September 15, 2020


Law-reading AI tips caseworkers on the optimal work process

Legislative clauses and IT systems set the framework for the work of many public employees, and often rigid flow diagrams put a stop to finding new and perhaps more efficient ways of solving a task.

But in the municipalities of Aalborg and Syddjurs, some departments use a tool that guides employees through the most optimal workflow that meets all legal requirements. They design the model themselves, and while employees navigate around the work tasks, AI intercepts data about the work process to map which paths give the best results.

"With a tool that reads and understands Danish legal sections, employees in the administration can design their own business process and be guided through the tasks in an interactive way that complies with the law."


Louise Holst Andersen

»We 'map' how employees work. How do they get from start to finish? This is the data on which we train the artificial intelligence, «says director Morten Marquard from DCR Solutions, the company behind the process optimization tool.

Morten Marquard explains that the basic thing is to make the system so that you are compliant by design. To make it so we are sure it complies with the rules, he says. But it is also about making a tacit knowledge visible.

‘Our AI can find some connections and it can be used to make things more productive. If an employee is sitting with a special case, then they know very well: 'I have to give this case to Peter - it's a little special, and he's better at it.' These are the patterns we can find and visualize, and then we can help them put it into words so that such cases can always get over there, "says Morten Marquard.

Not 'to do' - but 'ta da'

In the past, caseworkers in the two municipalities may have followed a long list without the possibility of deviating from the beaten path, but the tool DCR is rule-based, and this means that you can achieve the goal in several different ways, as long as the legal requirements are met, says Morten Marquard .

The engine behind the DCR technology is built in the functional language F #, while the tools themselves are written with a combination of the React environment and .Net.

“The very basic thing is that you get a to-do list. You will be told what to remember to do. One of the caseworkers in Syddjurs has said that it is not a 'to do' - it is a 'ta da'. That is because it distinguishes between tasks that you have to do and tasks that you can do, ”he says.

If a number of specialists have to perform tasks in a special IT system, an employee can describe the business process in a model using DCR's tools without having to use code himself. The end tool itself is a cloud solution with a portal interface, where the lawyer or office worker uses the relevant legal texts to create their model.

The three basic things the employee needs to make the model are the roles - who is involved in the work - the activities - what the caseworkers must do - and the rules of law - what conditions determine the process in different tasks.

»We have a tool in the cloud where you put the legal text in, and then you 'map' it over. Then you have the model, and then the IT systems themselves can pull it over because the model is digital. So the IT systems themselves snatch the model from our portal, «he continues and adds that some IT systems require that you copy the file with the model above.

Completely impractical, the caseworker's screen shows both a case about a citizen in a given IT system - and at the same time the screen shows the interactive task list from DCR, says Morten Marquard.

Once the model has been copied over and is an integral part of the IT system that caseworkers use, it can begin to collect data and train itself in identifying patterns in how employees work - and what is most effective.

»We make connections visible. It's based on data - what actually happened, not what we think or feel. It's not because the computer is necessarily right. But then an experienced employee can say: 'Yes, that's a very good point, but it just does not fit here'. "

Too far from description to implementation

DCR Solutions has developed the solution based on several years of research collaboration with several universities.

“We once had to digitize some processes for KL, which they had drawn in BPMN, and it was about writing children up for crèches. We got the model and we had three different interpretations of what was in the model. There I thought: There is something completely wrong here, "says Morten Marquard.

“If we humans are in doubt about the requirements specification, then I dare not even think about what happens if you put it into a computer. And that's where it started: Can we make some formal models that can actually describe what the work is? ”

The first system went into production in December 2013 and is still running today, says Morten Marquard. He founded the company DCR Solutions in 2018, and today they have 11 employees on the team, and four of them are researchers.

“It's not because IT systems or programmers are too bad. It is simply because there is too far from what we describe to what we implement. That's where it short-circuits. "

Morten Marquard calls the solution a "robot on steroids".

»Usually a robot can vacuum or make coffee. It can do one thing. But this technology can be used for really, really many things, "he says.

Still, there are areas where he would wish the solution was better. He is proud that the technology gives ownership of the process to the users themselves - they design the digitization themselves. But it is also a weakness, he says, because even though the team behind DCR sees the models as very simple, it is not always so straightforward for people who are not IT savvy.

“It's really very simple, but that simplicity does not capture people. We are not good enough to communicate it out. People say: Why is there more than one arrow? It is because there is something that goes backwards in time and something that goes forwards in time. And we are still fighting with that. "