What to Automate First in a Small Business, and What to Leave Alone

Most automations I am asked to fix were not badly built. They were built for the wrong task. Someone watched a tutorial, connected two apps, and automated something that happened twice a month and never followed the same pattern. Six weeks later it breaks, nobody remembers how it worked, and the owner decides that automation is overrated.

Automation is not overrated. It is just easy to aim at the wrong target. This post gives you a short test for deciding what is worth automating, the seven tasks I usually start with, three that I tell people to leave alone, and the arithmetic that shows whether any of it pays off.

I work as a virtual assistant with a technical background. That means I do the manual version of these tasks and I build the automated version, which gives me a fairly honest view of which is which.

The three-question test

Before you automate anything, ask three questions about it.

Does it happen often? A task that runs daily or weekly is a candidate. A task that happens twice a year is not, because the time spent building and maintaining the automation will exceed the time it saves.

Can you write the rule down? If you can describe the steps in plain sentences without saying “it depends” or “I just know,” the task can be automated. If the honest answer is “I decide based on a feeling,” a person should keep doing it.

Is a mistake cheap? Every automation gets something wrong eventually. If the cost is a reminder that goes out an hour late, that is fine. If the cost is a wrong invoice sent to your largest client, you want a human checking before it leaves.

A task needs a yes to all three. Two out of three does not mean forget it. It means the task needs a person in the loop, which is a legitimate design and often the right one.

Seven tasks to automate first

These are the ones I start with, roughly in the order that they pay back.

1. Lead capture into one place. Enquiries arrive through a website form, email, social messages, and calls. Route all of them into a single sheet or CRM automatically, with the source and a timestamp. This ends the “I think someone emailed about that” problem, and it gives you the data to see which channel actually brings customers.

2. Booking and reminders. A scheduling link, a confirmation message, and a reminder the day before. Missed appointments drop when reminders go out reliably, and nobody has to remember to send them.

3. Invoicing and payment reminders. Generate the invoice when a job is marked complete, and send a polite reminder a set number of days after the due date. Chasing money is awkward for people and trivial for a workflow. Keep a human approving the first invoice for any new client.

4. Inbox sorting. Label and route incoming email by type: quotes, support, suppliers, newsletters. Drafting replies to the common ones is possible too, but I would hold off on sending them unreviewed.

5. Weekly reporting. Pull numbers from your tools into one summary that lands in your inbox on Monday morning. Most owners either build this by hand on Sunday night or skip it entirely, and skipping it means decisions get made on memory.

6. Client onboarding. When a client signs, create their folder, send the welcome email, add the tasks to your tracker, and request the documents you need. The steps are identical every time, which is exactly why it works.

7. Social scheduling. Write content in batches, schedule it, and let it publish. Posting live does not make the content better, and it makes your week more fragmented.

Three things to leave alone

Tasks you do less than monthly. An annual tax pack or a quarterly supplier review is better served by a written checklist than by an automation. A good checklist captures most of the value at a fraction of the cost and does not break when an app changes its interface.

Relationship-heavy messages. The email to a client whose project is running late. The reply to a supplier you have worked with for years. The note to a customer who is upset. These are moments where tone carries the relationship. A person can draft them with help, but a person should send them.

Processes nobody has documented. If the steps only exist in someone’s head, an automation will lock in whatever that person happened to do on the day you asked. Write the process down first, run it by hand for two weeks, and only then decide what to automate. The documentation step is boring and it is where most of the value sits.

Rules first, AI second

Many useful automations need no AI at all. “When a form is submitted, add a row and send a confirmation” is a rule. Rules are cheap, predictable, and easy to debug when something goes wrong.

AI earns its place where the input is messy language. Reading an email and deciding whether it is a quote request or a complaint. Summarising a long thread. Drafting a first reply for a person to review. In those cases a rule cannot do the job, because people do not write in a fixed format.

My approach is to use rules for the structure and AI for the messy middle, keep the AI step narrow, and keep a human reviewing its output until it has earned trust. An AI step that can send things to customers without review is a risk you should take on deliberately, not by default.

Zapier, Make, or n8n

People ask which tool to pick, and the honest answer is that it matters less than the process underneath it.

Zapier is the easiest place to start. The interface is simple, the app library is large, and a first automation can be running in an afternoon. It can become expensive as your volume grows.

Make is more visual and handles branching logic and data manipulation more flexibly. For anything beyond a simple two-step flow, I often find it a better fit.

n8n can be self-hosted and gives the most control, including custom code where you need it. It asks for more technical comfort, so it suits people who have a technical person around.

For a first automation, choose whichever one you will actually maintain. Switching later is possible, and a clean, well-documented process moves between tools far more easily than a tangle of undocumented steps.

The maths

Here is a simple way to check whether something is worth automating. These numbers are illustrations, so replace them with your own.

Say a task takes you 5 hours a week. That is 260 hours a year. If your time is worth $30 an hour, the task costs you $7,800 a year. Now suppose an automation reduces it to one hour a week of checking. You recover 4 hours a week, which is 208 hours or about $6,240 of time per year.

The formula is frequency, multiplied by duration, multiplied by what your time is worth, minus the time spent supervising. If the result is smaller than the cost of building and maintaining the thing, leave it manual. Many small tasks fail this test, which is a good reason to be picky.

A worked example

Take a small design studio that receives about 40 enquiries a month. This is an invented example, but the shape is typical.

Today, each enquiry takes around 15 minutes of handling: copying the details into a sheet, replying with availability, sending a booking link, and creating a project folder once someone books. Forty enquiries at 15 minutes each is 10 hours a month.

After automation, the form writes straight into the sheet, an instant reply goes out with the booking link, and the folder is created when a call is booked. The human part left is reading the brief and writing a personal reply, perhaps 5 minutes each, so about 3 hours and 20 minutes a month.

That saves roughly 6 hours and 40 minutes a month, or around 80 hours a year. At $30 an hour, that is about $2,400. It is a real saving but a modest one, and I want to be honest about that. A single automation rarely changes your week. Five or six of them, built in sequence, do. The faster replies also matter more than the hours, because a lead who hears back in two minutes behaves differently from one who hears back in two days.

How I approach it

I follow the same order on every project, and the order is the point.

First, I watch. In the first week I run the manual process myself, or I shadow the person who does, and I write down every step including the awkward exceptions. Nothing gets built in that week.

Second, I write the standard operating procedure. A short, plain document that says what happens, who does it, and what counts as an exception. Often this document alone removes half the confusion.

Third, I pilot one automation on a small slice of the work. One lead source, one type of invoice, one client. I run it alongside the manual version and compare.

Fourth, I make failure loud. Every automation I build sends an alert to a human when something fails, because the dangerous failure is the quiet one. A broken connection that nobody notices for three weeks costs more than the automation ever saved.

Only after that do I widen the scope. The slow start is the cheap path, because it prevents rebuilding.

Common failures

Automating a messy process. The automation reproduces the mess, only faster.

Too many tools. Each new app is one more place for a connection to expire. Prefer fewer tools doing more.

No owner. Every automation needs a named person who knows what it does and gets the alert when it breaks. Without one, it decays.

No review of the output. Look at what the automation actually produced, not what you assume it produced. A monthly ten-minute check catches most problems.

Skipping the edge cases. The refund, the duplicate enquiry, the client with two email addresses. List them before you build, not after they break something.

What it costs

My virtual assistant plans start at $150 a month for 10 hours of admin support and go up to $850 a month for a full-time partner, which includes technical work and advanced automations. Standalone automation builds are scoped after I have looked at the process, because a three-step flow and a ten-step flow with approvals are very different jobs. Beyond my fee, expect small monthly costs for the tools themselves, and plan for an occasional hour of maintenance.

The cheapest automation project is the one that starts by deciding what not to automate. Many of the quotes I give end up smaller than the client expected for exactly that reason.

Where to start this week

You do not need a project. You need an hour. List the tasks you repeat every week, and write next to each one how long it takes. Run the three-question test on the top five. Pick the one that passes with the most hours attached, and write its steps down in plain sentences.

If that list shows you several tasks that pass, or you would rather hand the whole thing over, I can help. I am Souptik, a virtual assistant with a technical background and eight years of running operations for founders. I handle the admin work myself and build the automations around it, so the process and the system stay in step.

You can see how I work at souptik.site/process, check plans at souptik.site/pricing, or message me on LinkedIn with your list of repeated tasks. I will tell you which ones are worth automating, which are worth handing to a person, and which are best left alone.