What process automation actually means inside a company
"Process automation" sounds big, but in practice it describes something simple. You have a job that somebody does every day or every week, always the same way: copying data from an email into a spreadsheet, checking an invoice against an order, sending a report to the same three people. Automation moves that job onto a system that does it by itself, and the person is left with exceptions and checks.
Every company has jobs like this, whatever the industry. Invoices from e-Factura have to be read and passed to accounting. Shop orders have to be confirmed and handed to the courier. Quote requests from a form have to be split between colleagues. When the steps are known in advance, there is no need for a person in between. For a small company that means less time lost on copying and fewer typing mistakes. If you want to see what we build on this topic, the page about automations and integrations describes the types of projects.
One thing to keep in mind from the start: automation does not fix a bad process. If nobody knows exactly how a job is done today, or everyone does it differently, you first clarify it on paper. Only then can you hand it to a system.
Four kinds of automation and when each one fits
Not all automations are the same, and the wrong choice costs time. Below are the four kinds you meet most often, from the simplest to the most flexible. Many companies end up using two of them at once.
For flows between applications, the most common tool is the "workflow" type. In the n8n documentation, a workflow is a collection of nodes that automates a process, and a trigger is the node that starts it by itself on an event or on a schedule. The idea is the same in most tools of this kind: an event starts the chain, each step does one small thing, and at the end the result lands where it should.
RPA, meaning automation with software robots, is something else. IBM describes it as technology that carries out the repetitive tasks of an office worker, such as extracting data, filling in forms or moving files, based on rules. IBM also draws a clear line: RPA is process-driven, while AI is data-driven. In other words, RPA does exactly what you wrote for it, and AI copes even when the input doesn't always look the same.
| Kind | What it does | It fits when |
|---|---|---|
| Script or direct integration | Connects two programs through an API or files, with no interface. | You have one stable data transfer between two systems. |
| Workflow between applications | An event starts a series of steps across several services. | The steps are clear and you want to change them without a programmer. |
| RPA | A robot repeats a person's movements on screen, in programs without an API. | The old program has no other way in and the format doesn't change. |
| AI or agent | Reads free text, decides and acts within given limits. | The input varies: emails, invoices with a different layout, freely written requests. |
30 minutes, free. We'll tell you honestly if it makes sense.
How to pick the first process to automate
The temptation is to start with the most annoying process or the one most visible to management. It is rarely the right choice. A first project succeeds when the process repeats often, has steps you can write down, uses data that is already digital and, if it goes wrong, someone notices quickly and fixes it cheaply.
Context matters too. Since 2024, e-Factura has been mandatory for invoices between companies in Romania, so a lot of accounting data already arrives in a structured format. For that reason, receiving invoices and bank reconciliation are among the best candidates for an ordinary company. We described a ten-minute scoring method in the article on which processes to automate first with AI, and the criteria work the same without AI.
- 01Volume. If a process shows up twice a month, the time saved doesn't cover the setup work.
- 02Rules. If the answer to "how is it done" starts with "it depends", the process isn't ready yet.
- 03Data. Paper and photos on a phone slow down any automation. Files, emails and data in programs work well.
- 04Stakes. The first project shouldn't be the one where a mistake costs you a customer.
Five steps to start without getting tangled
You don't need a long project to begin. You need a short list, a stopwatch and the discipline of not changing three things at once. The steps below are the same for a company of five people and one of fifty.
- 01Write down the repetitive jobs in the company the way they are done today, with the name of the person who does them.
- 02Next to each one, note how long it takes once and how many times a month it happens. Time it once, don't guess.
- 03Pick a single process, the one with good volume and low risk, and describe it in numbered steps, including the exceptions your colleagues know about.
- 04Automate only that process and keep a person checking it for the first weeks, so you can see where the system gets it wrong.
- 05Compare the time afterwards with the time you measured at the start. If the difference is real, move to the second process. If not, understand why before you spend more.
Where an automation project usually breaks
Most failures don't come from the tool chosen but from how the ground was prepared. You see three patterns most often. First: a process nobody has described gets automated, so the system learns somebody's habits instead of a rule. Second: there is no measured "before", so nobody can say after three months whether it was worth it. Third: the automation runs with no human eyes on it, and the first small mistake repeats a hundred times before anyone sees it.
There is one more trap, specific to companies running older software. If the accounting program or ERP has no API, you don't have to replace it. You build alongside it: you read the data or use the official import and export flow. On SAGA, for example, one of our systems reads the database without writing to it and checks declarations before they are filed. We built bank reconciliation starting from the MT940 statement format, and the flow only flags what is not yet recorded in the ledger.
It also helps to know where other companies stand. According to Eurostat, in 2025 20.0% of EU enterprises with at least 10 employees used artificial intelligence technologies, and Romania had the lowest share in the EU at 5.2%. The figure is about AI, not automation in general, but it shows how much room is still left. A company that automates its most boring processes today doesn't need to be at the technological front, only more orderly than it was yesterday.
What it costs and when it isn't worth it
The price depends on the process you choose, not on the fashion of the tool. A simple transfer between two applications costs little and is done quickly. A flow with many exceptions, a program without an API or documents to read costs more and needs serious testing. That is why we don't give a price before we see the process. We work with a short discovery phase ending in a written report, followed by a fixed-price offer on a written scope. You can stop after the report.
There are also cases where automating isn't worth it. If a job happens rarely, if one person does it in twenty minutes a week, or if the rules change every month, a well-made spreadsheet and a checklist win. Automation is an investment in repetition. Without repetition, it has nothing to pay back.
Sources and further reading.
30 minutes, free. We'll tell you honestly if it makes sense.
Frequently asked questions
What does process automation mean?
It means a repetitive job with known steps is done by a system instead of a person: copying data, checking documents, sending reports. The person is left with exceptions and control.
What is the difference between classic automation, RPA and AI?
Classic automation and RPA run rules written in advance, on input with a constant format. AI steps in when the input varies, for example a freely written email or an invoice with a different layout. Many companies combine two of them.
Which processes can a small company automate?
The ones that repeat often and have clear steps: receiving invoices, bank reconciliation, confirming orders, recurring reports from the accounting program or the shop, and distributing requests received through a form.
Do I have to replace my accounting program or ERP in order to automate?
Usually not. Automation is built alongside the existing system: through an API where there is one, by reading the data or through the official import and export flow, without modifying the core of the program.
Let's see what can be automated in your business.
A free 30-minute session: we'll tell you what can be automated, how long it takes and what it costs, with a fixed price after discovery.