Working alone? What should you automate first?

The first thing to automate is work that repeats regularly, has clear rules, and is easy to fix if something goes wrong. The question I get asked most often is: "I want to use AI to automate my work, but I don't know where to start." My answer is always the same: don't start with the tool, start with your own work week. This post goes through five steps, from recording a week of work to documenting how you do things, along with the most common mistakes I see when helping small businesses.

Why automate repetitive work first?

Repetitive work is where a solo worker's time gets eaten up the most. Asana published in their Anatomy of Work Index report, surveying over 10,000 knowledge workers, that about 60% of time goes into "work about work" like finding information, status meetings, tracking progress, and copying data back and forth (Asana's page does not mention the year of the survey, read on 25/10/2026). This is a survey conducted by a software company itself, not a measurement for each solo worker, so treat it as a hint to measure your own. Here's a hypothetical example, calculated in three steps:

1. First, you spend 2 hours a day on repetitive tasks.
2. Next, a 5-day work week is 10 hours.
3. Finally, just cutting 50% gets you back 5 hours a week.

How do you pick the first task to give to AI?

Choosing has two steps. Step 1 is to record a real week of work: each time you finish a task, write down one line with what you did, how long it took, and whether it repeats. You don't need to be exact to the minute, the goal is to see where your time goes. By the end of the week, a solo worker usually sees a group of similar tasks: answering the same type of question, copying data from one place to another, making reports from a template, posting content to multiple channels.

Step 2 is to score each repetitive task with three questions:

1. Does it repeat regularly? Every day or every week without fail.
2. Are the rules clear? If you can write down steps for someone else to do it, AI can do it too.
3. Is one mistake serious? If doing it wrong costs money or customers, leave it for later.

Hypothetical example: answering messages asking for prices, repeats every day with about 20 messages, has clear rules and mistakes are easy to fix, so it goes to the top of the list. Meanwhile, making a quarterly tax report, mistakes cost fines, so it goes to the bottom.

Step 3: How do you build the smallest working version?

The smallest working version is the simplest version of a task, instead of building a whole system right away. First pick exactly one task. Next make the simplest version of that task. For example:

• When a message asks for a price, AI writes a draft answer, you just read it and hit send.
• When a new order comes in, it automatically logs to Google Sheets and sends you a notification.
• Every morning it gathers the tasks you need to do from your calendar and email.

I use n8n for most workflows like this because it's drag-and-drop, I can set it up myself, and it connects to most tools I know. If you don't want to set it up yourself, you can start with Make or Zapier. Hypothetical example: each price inquiry takes 3 minutes, 20 a day is 60 minutes; an auto draft cuts that to about 1 minute each, so you get back almost 40 minutes a day, about 67% of the time for that task. This number is just to calculate ahead, you need to measure it yourself with your weekly log.

Should you pick n8n, Make, or Zapier to start?

All three can build the smallest working version, they differ in how they count limits and cost. Three sources I've read: If you want to see a complete system running, read the AI system I use to run my solo work.

1. According to Zapier's official pricing page (read on 25/10/2026), the free plan has 100 tasks a month, and the Professional plan is 19.99 USD a month when paid yearly with 750 tasks.
2. According to Make's official pricing page (read on 25/10/2026), the free plan has 1,000 credit a month; I couldn't read Make's paid plan pricing table so I'm not listing it here.
3. With n8n, I set it up on my own server, so the main cost is the server and maintenance time; I haven't verified the license terms, so read them yourself before using it for business.

Hypothetical example: a workflow logging new orders to Google Sheets with 10 orders a day costs about 300 tasks a month if each order is one task, way beyond Zapier's free plan of 100 tasks.

Why You Need Someone to Review and Document the Process

Step 4 is for humans to review at the final stage, which means in the first few weeks AI only produces drafts and you review them. For example, when AI drafts replies to customers over Zalo or Gmail, you'll see where AI tends to make mistakes, adjust the instructions, then gradually trust it with lower-risk tasks. For things like sending money, deleting data, or dealing with difficult customers, you should keep the review step long-term.

Then at step 5, once each process runs smoothly, write down one short page: what the process does, where it gets data from, where to look if it breaks. After a few months you'll have many processes, and if you don't document them, you'll forget how they work yourself.

What are the biggest mistakes I see most often?

The biggest mistake is starting with tools instead of starting with work. Three mistakes I see most often when working with small businesses are:

• Buying tools before you know what work needs to be done. A nice tool like Notion or Zapier can't replace a clear task list.
• Automating work you do once a quarter. The time to set it up takes longer than doing it by hand.
• Handing AI work you haven't done well by hand yourself yet. If the process is already messy, AI will just make the messy process faster.

Start small, measure by how much time you get back, and expand gradually. That's how my system grew, and it's also how I work with small businesses when rolling out AI Automation.

Source: official pricing pages from Zapier and Make (accessed 25/10/2026); Asana's Anatomy of Work Index report (survey conducted by the company, year unclear). Examples of hours worked and order numbers are assumed; n8n licensing terms and Make's paid pricing table have not been verified. Experience is the author's.