Key takeaways
- Score every recurring task as times per week x minutes x how rule-based (1-5), divided by the cost of a mistake (1-5). The top score is your first automation.
- Keep a one-week 'did it again' log before you buy or build anything. Mine showed I was checking email for leads about 20 times a day.
- My own order: lead alerts to my phone, then capped daily outreach, then bookkeeping categorization, then the storm-day route app.
- Three things stay manual in my companies: pricing conversations, upset customers, and the final number on a quote.
- Every automation keeps a human check. A suspected-spam lead still pushes to my phone with the caller's details; nothing is dropped silently.
The 4-question test for what a small business should automate first
'Start small' is the advice everyone gets, and it tells you nothing about which small thing. Here are the four questions I ask instead. Answer them for any task you repeat and you get a number you can rank.
1. How many times did you do it last week?
Not how many times you think you do it. Count. Under five and it is probably not worth touching yet, no matter how annoying it is.
2. How many minutes does it take each time?
Include the walk to the truck for the paperwork and the two minutes finding the last text you sent them. Short tasks done forty times beat long tasks done twice.
3. Could you write the rule on an index card? (1 to 5)
'When the website form fills, send me the name and phone' is a 5. 'Figure out whether this customer is worth a discount' is a 1. Software is good at 4s and 5s and dangerous at 1s and 2s.
4. What does a mistake cost? (1 to 5)
A duplicate text is a 1. A wrong number on a signed quote is a 5. This is the question the listicles skip, and it is the one that decides what stays human.
The score
Multiply times per week by minutes by the rule score, then divide by the mistake score. In my shop the thresholds are: over 100, automate now. 25 to 100, automate later. Under 25, leave it alone. Anything with a mistake cost of 5 is 'keep human' regardless of the number.
Checking my inbox for new leads scored 100 times a week x 1 minute x 5 / 1, which is 500. Deciding a remodel price scored 5 x 20 x 1 / 5, which is 20. That gap is the whole method.
Keep a one-week 'did it again' log before buying anything
The test only works on real counts, and owners are terrible at guessing their own counts. So before I bought or built anything I kept a log for one week.
It was one note on my phone. Every time I caught myself doing something for the second time that day, one line: what it was and a tally mark. No categories, no app, no spreadsheet. The friction has to be near zero or you stop by Wednesday.
Mine surprised me in three places:
- I opened email to see if a website form had come in about 20 times a day, for a form that filled a few times a week.
- I retyped the same 'on my way, about 15 minutes' text from scratch every time, including looking up the last one to copy the wording.
- I logged into the bank site most mornings to see what had cleared, then did nothing with the information.
Two rules for log week. First, do not buy software during it; every tool looks essential when you are tired of a task. Second, at the end of the week run only the top three tallies through the four questions. The rest can wait for next quarter's log.
If you want the money side of the top tally, the calculator below turns hours per week into dollars per year and per month.
The order I automated my own companies in, and why lead alerts came first
Four systems, in the order they went live. Each one is something I use every week, not a demo.
1. Lead alerts to my phone
A form fill on one of my sites now pushes to my phone within seconds through ntfy, a free push service, with the person's name, phone, email and message, plus a tap-to-call button. It scored highest on the test and took an afternoon to wire, so it went first.
The first version got it wrong. For privacy I had the push name only the town and nothing else. The first real inquiry arrived while I was on a job, the push said a town, and I had no idea the details were sitting in an inbox I was not watching. The push is the channel now, with everything on it.
That afternoon of work is also the one I would hand to almost any owner first, and I wrote it up step by step in get a text when someone fills out your website form.
2. Daily cold outreach, capped
This was a task I was not doing at all, which is the other reason to automate: the rule-based job you keep skipping. Style Marking now sends 25 cold emails a weekday, from a scheduled job on my Mac at 8:00 a.m., with a hard 25-a-day cap, a 90 to 180 second gap between sends, and bounce and unsubscribe lists it checks before every message.
If the Mac was asleep at 8, the run fires when it wakes. I get one tally on my phone when it finishes.
3. Bookkeeping categorization
Third because it sat in the 'automate later' band until the CPA package forced it. I built Fischer Books, a desktop app that reads my bank statements for three sets of books, categorizes the recurring vendors by rule, and puts whatever it cannot place into an inbox.
On the first load 148 of about 1,200 rows needed me, and every answer teaches the app an alias so the next statement needs less. More on that build in bookkeeping without QuickBooks.
4. The snow route and quote app
Lowest score of the four, because the task only exists on storm days, so it went last. For Revisions Unlimited it orders the night's stops by road time, shows one stop at a time with Navigate, On my way and Done, and texts a price from my tiers when someone signs up. Two weekends, and the whole thing is in the snow route app post.
When two tasks both land in 'automate now', do the faster build first. Lead alerts were live the same day; the outreach engine took three weeks and a warmed-up inbox.
What I deliberately kept manual
Three jobs scored a 5 on cost of a mistake, so they stay with me no matter how often I do them.
Pricing conversations. My snow app will assemble a price text from my tiers, and I tap send. But the conversation about why it costs that, what is included, and what happens on the fourteenth storm is a phone call or a driveway chat. A customer who has questions about price is deciding whether to trust me, and a bot answering that loses the job.
Upset customers. A missed-call text that says 'on a roof, will call you back in an hour' is fine for a new lead. It is the wrong thing to send to someone whose floor is wrong. Nothing in my companies auto-replies to a complaint. I read it and I call.
Final quote numbers. QuoteShield, my quote tool, drafts the document and the line items. The number at the bottom of anything I am liable for is typed by me after I have stood in the room. Software can draft a quote; it cannot be the one who eats the mistake.
Notice none of these scored low on frequency. I have pricing conversations every week. They stay manual because the fourth question overrules the first three.
The manual check every automation still needs
Every system above keeps a human in the loop at the one place a mistake would be expensive. This is the part I would push back on hardest if a vendor told you their tool 'just handles it'.
Suspected spam is flagged, never dropped. When a lead looks like spam, it still pushes to my phone, titled 'Possible spam' with the full contact details, and only the auto-reply email is skipped. I decide; the code never discards.
I learned this when a hidden anti-spam field named 'company' got filled by a browser's saved-address autofill, so real leads were being thrown out with no log at all. The field has a meaningless name now and every hit is logged.
The outreach engine defaults to a dry run. Sending for real takes a flag and typing the word 'send'. Failed sends do not count toward the daily cap, so a bad morning cannot burn the whole day's quota on errors.
Books that need a human block the CPA package. An early version of my statement parser booked a $15,000 transfer between my own accounts as dividend income. The review inbox caught it. The year-end package for my accountant will not export while a single row is unreviewed.
The snow app refuses bad GPS. If my phone's location is worse than 250 meters, the route starts from my shop instead of from a wrong pin, and the app tells me why.
The pattern is the same every time: let the automation decide the easy 90 percent, show you the rest, and never delete anything silently.
Do you need a developer? When no-code is enough and when it isn't
I build custom systems for a living, so weigh this accordingly, but no-code tools are the right answer more often than my industry admits.
No-code is enough when:
- It is one trigger and one action: form fills, text me.
- The volume is low and nothing breaks if one fires twice.
- A generic tool's way of doing the job is close enough to yours.
- No money moves and no books change.
You need a developer when:
- Two things can happen seconds apart and the second must not overwrite the first. My snow app had a Done tap on bad signal nearly wipe out the next one; fixing that took version checks on every write, which is not a Zapier box.
- A hard cap or a rule has to survive errors, like 25 a day and never 26.
- It touches money, quotes or your books.
- You want one system built around how your business works, not five apps that only you know how to stitch together, which is the whole point of the key-man risk argument.
Build time, from my own four: an afternoon for lead alerts, two weekends for the route app, about three weeks for outreach once the inbox was warmed, and the bookkeeping app was the biggest at roughly a month of evenings. Most owners' first automation is the afternoon one.
A free quote lays out the set build price and the monthly plan for your first one, and the automation services page lists what I have wired for other owners. I work with businesses anywhere by video and screen share, and the team here in Minnesota can come out in person when a walkthrough of the shop is the fastest way to understand the job.
Your scorecard and what to do with the top result
Here is the whole method on one card. Draw six columns on paper or in a notes app:
- Task, in five words or less.
- Times last week, from the log, not from memory.
- Minutes each time.
- Rule score, 1 to 5.
- Mistake score, 1 to 5.
- Score: times x minutes x rule, divided by mistake.
Sort by the last column. Cross out anything with a 5 in the mistake column. Over 100 is this month, 25 to 100 is this year, under 25 is not worth your attention.
Then match the top line to the post that walks through it:
- Checking for leads: a text when the form fills.
- Missed calls while you are working: missed-call text-back.
- Getting back to real estate leads: responding to leads faster.
- Chasing invoices: getting paid faster.
- Entering statements: bookkeeping without QuickBooks.
If your top line is something I have not written up, send me the card through the contact page. I come back with a clear scope for that one task: a set build price, quoted up front, and a monthly plan for just the services it uses, not a platform you have to grow into. My development team and I build it around how you work, and it keeps a human check where the mistakes are expensive.
Frequently asked questions
The task you did more than five times last week, that follows a rule you could write on an index card, and where a mistake is cheap. For most one-to-ten-person businesses that is getting new leads to your phone instantly, then follow-up texts for missed calls and unanswered quotes, then invoice reminders.
Anything that is one trigger and one action with no judgment in between: a website form fill that texts you, a missed call that sends 'on a job, will call back', an invoice reminder that goes out seven days after the due date, a confirmation text the night before an appointment. Each of these can be live in an afternoon, and in my own companies the lead alert was the first one I built.
Count it for one week instead of guessing. Keep a note on your phone and add a tally every time you do the task again. At the end of the week, multiply times by minutes by how rule-based it is on a 1-5 scale, then divide by the cost of a mistake on a 1-5 scale. In my shop a score over 100 gets automated this month, 25 to 100 this year, and under 25 is left alone.
Anything where a mistake costs you a customer or money you cannot get back: pricing conversations, replies to upset customers, and the final number on a quote or contract. In my companies those three stay manual even though I do them every week. Software can draft the quote and assemble the price text; a person should be the one who commits to it.
From my own four builds: a lead alert to your phone takes an afternoon, a route or job-tracking app takes about two weekends, a capped daily outreach system took about three weeks because the sending inbox had to be warmed up first, and a bookkeeping app that reads statements was roughly a month of evenings. Most owners' first automation is the afternoon kind. Anything quoted in months for a first automation is too big a first step.
Yes, for the simple ones. A form-fill text, a missed-call text-back and an invoice reminder can all be done with no-code tools and a monthly fee. You need a developer when two things can happen seconds apart and must not overwrite each other, when a hard cap or rule has to survive errors, when it touches money or your books, or when you want one system built around how your business works instead of generic apps stitched together.
Want help picking the first one?
Send me your 'did it again' log, or just the three tasks you are sick of. I'll come back with a clear scope for the top one, with a set price for the build and a monthly plan for just the services it needs. No templates, and you talk directly with the person who builds it.
Founder, Style Marking · owner, Revisions Unlimited. Jacob builds custom websites, apps and operations systems for small-business owners — and runs a remodeling company on the same tools he builds for clients. Every post here comes from systems he has actually shipped.