The difference between digitalisation, automation and AI

The difference between digitalisation, automation and AI

The difference between digitalisation, automation and AI is what each step does with the same piece of work. Digitalisation means the information exists as a file the office actually uses. Automation means a rule you can write down does the step the same way every time. AI means a model reads what no rule can describe, and a person checks the result until the step is reliable. Each later step needs the one before it.

This is for owners and office leads in a Swiss SME who have been told to do something with AI and want to know which of the three is actually missing. One supplier invoice runs through all three as a running example. It is not a client story.

Digitalisation: the invoice becomes a file

English splits a distinction that German folds into one word. Digitisation is the conversion: paper becomes a file, or a scan becomes a PDF. Digitalisation is the office working from that file instead of from the cabinet. Neither is automation yet. Someone still opens every file, reads it, decides who approves it and types the amount in somewhere.

That is already useful: the invoice can be searched, forwarded and kept. The work around it stays the same.

In Switzerland, part of the conversion is already done for payments. The QR-bill replaced the old payment slips on 1 October 2022. Its Swiss QR code holds the payment details in digital form, next to the text on the paper (SIX). A file is not finished forever. SIX says QR-bill version 2.3 has applied since 22 November 2025, and the code now allows only the structured address. A PDF from 2023 can fail the current rule and still be “digital”.

Automation: a rule moves the file

Automation means a rule does the step, the same way every time. Every PDF that lands in the invoices mailbox is saved in this month’s folder. Invoices from a known supplier go to the person who ordered from them. Above an amount you set, a second person has to approve. When both have approved, the file goes to bookkeeping.

The rule does not understand the content. It checks a sender, a folder or an amount and acts. That is its strength: same input, same result, easy to check afterwards.

Vendors often put another name where you would expect the word AI: RPA, robotic process automation. It is software that repeats a fixed sequence of clicks: open the mailbox, save the attachment, type the amount into a field. That is automation. When the layout changes, the sequence breaks.

A rule can read the amount from a QR code, because the field has a fixed place and a fixed meaning. It cannot tell from a free-text line which project a delivery belongs to. The rule has no fixed field to read.

AI: a model reads what has no fixed shape

AI, in this split, is the step that infers from the input what a rule cannot fix in advance. The OECD recommendation, as updated in 2023, defines an AI system as a machine-based system that, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions. Inferring is the difference from a rule, which only follows. The definition is a frame, not a product description.

In the invoice example: forty suppliers, forty layouts, some invoices only as text in an email. A model returns supplier, amount, due date and a suggested cost centre. It is usually right. Not always. So the result goes into the same approval step as before, and a person confirms until you have seen enough runs.

Pasting an invoice into a chat window and typing the answer into the accounts is not automation, and often not a reliable step in the process either. It is a person with a tool. The step is in the business only once the input, the check and the destination are fixed.

The three steps in one view

The common mistake: a model, while an earlier step is missing

The common mistake is to buy “AI” while invoices still sit on one desk, or while nobody has decided who approves them. In the first case the model has nothing to read. In the second it extracts fields that go nowhere. The money goes into the third step while the first or the second is missing.

Three questions, in this order. Is the input already a file in a shared place? Can the next step be written as a rule on one page? Is there a remainder that someone has to read and interpret? Only the third answer calls for a model. Many office processes stop at the second. The rule is cheaper per run and gives the same answer every time.

Write the three answers next to an invoice that already passed through your hands twice this week. If the first answer is no, or the step already fits a rule and nothing is left over, a model is the wrong purchase this week. RPA stays in the middle column, even when the brochure prints the word AI on the licence.

Who should skip the third step

If the invoices are still paper on one desk, skip the model and make one shared place for the file first. Building a product or testing an idea is a different job from sorting one office process.

What it costs

A franc figure for “introducing AI” depends on how many documents pass through and whether a person still reads the result. A rule has no per-invoice model fee. A model does, and the check stays.

What can be priced is the hour in which you sort one process. The Process Check is 60 minutes online and costs CHF 99. You bring one recurring process. If the session takes place and produces no tangible case, the fee comes back. If you later book an implementation, the check price is credited in full.

If the whole team needs to use the same tools under the same rules, the AI in business page is the place to look. That is a workshop. The Process Check hour sorts one process.

What to do next

Take one process you still do by hand every week and mark which of the three steps is missing. If the three questions above are enough, you do not need anyone for that. If the process is concrete and the answers contradict each other, the hour on the Process Check page is what that is for.