native to LinkedIn and the email sounds like an email.

Claude is good at this because the task is a reading-and-rewriting task, which is what large language models are for. It is not good because it is AI. It is good because writing one tailored message takes seven minutes of human time and about two seconds of model time.

Real personalisation is writing a message that could not have been sent to the person in the next row of the spreadsheet.

What the operator actually needs to see

Every outbound tool ships with a dashboard, and most of them are badly designed. They show you vanity metrics (opens, hovers, views) which are affected by mail clients and spam filters more than by how well the campaign is going.

The numbers worth putting on an operator's screen are fewer than you think:

• Volume per channel: are we actually sending what we said we would send?

• Reply rate per channel: which one is pulling its weight and which is being carried by the others?

• Booked meetings: the only number clients really read.

• Stuck threads: conversations that went warm and then went quiet, ordered by how long they have been quiet.

That last one is where most outbound tools fall down. They get you to the first reply and then lose interest. The person who replied and then went quiet is the best lead on your list. A dashboard that surfaces them at the top is doing its job.

The other thing worth tracking, quietly, is bounce and spam rate by domain. Not because you look at it every day, but because the day it spikes is the day the whole operation is about to go backwards.

Outbound is not dying. Single-channel, template-driven outbound is dying. The replacement is something that looks a lot like how people already sell to each other in real life: show up in a few places, say something specific, and do it without wearing a human out.

If your team is running an email sequence and it has stopped working, the fix is almost never a better subject line. Tell us what you are already sending and we will look at what a second and third channel would add to it.