What should I automate with AI first?

Start with the task that is highest-frequency, most repetitive, and lowest-judgment — not the flashiest one. Run every candidate through three filters: how many hours it eats each week, how repeatable the steps are, and how little judgment it needs. The winner is usually something boring, like lead follow-up or appointment reminders, not a clever agent. Fix the process before you automate it, or you just make the mess run faster. Then add the next one only when the last is working. That sequence, not the tool you pick, is what makes automation actually compound.

Why is automating the wrong task first the costliest mistake?

It’s in our nature to try to do the most clever or impressive thing first. That’s exactly how a lot of founders reach when it comes to AI and automation. We go for the thing that looks the coolest and we can brag on social media for it, but in reality it’s not giving you any value in return. I can mention fully automating social media output, but that’s not even what I’m talking about here. I’m more just talking about automating the actual work that you do before automating admin tasks. You want to be automating the boring high-frequency, low-impact ones before the high-impact ones. The high impact tasks are much more complicated. They require certain skills, certain knowledge, certain nuance, and it’s very specific in how they operate. When you try to get those intricacies built into an automation you run the risk of messing it up. If AI doesn’t understand what you’re trying to do, its default is to just assume. But what if you can’t let it assume? You tell it everything, but you forget something or it doesn’t actually have the right skills or tools to go do the automation and you get output that sucks. And now you’ve essentially wasted your time building that automation, and you’ve spent just as much time building it as it would have just to do it yourself.

Versus what if you spent the time automating the admin tasks? Around 36% of an entrepreneur’s week goes to small admin tasks. And if you have managers, 55% of them lose a full workday per week to manual repetitive tasks. But these tasks all happen the same way every time. When you’re training somebody new on them, they follow a standard operating procedure that has a numbered list of instructions. And it’s exactly those instructions that you can give to AI to make automating that task super easy. The tasks are so straightforward that they are actually easy to automate and immediately provide value once they’re built. They very rarely have to be rebuilt. And when they do, it’s as simple as, hey, change this step in the process.

What’s the three-filter test for choosing your first automation?

Here’s a test you can use on deciding whether or not a task is worthy of an automation.

Firstly, count how many hours it eats. The more hours per week it eats, the more likely an automation for it will be valuable. Something that takes you one hour a week should probably still take you that one hour a week. Whereas something that takes you 10 plus hours a week serves you much better in being automated simply because it’s eating more than a day.

Next is how repeatable the steps are. I’ll take this back to the standard operating procedure (SOP) note. An SOP is simply a set of instructions that someone can follow to perform a task. You can directly give AI an SOP and tell it to turn it into an automation. If the steps are repeatable the AI can write them as almost exact instructions as well.

The third test is judgment. The higher the judgment required the more likely AI is to get it wrong. If we take a look back at the repeated instructions, say we have an instruction where we need to make a decision. If that decision requires that nuance or expertise, then AI is less likely to be able to reliably make the right decision. Inversely, if we have a decision that is purely based on information and it’s a guaranteed answer, AI can simply find the answer and you are good.

Putting it all together, the tasks that are best automatable are the highest hours tasks that have very repeated steps and if it requires decision, that decision is based on data not expertise. The winners of this, to the biggest surprise, are the admin tasks. They are done the same every time, they take up the most hours, and there’s very little judgment to them.

Measure yours in 3 minutes: the free Task Audit scores how much of your recurring work could run without you.

Why should you fix the process before you automate it?

A system or process, in the traditional business sense, is just a repeatable sequence of activities designed to achieve a goal. But, if you are running a process, and that goal isn’t achieved, or if it’s only sometimes achieved, then the process is broken. If you were to automate a broken process, it doesn’t magically fix it. Automating just moves you faster to throwing the results in the trash like you did before. You just get a faster mess. In fact, roughly 73% of automation projects fail because teams automate a broken process instead of fixing it first. Automations are meant to speed up what’s already good and running smoothly. When applied to efficient operations, automation magnifies the efficiency. When applied to inefficient ones, automation magnifies the issues.

To make sure you are not making a mess when automating, build the process correctly first. A process should always result in a correct outcome and it should be easily repeatable and followable by almost anyone. A process should be trusted and essentially THE way of doing something in the business before it is even thought about being automated. When it’s a person running the process, you don’t really have to worry about the issues too much. If the result is wrong, hopefully, the person will catch it and fix it. They’ll notice, oh, the email didn’t send, let me resend it. They’ll generally notice what’s wrong and they absorb the exceptions without even thinking about it. AI, on the other hand, removes that safety net entirely. AI simply acts how you tell it to, and, unless you are really good at prompting AI, it is very unlikely to catch the issues when they come up in a broken process. It doesn’t know why the process is broken. It doesn’t know it forgot things. Generally, it creates broken output at scale with nobody watching.

What’s the boring first automation most small teams should pick?

The lowest effort, highest value automations you can employ to benefit your business aren’t even AI. They’re the ones that are built into the tools you already use. If you use a scheduling tool like Acuity or Calendly, you have built in reminder automations. You can customize your reminders and intake forms and they work out of the box. If you are not doing this you are missing out on the features you are paying for. When you set up reminders for your meetings, it cuts no-shows by 20-35%. When adding intake forms, you can prepare yourself for every conversation you have. Search through all of the tools you already use for things like these. If it’s not replacing human judgment and empathy (an automated follow-up email doesn’t come across well), then add it. Done-right automations recover ~10-15 hours/employee/week, and generally, the tools you already use have something that has immediate value.

After your owned apps you start to move into territory that AI can handle. An example of this would be client intake and onboarding. A good client experience is a predictable and repeatable process. They sign the contract, pay the invoice, and get their onboarding scheduled and material sent. This should be the same across all clients and the only thing that changes is the client information. A good first automation here could be automatically sending something within that process when the item before it is completed. As in, when the contract is signed they get sent the invoice, or when the invoice is paid, they get sent an onboarding email. If you are no longer filling in and sending these items, that’s valuable time back on your plate to go sign another client. The reason it’s a good task for AI is because it’s simply just sending data to a new area. It’s boring data entry and monitoring. A contract gathers the client’s name and email for the invoice and onboarding email, the AI watches for each item coming back, fills in the new one and sends.

How do you add the next automation without creating tool sprawl?

We want to be automating one thing at a time, and getting it right before we move on. And we want to pick small, not the entire process. If our goal is to automate the whole process, we want to look at each individual step as its own possible automation. Pick one task, work with AI to automate it, get it 100% right and run a few confirmation tests, then you can move to the next task. One in when the last works. Why? Because an automation is meant to operate without you, so if it’s not correct and we spend more time fixing it, it’s not doing its job. Our goal is hours back and not a clever automation stack that doesn’t work. So, there’s no need to rush into this. Move slowly, automate one thing at a time, while continuing to do all the other processes. Once you finish one automation make sure it works and move to the next. With each automation that you complete, you’ll notice something. You’ve started to get more and more time back, while also getting better at automating your business. And so creating the automations actually becomes easier. The time it took you to make your first one is much longer than the time it took you to make your fifth. And by the time you’ve made your fifth one, you’ve automated enough that you’ve gotten a few hours back a week, which you can then spend automating more of your business. And so you can see it starts to compound for you. One automation gives you a few minutes back, but the fifth gives you an hour and the 10th gives you an entire day back a week. Being smart with what you automate will create much more value than rushing into it because you think you can create something that looks flashy and only maybe gets you an hour back a week.

Frequently asked questions

Why is automating the wrong task first the costliest mistake?

Founders reach for the clever, impressive automation they can brag about, but high-impact work takes skill, knowledge, and nuance, so AI is likely to get it wrong and you waste as much time building it as you’d spend doing the task yourself. Admin tasks happen the same way every time, so they’re easy to automate and pay off immediately. Start there, not with the flashy thing.

What’s the three-filter test for choosing your first automation?

Run every candidate through three filters: how many hours it eats each week, how repeatable the steps are, and how much judgment it needs. The best tasks to automate are high-hour, highly repeatable ones where any decision runs on data, not expertise. To most people’s surprise, that usually points straight at admin work.

Why should you fix the process before you automate it?

A process is a repeatable sequence of steps meant to hit a goal, and if it doesn’t reliably hit that goal, it’s broken. Automating a broken process doesn’t fix it, it just gets you to the same mess faster, which is why roughly 73% of automation projects fail. Build the process so it produces the right outcome every time and anyone can follow it, then automate it.

What’s the boring first automation most small teams should pick?

The best first automations often aren’t even AI. They’re the reminders and intake forms already built into tools you pay for, like Acuity or Calendly, and turning on reminders alone cuts no-shows by 20 to 35%. After that, point AI at client intake and onboarding, having it send the next step when the last one completes, like firing the invoice the moment the contract is signed.

How do you add the next automation without creating tool sprawl?

Automate one thing at a time and get it fully working before you move on. Pick a single step, not the whole process, and only add the next when the last one runs without you. The goal is hours back, not a flashy stack that doesn’t work, and it compounds: your first automation gives a few minutes back, your fifth an hour, your tenth a whole day.

Not sure which task to automate first? A free Task Audit maps where your hours are going and shows you the one to start with: https://systemssuccesspro.com/task-audit/

Photo by Annie Spratt on Unsplash.