The Line You Draw

Nobody decides to automate their business. It just happens to them, one convenient tool at a time, and then one day they look up and something they used to care about is being handled by a scheduled task.

I want to talk about the list that prevents that. It's short, it takes an hour to write, and I'd guess fewer than one in fifty operators has one.

It's a list of things you will never automate. Written down. On purpose. Before the question comes up.

The reason this is a strategy piece rather than a philosophy piece is that the economics changed. Two years ago you didn't need this list, because capability drew the line for you. Plenty of work simply couldn't be automated, so there was nothing to decide. That's over. Most of what happens in a small business can now be handled by some combination of a model and a workflow, at a quality level that's good enough that nobody complains.

When capability stops drawing the line, somebody has to draw it deliberately. If you don't, the line gets drawn by whatever was easiest to set up that quarter, which is a terrible way to make decisions about your business.

Let me describe the pattern, because it's specific and you'll recognize it.

You automate something small and obviously correct. Appointment reminders, say. Works great, nobody misses it, you get twenty minutes back.

Then something adjacent. The follow up after the appointment. Also fine.

Then the check in at thirty days, which used to be a phone call you made yourself. Now it's an email that goes out automatically. Still fine, mostly. Response rate drops a little and you notice and decide it's a seasonal thing.

Then the win back sequence for people who went quiet, which used to be you noticing somebody had gone quiet.

Two years in, you have not spoken unprompted to a customer in eleven months. Every touch is triggered. Nothing in your business surprises you anymore because nothing reaches you before it's been categorized.

At no point did you make a bad decision. Every single step was individually correct. That's what makes drift dangerous. It's not made of mistakes. It's made of small good choices that add up to a business you wouldn't have designed.

The list is the thing that catches it. Not because a written list has magic power, but because it forces you to notice when you're about to cross something off it, and noticing is the whole game.

Here's how I decide what goes on mine. Three questions, and anything that fails any of them stays human.

Is the relationship itself the product here?

There's a category of interaction where the content of the message is almost irrelevant and the fact that a specific person sent it is the entire value. A note after a client's father died. A call when a project went sideways. The check in with the customer who's been with you six years and hasn't ordered in three months.

You can automate all of these. The output will be competent. And if the recipient ever finds out, and eventually somebody always finds out, you don't just lose the value of that message. You retroactively lose the value of every similar message you ever sent, because now they're recategorized as maybe automated too.

That's the asymmetry that makes this a business decision rather than a sentimental one. The upside of automating is a few minutes. The downside is a portfolio wide revaluation of your sincerity. Bad trade, every time.

If this goes wrong, how bad is it, and can I take it back?

Two dimensions here and both matter. Something can go wrong often and cost nothing, which is fine. Something can go wrong rarely and cost everything, which is not.

Pricing decisions, hiring decisions, firing a client, anything with legal weight, anything involving a promise about the future. These fail the test not because a model would decide badly, but because the failure mode is irreversible and the cost of being wrong isn't symmetric with the benefit of being right.

A useful version of this question: if this decision goes wrong, will I be able to explain how it was made? If the answer is "it was in a workflow somewhere," you've automated something that needed a name attached to it.

This is the one almost nobody thinks about and the one I'd argue is most important.

Does doing this work myself keep me good at something I need to be good at?

Some work is valuable for what it produces. Other work is valuable for what it teaches you while you do it. Sales calls teach you what objections are forming this quarter. Reading support tickets teaches you where the product is confusing. Writing your own content teaches you what you actually think.

When you automate work in the second category, you get the output and lose the education. And the education was the point, because the education is what allows you to supervise the automation.

This is the failure that sneaks up on people. You automate customer communication, so you stop reading customer communication, so you lose your feel for how customers talk, so you can no longer tell whether the automated communication sounds right. You've removed your own ability to detect the failure of the thing you built. You'll find out it broke when revenue moves, which is six months late.

Anything that fails the learning test doesn't have to stay fully manual. But you need a sampling rule. Read ten percent. Take one call a week. Write one piece a month yourself. Keep enough contact to stay calibrated.

An hour. Do it today if you can.

Start by listing every recurring activity in your business that touches a customer, a decision, or money. Twenty to forty items for most small operations. Don't organize, just dump.

Run each one through the three tests. Most will pass all three cleanly and those are your automation candidates, whether or not you've gotten to them.

What fails goes on the list. And here's the constraint that makes this useful: cap it at seven items.

The cap is doing real work. A list of thirty things you won't automate isn't a strategy, it's a mood, and you'll ignore it within a month. Seven things you'll actually defend under pressure is a strategy. If you've got twelve candidates, rank them and cut five, and notice how much you learn about your own business from which five you're willing to lose.

Write each one specifically. Not "customer relationships." That's not a line, it's a sentiment. Try "the first call with any new client, always me, no exceptions" or "any message where somebody is receiving bad news" or "the decision to fire a client." Specific enough that you'd know immediately if you'd crossed it.

Then put the list somewhere your team can see it, and tell them what it's for. Half its value is that other people know it exists, because they're the ones who'll be building the automation that quietly crosses it while you're not looking.

I don't want this to read as a case for doing things the hard way. Most sentiment about manual work is nostalgia wearing a suit, and nostalgia is expensive.

So here's the other list. Things people protect that they absolutely should not.

Scheduling. There is no relationship value in coordinating calendars, only friction, and the friction is costing you meetings.

Data entry of any kind. Nobody has ever built a customer relationship through accurate CRM hygiene. Wire it up in Make and never think about it again.

Meeting notes. The recording and the transcription and the first pass summary should all be machine work. Fathom does this well and the thing you should keep for yourself is reading the summary and deciding what it means, not producing it.

First drafts of anything routine. Proposals, reports, recurring updates. The judgment is in the edit, not the blank page.

Reminders and follow up scheduling. The thing that has value is the follow up conversation, not remembering that it's due.

Time tracking and the reporting on top of it. I run Rize for this and the only human part is looking at the numbers once a month and being mildly horrified.

If any of these are on your do not automate list, take them off. You're protecting effort rather than value, and those feel identical from the inside.

Here's the strategic argument, and it's the reason this is a Sunday piece.

As automation becomes universal in your category, the automated parts of everyone's business converge. Your competitors get the same tools you have, running the same patterns, producing similar output. That's not a prediction, it's just what happens when capability becomes cheap and available.

What doesn't converge is the seven things. Those stay differentiated by definition, because they're the parts where a specific person is doing something specific, and that can't be copied by buying software.

So the list isn't a constraint on your efficiency. It's a description of where your margin is going to live in three years. Everything you automate becomes table stakes. Everything you deliberately don't becomes the reason someone picks you.

That reframe changes how the list feels. It stops being a set of things you're sacrificing efficiency on and starts being the actual asset, with the automation existing to protect it. You automate ruthlessly so that you have the time and attention to be excellent at seven things. That's the whole design.

Put a recurring block on the calendar for the first week of each quarter. Twenty minutes.

Read the list. Ask two questions. Did I cross any of these without noticing? And is anything on here that's become nostalgia rather than strategy?

Both directions matter. Sometimes you'll find you've drifted across a line and need to pull something back. Sometimes you'll find an item that made sense two years ago and is now just habit dressed up as principle. Cut those without guilt.

The list should feel a little uncomfortable. If defending every item is easy, you've made it too safe and you're probably protecting things you should have handed off.

Take the hour today. Twenty items dumped on a page, three tests, cap at seven, tell your team.

And I'd add one thing about how to pick. When you're stuck between two candidates, ask which one you'd be embarrassed to have automated if a customer found out. That question cuts through more than the three tests combined, and it takes about four seconds.

New week starts tomorrow. See you then.

Jordan

The AI Newsroom is written by Jordan Hale. This issue contains affiliate links to tools I actually use. If you sign up through them I may earn a commission at no extra cost to you.