How to Pick Your First Automation Instead of Boiling the Ocean
A simple scoring method for choosing the one process to automate first, so you get a paid-for win instead of a half-finished platform.
Every owner who decides to "get organized" makes the same mistake. They try to fix everything at once.
They map the whole business. They shop for a platform that does all of it. They spend six weekends configuring. Three months later, half of it is live, the team has quietly gone back to the old way for the parts that don't work, and the owner has a monthly bill and a bad taste in their mouth.
The fix is embarrassingly simple. Pick one thing. Finish it. Let it pay for itself. Then pick the next one.
The hard part is picking the right one, so here's how I do it.
Score every candidate on four numbers
Make a list of every repetitive process in your business that annoys you. Missed calls. Quote follow-up. Scheduling. Invoicing. Review requests. Onboarding a new customer. Answering the same five questions from staff. Whatever's on the list, it's on the list.
Now score each one, one to five, on four things.
Frequency. How many times a week does this happen? Under five, score it a 1. Fifty or more, score it a 5. Automation is a fixed cost against a variable payoff. Low-frequency work does not repay it.
Bleed. What does it cost when this goes wrong? Not a feeling. A number. A missed call that never gets a callback costs you the average value of a job times your close rate. A quote that never gets followed up costs the same. An invoice sent late costs you float. If you can't estimate the number, that's a signal in itself.
Clarity. Could you write down the rule in five sentences? "If a call comes in and nobody answers within 25 seconds, text the caller. If they reply, alert whoever's on duty." That's a 5. "It depends on the customer and how busy we are and what mood Dave's in" is a 1. Unclear processes don't get automated, they get formalized first, and that's a different job.
Ownership. Is there a person who currently owns this and will notice if it changes? A 5 means someone cares. A 1 means it's nobody's job, which is usually why it's broken, and also why an automation for it will rot within a month.
Multiply frequency by bleed. Add clarity. Add ownership. Rank the list. Do the top one.
Why frequency times bleed and not frequency plus bleed
Because a low-frequency, high-bleed event is a training problem, not an automation problem. If you lose one enormous deal a year to a dropped follow-up, don't build a system. Put a reminder in your calendar and pay attention.
And a high-frequency, no-bleed task is just noise. Automating it feels great and changes nothing. Plenty of people automate their Slack notifications and call it digital transformation.
The money is in things that happen constantly and cost you a little every time. Those are invisible, because no single instance hurts enough to notice. They're also where nearly all small-business revenue leaks. Ten missed calls a week, at a 30% close rate, at a $600 average job, is $93,600 a year walking out of the building without slamming the door.
Do that arithmetic with your own numbers before you do anything else. It will pick the project for you.
The one-week rule
Whatever you pick, it should be shippable in a week. Not designed in a week. Live, running, doing the work, in a week.
If it can't be, you picked something too big and you should cut it down until it can be. There is always a smaller version.
- "Automate our whole sales process" becomes "every missed call gets a text in 60 seconds."
- "Build a customer portal" becomes "send an automated status update at three fixed points in the job."
- "AI assistant for the team" becomes "a chat that answers questions from our pricing sheet and service area doc."
The smaller version is not the compromise. It's the actual project. Whatever you learn from the smaller version is what tells you if the bigger one is worth doing, and it usually changes the design of the bigger one substantially.
Instrument it before you build it
Before you turn anything on, write down the number you expect to move and how you'll read it.
- Missed-call recovery: how many missed calls last month, how many turned into a conversation, how many turned into a job.
- Quote follow-up: how many quotes went out, how many got a response, how many closed.
- Review requests: how many completed jobs, how many reviews.
Write today's numbers down on paper. Put the date on it. Because in six weeks, if you didn't write it down, you will not remember, and you will have no way to tell whether the thing you built was worth it. You'll rely on vibes, and vibes always say the new thing is great for the first month.
This is also how you fire an automation that isn't working, which is a thing you should be willing to do.
What "finished" means
An automation is finished when:
- It runs without anyone remembering to run it.
- Someone owns it by name and gets told when it breaks.
- It fails loudly, not silently. Silent failure is worse than no automation, because you've stopped watching.
- The person doing the work the old way has actually stopped doing it the old way.
That last one kills more automations than any technical problem. If your office manager still calls back every missed call manually because she doesn't trust the text, you didn't automate anything. You added a step. Watch for it. Ask.
Then do the next one
Here's the compounding part. Once you've built the first one, you've done three things you didn't have before. You have a place where lead data lives. You have a way to send messages. You have a habit of measuring.
The second automation costs half as much effort as the first, because it plugs into those. The third costs less again. That's the whole reason to go one at a time instead of buying a platform: you build the plumbing as a byproduct of solving real problems, instead of buying plumbing and then hunting for problems to justify it.
Six months of that, one project at a time, and you have a business that runs on systems. No big bang, no six-figure implementation, no weekend spent in a settings menu.
Pick the top of your list. Give it a week.
If you want help ranking the list, or a straight answer on whether the top item is even worth doing, tell me what's on it. You can also see the kinds of systems I build if you want a sense of what a finished one looks like.