In one line

Hand the repetitive distance to machines. Keep the last mile — judgement, voice and relationship — with a person. Confuse the two and you get work that is fast, fluent and worthless.

Most arguments about AI in marketing and sales are really arguments about where to draw one line. On one side is the repetitive distance: research, drafting, logging, compiling, monitoring, scheduling. On the other is the last mile: deciding whether something should be said at all, how it should sound, and whether this is the right moment to say it.

The principle is simple. Machines do the miles. Humans do the last one.

Getting that line right is most of the game. Get it wrong in either direction and the results are bad in completely different ways.

The miles are eating the week

Salesforce’s State of Sales research has repeatedly found that sales reps spend only around a quarter of their working week actually selling — roughly 28% in its widely cited figure. The rest goes to research, data entry, internal reporting, chasing information and administration.

Marketing teams have their own version of the same problem: compiling reports that nobody reads before the meeting, pulling the same numbers from three platforms every Monday, reformatting assets, monitoring dashboards for the one metric that moved.

None of that is where the value is. It is the distance you have to travel to get to the value. And it is exactly the kind of work machines are now genuinely good at: repetitive, rules-bound, tolerant of a first draft, and easy to check.

What counts as the last mile

The last mile is the part that carries judgement, voice or relationship. In practice:

Two ways to get the line wrong

Humans doing the miles. This is the default state of most teams. It is slow, inconsistent and exhausting, and it produces a particular kind of failure: the important work does not get done because the urgent busywork got done first. Follow-up lapses. Research gets skipped. Response times drift from minutes to days.

Machines doing the last mile. This is the newer failure, and it is spreading fast. Fully automated outreach, generic AI content published unedited, sequences that send whether or not anyone looked at them. The output is fast and fluent and worthless — and worse than worthless, because buyers learn to recognise it and discount everything that looks like it. When automation makes average content free, average content stops being a differentiator. That idea gets its own treatment in educate to differentiate.

Approval gates are how the line is enforced

In practice, this principle shows up as a small set of operational rules:

The common objection is that approval gates slow everything down. They do not, if the drafts are good. Editing a strong draft takes seconds. Writing from a blank page takes twenty minutes. The approval step is where the human last mile lives, and it is far faster than doing the whole journey on foot. Speed-to-lead is the clearest example: an automated, researched first touch held for one-click approval gets you response times measured in minutes without giving up the human voice.

A useful test for any step

When you are not sure which side of the line a task belongs on, ask this: if the buyer found out a machine did this step, would they feel relieved or deceived?

If relieved — nobody is upset that a machine compiled the meeting brief or logged the call — it is miles. Automate it. If deceived — a heartfelt note that turned out to be a template, a “personal” recommendation nobody actually considered — it is the last mile, and a person needs to own it.

Where to go next

The six pipeline automations is a practical list of the miles worth handing over first. Guardrails that make speed safe covers the rules that keep the line from drifting. And if the question is content rather than pipeline, Sold on Social is the practice built around the human half of this principle.