Process automation with AI
Process automation with AI, and your people decide.
Much of the work in a process is reading, researching, drafting and checking. An AI colleague can prepare that work. We set it up so that your people decide at fixed points, and who decided what is recorded.
ForDirectors and operations managers of mid-sized organisations and professional services firms.
What goes where
Not every step is work for AI.
When you automate a process, you look at what happens in each step. We distinguish three kinds of steps. Each calls for a different tool.
A fixed rule
Same input, same outcome: saving a file, sending a reminder on a set day. Ordinary workflow tools and RPA are cheaper and easier to test there than AI.
Judgement on messy input
Reading a case file, organising documents, writing up findings, drafting. This is where an AI colleague is strong: it reads and weighs what the documents say.
A decision with consequences
Giving advice, rejecting an application, sending something out. That stays with a person, with the prepared work on the table.
Rule of thumb: if you can write the decision rule on one sheet of paper, that step does not need AI.
How it works
Prepared by AI, decided by people.
On the Prudai platform, AI colleagues work through the steps of a workflow and stop where a person needs to look. This is what the pattern looks like in LEO, our platform for legal work.
- 01
A person starts the work
An employee confirms the assignment and chooses the documents. The AI cannot start a workflow on its own.
- 02
A plan first
An AI colleague turns the assignment into a short work plan: which questions, which documents, what is missing.
- 03
An approval point
You choose: approve, revise with an explanation, or stop. In most workflows, the work waits for that decision.
- 04
Specialists and a counterpart
Specialised AI colleagues then each work on a part. In many workflows, a counterpart role reviews the work from the other side’s point of view.
- 05
Review and delivery
The work is weighed and brought together into a draft. You review that draft.
This is the pattern. Not every workflow has every step.
How it works in our products
Four examples, one way of working.
The same principle runs through several of our products: the system prepares, a person decides.
LEO
Legal case files
Workflows for liability analysis and building a statement of damages, among others. In most workflows, the lawyer decides after the plan whether the work continues.
LEO use cases (Dutch)IRMA
Processes and risks
In the process editor you draw a process with steps and control points. IRMA can propose a risk or measure per step, with a rationale and a source. Only after an employee approves it does it enter the register.
About IRMAVERA
Construction files
VERA organises the Wkb file, points out what is still missing and can prepare a draft. The quality assurer issues the declaration themselves.
About VERAZIA
Care reporting
With the Nedap ONS connection, ZIA can write back a daily report. ZIA shows the draft and always asks for your confirmation first. The connection is off by default and needs a separate write permission from Nedap.
About ZIA
How to start
Start with one process, not with a tool.
One first process is enough to learn where AI helps and where it does not. This is how we go about it.
- 01
Map the process
Together we map the process: which steps, who does what, where the reading and research work sits, and where a person has to decide.
- 02
Capture the knowledge
In short working sessions, your specialists write down what they know. They review it before it goes into the knowledge base.
- 03
Configure on the platform
We configure an AI colleague for those steps, with approval points where the direction of the work is set or where it leaves the building. The platform already exists: configuration, not a development project.
- 04
Train and adjust
Your people learn to work with the AI colleague. After that, we adjust together.
Frequently asked questions.
What is process automation with AI?
AI colleagues carry out the steps of a business process that require reading, organising and drafting. Fixed rules stay with ordinary automation, and a person decides where there are consequences.
How is this different from RPA, Power Automate, Zapier or n8n?
Those tools carry out fixed rules: same input, same outcome. For steps without judgement they remain the best choice. AI adds something where the input is messy, such as free text and documents. In one process they can work side by side.
Does the AI make decisions on its own?
In a workflow, the work only starts once an employee confirms the assignment. In most LEO workflows, the AI then waits for approval of the plan. In a few fact-finding workflows, the work continues without a response after a set period; that is fixed in advance.
Which processes are suitable?
Processes with a lot of reading and research and a clear end product: screening case files, organising documents, finding what is missing, drafting. Steps that can be captured entirely in rules are better automated without AI.
How do we start?
With one process. In an AI workshop we explore the opportunities and bottlenecks; a business case follows. You decide on the next step after each one.
Where is our data?
Customer data is stored on our own servers in EU data centres, in Germany and Finland. For AI processing we work with selected subprocessors. You decide which documents and systems are connected. Details are on trust.prudai.com.
Which process in your organisation runs on reading?
That is where the conversation starts: one process, the steps in it, and the moments at which your people want to decide.
Schedule a conversation