Intelligent process automation: workflow tool, RPA or AI employee, and which fits when
Intelligent process automation is an umbrella term. What's usually meant is automation that can do more than fixed if-then rules: it can also read letters, emails or documents that look different every time, and then decide what happens with the document. If you search for the term, you'll find software, platforms and consulting offers, along with words like "intelligent automation" or "hyperautomation". For an owner with a team, a simpler question gets you further: which of three ways fits which process? In this article, we sort out the three ways, without testing tools or rating vendors, and show three questions you can use to decide for your own process.
What's behind the words
We read "intelligent automation", "hyperautomation" and similar words as umbrella terms, the way vendors use them. They usually describe a mix of several techniques: fixed rules, software robots and AI, often on one shared platform. That's not a bad idea. But for the decision in your business, the word says little. A product with "intelligent" in its name can be exactly right for your process, or not fit at all.
The same goes for "process automation software" and "process automation tools". Both cover very different things. That's why we separate by way of working: by what a way does, and by what it needs. Which product you use for it is a second question.
Way 1: fixed rules
The first way of working connects programs with if-then rules. When an order comes in through the online shop, create the customer in the accounting system. When a form is submitted, send a confirmation of receipt and add the inquiry to a list. Rules like these are usually built in a workflow tool. This category includes, for example, Zapier, Make or Power Automate.
A fit when the input is structured the same way every time, with the same fields, and the programs involved have an interface. Then a fixed rule is often set up quickly and does exactly what it says.
It needs someone who can build the rules and adjust them when something changes: a new field in the form, a new program, a special case. Anything that doesn't fit a rule either sits there or goes to a person, depending on how the rule is built.
Way 2: click robots (RPA)
RPA stands for robotic process automation. A software robot copies what a person does on screen: open a form, click a field, enter a value, save. It doesn't need an interface for that. It works with the user interface.
A fit when a program has no interface and the workflow in it is stable, for example the same forms in the same order every time. With older programs in particular, this is often the obvious way to stop entering data by hand.
It needs someone who can adjust the robot's workflow when a form changes. And just like with fixed rules, the robot follows the workflow it was built for. If a case comes up that has no workflow, the way it was built decides whether it stops or carries on.
Way 3: AI employees with a job description
An AI employee can also read input that has no fixed form: a letter, an email with an attachment, a notice with its own layout. It recognizes what it is, decides by the rules your team has taught it, and hands over anything it can't match with certainty. This way is our way.
For this to hold up, we hire an AI employee like a new member of staff: with a job description, a trial task and a permanent position. Role, tasks, permissions and limits are put in writing, along with a contact person on your team. After that, it's led like a team, with fixed rules, reports and review steps.
A fit when the input looks a little different every time, when there are many variants, and when your team can demonstrate the rules, even if it has never written them down.
It needs training and sign-off: watching how your team does it, showing every variant once with someone sitting alongside, asking through the escalation path, signing off after four clean runs. And it needs looking after. Every new case the AI employee doesn't know goes to a person and gets trained in as a rule in the feedback session.
Others call something like this an AI agent. We say AI employee, because with us it has a job description, fixed rules, an escalation path and a report, and it's led the way a person on the team is led. If you want to know what's behind it technically, Kevin explains it on his blog: What Is an AI Agent? Explained Simply.
Three questions that decide
Which way fits a process depends less on the technology than on the process itself. Three questions help you sort it out.
Is the input structured?
Does the work arrive in the same form every time, as a form, a data record, a row in a spreadsheet? Then fixed rules are often enough. Does it arrive as a letter, an unstructured email, a PDF with a changing layout? Then you need something that can read. The line isn't sharp. An invoice from the same supplier is more structured than mail from twenty different senders.
Are there variants?
Count how many kinds of case the process has. If there are two or three, they can be built as rules. If there are dozens, and new ones keep coming, every new variant turns into a new rule or a new workflow that someone has to build, with fixed rules and with click robots alike. With an AI employee, a new variant reaches a person through the escalation path and gets trained in afterwards.
Who looks after it?
Each of the three ways needs looking after. The question is who does it and what knowledge it takes. With fixed rules and click robots, looking after them takes someone who can build rules or workflows in the tool. With an AI employee, the person on your team who knows the process explains the new case. Who in your business has time for that is an honest question, and the answer is part of the decision.
In summary, by way of working:
| Fixed rules | Click robot | AI employee | |
|---|---|---|---|
| Works with | fields and interfaces | on-screen forms | text and documents, even without a fixed form |
| A new variant means | building a new rule | extending the robot's workflow | training in the case through the escalation path |
| Looked after by | someone who can build rules | someone who can adjust the robot | the person who knows the process, in the feedback session |
| With an unknown case | whatever the rule provides for | whatever the workflow provides for | escalation path to a person |
What all three ways need
Before you decide on a way, the process itself has to be described. That applies to every tool. We define every process by five points: trigger, start, work for each variant separately, end and reporting. Then there's the escalation path, meaning what happens when a case doesn't fit.
Once you write these points down, you often notice along the way which way fits. If all the variants can be stated in two sentences, a fixed rule may be enough. If the list of variants takes a whole page, and your team says "it depends" for every second one, a lot speaks for an AI employee. And if a clear end is missing, the process isn't ready yet for anything to take it over at all.
Combinations are normal
The three ways don't rule each other out, and some tools now combine several ways of working themselves, for example fixed rules with a step in which AI reads a document. Then the lines blur, and the more important question becomes who manages the rules, the escalation path and the report. An example of a combination: a fixed rule puts an email with an attachment into a folder, an AI employee reads the document and matches it, and at the end it lands in the document management system you already have. If you already use a tool that works well, you don't have to replace it.
What matters is that the whole workflow has one owner: someone who takes over the whole workflow, from the letter to the filing.
Where a person stays involved, whichever way you choose
Whichever way you pick doesn't change one rule: we keep a human involved when decisions are about people, about money or about anything with legal consequences. That's a house rule for our work, not legal advice. On top of that, every step that can't be taken back has an approval point in front of it: the process stops until a person explicitly says yes, for every single case.
What that means at the start of a process, when your team checks every run, is covered in Autonomy Level 1, Assisted: when your team checks every run. How an AI employee handles input differently from fixed rules is described in detail in AI process automation: what changes with AI compared to fixed rules. And the overview of process automation in a business with a team is in Process automation in a business: what works without developers and where a person stays involved.
What this looks like for us
At a tax advisory firm, an AI employee files tax assessment notices, both digital ones and those that arrive on paper. The notice process distinguishes around 45 rules. The reason is in our records: many types of assessment, different types of tax and of letter, and letters from the city that look different from each other.
That's exactly the situation in which we answer the questions above differently than we would for a form. The input isn't uniform, there are many variants, and the person who knows them works in the office team. That's why an AI employee works there. First, Kevin watched the office team: which types of assessment there are and how each one gets filed. Then the AI employee was trained, one type of assessment at a time, with someone sitting next to it. Anything it can't match with certainty goes to the office team by email and gets trained in afterwards.
A second AI employee checks the inbox every 30 minutes, spots the emails with paper notices and hands them over to filing. Filing happens in the document management system the firm already uses. Both processes are at Autonomy Level 2, Autopilot (as of September 2026).
What we haven't done: we haven't built the same process with fixed rules or a click robot in parallel and compared the results. That's why the comparison above is our assessment, not a test result.
Runs without me
Every two weeks, one process that runs without the owner: what came in, what the AI did, where a person decided. Two minutes to read.
Worksheet
Process Potential Check
With five questions and a first process list to fill in, you find the process that could run on its own first. Comes as a PDF by email once you confirm your sign-up.
Strategy session
In a 20-minute strategy session, you tell us where things get stuck. We'll tell you honestly whether an AI employee is a good fit for you.
Book a strategy sessionFrequently asked questions
Is intelligent process automation the same as AI?
It usually means automation that uses AI for reading and sorting and combines that with fixed rules. It isn't a fixed term with a clear boundary. For every offer, ask what exactly is meant by it.
Do I need dedicated process automation software?
Not necessarily. An AI employee works with the filing system you have today: folders, a document management system or your industry-specific software. Whether your software can be connected is something we check on the call.
Can I start with fixed rules and switch later?
Yes. A fixed rule that reliably handles a uniform workflow can stay. An AI employee comes in where input arrives without a fixed form or where the variants become too many.
What is hyperautomation?
We read it as an umbrella term that vendors use to describe combining several kinds of automation on one platform, for example fixed rules, software robots and AI. For a business with 3 to 15 employees, the more important question is which process goes first and who looks after it.
Your next step
In the end, intelligent process automation comes down to the input, the variants and who looks after it. If you want to know whether an AI employee is the right way for one of your processes, we'll look at it together on a call.