A process is ready to automate once you’ve done it enough times to know its exceptions. Before automating, ask: Where are the exceptions? How often do they happen? Is this actually how you want to keep doing this? If yes, you’re ready.
What you should (and shouldn’t) automate in your business
Automation is having another moment.
Between artificial intelligence, agentic tools, CRMs, workflow platforms, and every software company promising to save you hours a week, there’s a time gold rush afoot that can make you feel like the goal should be to remove yourself from as much of your business as possible.
But instead of asking, “Can I automate this?” I would challenge you to start with, “Do I understand this process well enough that automation will actually make it better?”
Automation should come after you’ve figured out how the process works, not before.
Automate the things you’ve already figured out
The best candidates for automation are mature processes. That means things you’ve done multiple times, intend to keep doing roughly the same way, and don’t need to reinvent every time they happen.
🔎 Read more: When AI makes sense for your business (and when it doesn’t)
Think recurring invoices and payment reminders, receipt collection, welcome email sequences, follow-up surveys, and lead information moving into your CRM. These are processes where you can say, “This is how we do this every time,” or close enough that the exceptions are already understood.
That doesn’t mean you set it up once and never look at it again. I still check automations because things break. But I don’t need a human making a fresh decision every single time the process runs.
But when is a process actually mature?
It can be all too easy to call a process done and repeatable the first time it works. Now this is how you’re going to do it every time, right?
Except the first time you do something, you almost always learn something from it. The second time, you probably learn something else. There are exceptions you didn’t know about yet, things that took longer than you expected, or steps that made perfect sense on paper and were kind of a pain once you actually did them.
🔎 Read more: How to run a quarterly business review (QBR) that works
Unless you’re building an airplane or a space shuttle, where getting it wrong the first time has some pretty massive repercussions, you probably aren’t going to spend years testing a business process before you use it. Most of us move faster than that.
So before you automate something, consider:
- Where are the exceptions?
- How often do they happen and when?
- Is this actually the way we want to keep doing this?
Then you can say, “Okay, we’re going to do it this way until we decide we don’t do it this way anymore.”
Don’t automate the parts that still need you
A process can be repeatable without every single step being something you can automate, and that’s not a bad thing.
Take my proposals. There are four sections that are basically the same every time, which is perfect for automation. But there’s also a section at the beginning that I write based on the discovery call, and the whole point of that section is for someone to read it and think, “Wow, you really heard me. You understand what I’m trying to solve.”
We tried giving AI the transcript and asked it to pull out what we needed for the proposal.
It didn’t work.
🔎 Read more: Here’s what I do for my self-funded founder clients
I still need to spend a few minutes recapping what I heard and framing it so it builds my relationship with the recipient, and puts them in the frame of mind for evaluating my proposed solution.
I have the same issue with financial models. I can give Claude a P&L and ask it to put the data into a standard cash flow format, and it can get me most of the way there. That used to take hours. But I still have to know what a cash flow statement is supposed to look like, check the assumptions it made, and add anything that hasn’t happened yet that’s based on an assumption or verbal commitment.
That’s a pretty good test for your own processes. If every time you do something, you still have to put some handling into it, there’s probably a good reason.
Automations can still cause problems
I’ll be honest: this was a funny topic for me to talk about right now because some of my own automations are currently making my life harder. I’ve been through more CRM systems than I care to count, and there are days when I look at what I’m manually tracking and think, “I had a spreadsheet in 2016 that did this wayyyy better.”
I also recently had one system automation that was supposed to send me a list of everything due in the next 48 hours. Sounds like a great idea, right?
🔎 Read more: How to delegate tasks (a decision-making framework)
Except that it sent me a giant list of task numbers with zero context.
Me: I don’t know what task 888158 is, can you produce this list with task names?
System: It is task 888158.
Thanks, robot overlords.
I’m not saying automation is bad
I’m saying that expecting the state of software and AI to allow you to automate everything, all the time, and not have to be involved isn’t there yet. And there will always be one-offs, because we say yes to getting things done and making money.
Sometimes you’re going to spend more time than you expected tweaking an automation until it actually does what you need it to do. Sometimes it’s going to break and you’ll fix it and push a button to re-run the collected errant events. And sometimes you’re spending enough time reworking the thing that you should just do it manually and try again next time.
🔎 Read more: How to take time-off as a founder (without the guilt)
So if an automation saves you time, reduces errors, or closes a few tabs in your brain, great. Keep it. If you’re spending more time babysitting the automation than you spent doing the thing manually, maybe it’s not actually helping.
Don’t stress about labels
Don’t get too hung up on whether you can call something “automated.” Ask whether it made the process better. Did it save you time? Did it reduce mistakes? Did it take something repetitive off your plate without creating a whole new job of managing the automation itself? Whether you call it a process, an SOP, an automation, or an agent, the best outcome is that you can call it done.
