These are the questions we get most often. If yours is missing, email hej@kapaciti.se.
An AI employee holds a role in the business the same way an employee does. It works in the systems you already use, takes care of the recurring work and hands over a finished draft. Nothing sensitive leaves the building until someone on your side has said yes, and every decision is written to a log.
A chatbot answers questions. An agent carries out tasks, though the word itself says nothing about approval or accountability. An AI employee holds a role, finishes the task across several systems and comes back once for a yes. What separates them is who acts, who approves, who is accountable and what is logged. The answers to those four questions tell the products apart, not the name.
RPA follows a recorded path and stops when something looks different, because it clicks rather than reads. An assistant inside a tool helps where it lives, but it cannot see the case in your other systems. An AI employee reads the material, works across several systems and comes back to you for a yes before anything happens.
No. The connection is built during onboarding, in the systems you already use. If your system is not on the list, we build the connection.
Recurring work that has clear input and a clear result: the monthly report from the books, replies to routine email, suggested entries in the public records register, deadline monitoring and first-line customer service. It is a poorer fit where the judgement is the whole job. The examples we show are fictional and labelled as examples.
It starts with half an hour where we go through what your week looks like and what must never be touched. Then comes a shadow week, where it reads and suggests without being able to act, so you see the result in your own cases before you decide. After that it works for real inside the limits you have set, and everything that leaves the building passes a person who says yes.
You decide where your yes sits. Anything meant to leave the building is prepared, put in front of you and waits. The log shows what was proposed, who approved it and when, and it can be audited afterwards without asking us.
Accountability stays with the business, exactly as when an employee gets it wrong. That is why the approval and the log are the whole point: a yes is tied to one specific text and one specific person, and can be checked afterwards. Without a trail, accountability cannot be carried, however good the model is.
That depends on the solution you choose. If we run in the cloud, operations sit in a Swedish or Nordic region. For organisations with stricter requirements, such as banking, healthcare, the public sector and critical infrastructure, a server of your own is standard, and then no data leaves your own data centre. We write down where each part runs before you decide, so the answer holds even when someone else asks you.
It depends on which personal data is processed, which data processing agreement is in place, where the model runs and how the use is classified under the rules. An answer without those conditions is worth nothing, and no supplier can promise you compliance. ALEX is built for human oversight and for everything to be logged, and what applies in your case is checked against the legal text before you decide.
The price is set once we know what is to be done, at what scale and in which systems. We do not put a price list here, because a first flow and a programme for a whole office cannot be compared. After the mapping you get a price in writing before anything is booked.
We work in three formats: half a day with the management team on where a yes should sit, training for staff on their own cases, and open seminars for anyone who wants to see before anything is decided. It takes place at your offices or online, in Swedish or English. The angle is that staff should learn to lead an AI employee, not just to write in a box.
Management practises setting limits and the order of approval, finance people practise reading material with a source and a deviation, case officers and registrars practise judging a proposal before it is registered, and customer service practises answering in its own voice without letting anything unchecked out the door. We split the groups by what people actually do all day.
They can ask for something in a way that makes the answer useful, read a draft critically and tell the difference between an answer with a source and an answer that merely sounds right. They know what must never be entered into a tool, and they know who approves what in your organisation. The working methods we go through are left behind in writing.
We teach the working method, not a product, and we practise in the tools you already license. If the training runs in your own systems, we are careful about what may be entered and what never may. If you are going to work with ALEX, we practise in that too.
The rules require anyone who uses AI at work to understand enough to stand behind what is done, and the level depends on role and risk. Exactly what applies, and from when, is checked against the legal text before we claim anything about it, so we do not sell a course with a paragraph in hand. If you need the documentation to show what your staff know, we produce it together.
Newsletter
A summary roughly once a month. What has changed in the rules and which vendor shifts affect Swedish companies. Sent by us, not by an automation pretending to be us.