Which tasks in your business you can automate (and which you shouldn’t)

It’s half past nine on Monday and someone on your team is already deep in the weekly report: copying figures from three spreadsheets, checking the totals and sending it round. On Tuesday, orders that arrived by email get typed in by hand. On Thursday, supplier invoices are checked one by one. Nobody complains. It’s always been done this way.

A machine could do most of that, though not always with artificial intelligence. This guide helps you spot those tasks, choose between simple automation and AI, and decide what should stay in human hands.

How to tell if a task can be automated

A task can be automated when it’s repetitive, follows clear rules, happens often and, done by hand, invites mistakes. If it ticks at least three of these four boxes, it’s a good candidate.

• It’s repetitive. Same steps, same order, same programs.
• It follows rules. You could explain it to a new starter on one page: “if the order is above a certain amount, tell John; if not, send it to the warehouse”.
• It has volume. Daily or weekly, not twice a year.
• It invites mistakes. Copying figures, typing product codes, reconciling amounts: exactly what a tired person gets wrong at six on a Friday.

A quick test: if the person who does it says “I could do it in my sleep”, a machine can probably do it.

What these tasks really cost you

A repetitive task costs the time it takes, what your team isn’t doing in the meantime, and the mistakes that slip through. Those hours are hard to see because they’re spread out: ten minutes here, half an hour there.

The mistakes show up later: a duplicate invoice paid twice, a mistyped order that reaches the customer short. And there’s a quieter cost: the people who know your business best spend part of their week acting as a photocopier. That wears people down.

Automation or AI: when to use each

If the data arrives in a tidy format and the rules are clear, ordinary automation is enough; AI only pays off when something has to understand messy text.

Ordinary automation is a program that follows fixed instructions: take this data from here, put it there, and flag anything that doesn’t add up. It can produce the Monday report from your ERP (the system you use for orders, stock and invoicing), send reminders for overdue invoices or match bank payments against your invoices. It’s cheap to maintain, predictable and easy to check.

AI genuinely helps with four kinds of work:

1. Reading documents and emails. PDF invoices from different suppliers, each with its own layout, or orders written out in an email.
2. Sorting. Deciding whether an email is an order, a complaint or a question, and sending it to the right person.
3. Drafting. A first reply to a customer or a summary of a long contract, which a person then reviews.
4. Answering simple questions. “Where is my order?”, “Could you resend the March invoice?”. Anything complicated goes to a person.

Outside those cases, AI is usually a more expensive and less predictable way of doing what a rule does well: you pay for every use, and its mistakes aren’t always obvious. Often the best answer combines both: AI reads the invoice and ordinary automation does the rest.

What should stay with people

Anything that needs judgement, has consequences that are hard to undo or depends on trust should stay with a person.

• Decisions involving money or risk. Approving a large payment, extending credit to a new customer, changing a price.
• Exceptions. The unusual order, the long-standing customer whose problem doesn’t fit any box.
• Relationships. A negotiation, an angry customer, a difficult conversation with a key supplier.
• Final responsibility. If something goes out under your company’s name, someone on your team should have seen it.

The practical rule: the machine does the dull part and flags what needs a human look. The person decides.

Common mistakes when automating

The most common mistakes are automating a process that already works badly, starting with the showiest idea and not measuring.

• Automating a mess. If everyone does the task their own way, agree on one way first. Otherwise you just get the same mess, only faster.
• Starting with the showiest idea. An AI assistant on your website sounds good in a meeting. Giving someone their Monday morning back is less glamorous, but you feel it sooner.
• Not measuring. Without knowing how long the task took before, you can’t tell whether it was worth it.

How to decide where to start

Start with the task that saves the most time, happens most often and does the least damage if something goes wrong. Ask three questions:

1. How long does it take each time?
2. How often does it happen?
3. What happens if it goes wrong? Is it fixed in five minutes, or does a wrong payment leave the bank?

Time multiplied by frequency gives you the hours at stake. Risk doesn’t rule a task out: it tells you how much human review it needs. The best first candidates have plenty of hours at stake and little risk.

Imagine a 60-person distributor that receives orders by email and prepares a sales report every Monday. The report is the first candidate: it repeats, follows fixed rules and, if something breaks, you see it straight away. The email orders would come next, with AI reading them and a person checking before confirming.

Start with a single task. For a couple of weeks, note how many hours it takes, then automate it and measure again. If it saves what you expected, move on to the next.

A checklist you can use this week:

• Ask each department head for three tasks their team repeats every week.
• For each one, note how long it takes and how often it happens.
• Mark which follow clear rules and which depend on reading messy text.
• Park, for now, anything that would cause serious damage if it failed.
• Pick one. Just one.

How we approach it at vitamina.dev

At vitamina.dev we start by sitting down with whoever does the task, not by choosing a tool. We look at how long it takes, where mistakes creep in and which exceptions come up, and we pick the first task based on the hours it saves and the risk involved.

Then we use the simplest solution that works. Our experience with management systems, financial platforms and AI integration lets us tell when AI adds value and when it isn’t needed. Where there’s risk, a person reviews. And we measure the hours saved before moving on.

Frequently asked questions

What business tasks can be automated?

Tasks that are repetitive, follow clear rules and happen often. For example: moving data from emails into your ERP, producing regular reports, sending payment reminders or sorting incoming emails. Anything that calls for judgement or carries serious risk is best left with a person.

Do I need AI to automate tasks in my business?

Not always. If the data arrives in a tidy format and the rules are fixed, ordinary automation is cheaper and more predictable. AI pays off when text has to be read or written: invoices in different layouts, orders that arrive in an email or replies to customers.

Can I automate without replacing the software I already use?

In many cases, yes. Many management systems, banks and email tools let data be exchanged automatically. If an older program doesn’t allow it, look at alternatives, but switching systems shouldn’t be your first step.