What you shouldn’t automate might matter more | By Martin Soler

One of the most helpful ways to define automation is by looking at what shouldn’t be automated.

It’s tempting to automate everything. Tools keep getting better, AI keeps getting faster, and workflows keep promising “10x productivity.” But in the process, we risk automating the very things we shouldn’t.

BTW - subscribe to my weekly newsletter for more. 

Here’s a principle I think could work: if the end goal is to move a process forward, automate it. If the end goal is to move a person, don’t.

We should absolutely automate repetitive procedures: rate updates, inventory pushes, internal reports, invoice emails, check-in messages. These are inputs and outputs that follow a predictable path. Systems love predictable paths.

And humans do too.

However, human communication doesn’t (whether it’s a guest email, a marketing post, a social media reply, or even the photo you choose to lead an ad) isn’t just an output. It’s a signal. And people have a sixth sense for knowing when that signal is fake. Just like we notice when the physics in a CGI jump are a bit off, or when an actor lands weirdly in a motion graphic, we might not know why, but we instantly know that something’s wrong.

This is exactly what happens when content meant to connect with a person is pushed out automatically. Sending out an invoice or a confirmation isn’t meant to connect.

AI imagery isn’t the issue. Automation isn’t the issue. The problem is when the communication or action is triggered by automation. When the machine decides the timing, the tone, the delivery (while it’s meant for a human) we can feel it is off.

Even hyper-personalized recommendations (yes, Spotify, Netflix, Youtube, I’m thinking of you) still feel mechanical. You know it’s a machine pretending to know you, not a person who actually does. We have accepted that the machine is about 20% right, but we wouldn’t accept a human doing that.

I think the most dangerous form of automation (reputationally) is the one that tries to fake human intention. Because humans feel intention. And when there is none, trust breaks.

So, automate all you want, just don’t automate the human parts. The email that’s meant to move someone. The social media post that’s meant to engage. The welcome that’s meant to help them feel appreciated. The concierge that is just a search bar. Those need intention. And intention is still (for now) a human job.

Because the moment people realize they’re on the receiving end of a system (not a person) they disengage. That’s when automation becomes spam.

So looking at what not to automate might give you more ideas of what you should automate. And as Chaplin satirizes well above, we can’t quite automate everything.