Skip to content

Case Finnish Defence Forces

Beating ticket system challenges with the aid of machine learning

In the Finnish Defence Forces, up to 5,000 online support and service requests are made and resolved every month using a ticket-based system. However, service requests do not always get sent to the correct address. With the aid of funding from the Ministry of Finance, Digia and the Finnish Defence Forces Service Centre implemented a pilot in which machine learning and text analysis were used to automatically resolve and categorise tickets.

 

Large organisations have a great number of both users and IT systems, and a variety of problems and faults will often occur. In the Finnish Defence Forces, up to 5,000 online support and service requests are made and resolved every month using a ticket-based system. Most of the people who make service requests use the ticket system rarely, and therefore service requests do not always get sent to the correct address. This generates additional work and delays in the service process.

The Finnish Defence Forces Service Centre has long been attempting to harness machine learning to solve this problem. An opportunity arose in autumn 2018, when the Ministry of Finance opened a call for funding applications that was aimed at encouraging government agencies to streamline and automate their services.

“We had already been playing with the idea of harnessing machine learning in service requests. After the Ministry of Finance’s call for applications, we began clarifying this idea and received funding for a project that sought to reduce administrative workload with the aid of artificial intelligence and text analysis,” says Jaakko Vesanen from the Finnish Defence Forces Service Centre.

In spring 2019, this resulted in a proof of concept (POC) in which the user describes their support or service request to a chatbot linked to a Skype channel. The chatbot helps users to direct their service request to the correct place.

The pilot gave us good reasons to continue developing chatbots. We received plenty of input for our other budding chatbot projects, as well as new understanding and experiences that we’ll be sharing internally.

Jaakko Vesanen, Service CEnter, the Finnish Defence Forces

Chatbot harnesses existing components to boost service efficiency

The concept was initially developed at weekly workshops that used discussion and iteration to crystallise ideas into a final concept. The Defence Forces were able to start working on the proof of concept in agile collaboration with a familiar partner: Digia.

“Our initial challenges included conceptualising the new process and then considering how we could use text analysis and machine learning to our best advantage.” This was no traditional chatbot – we wanted more behind it,” says Vesanen.

Usability lies at the heart of the planned chatbot: when users detect a problem, instead of searching for the right report form, they will be able to freely describe the problem to the bot. With the aid of text analysis and machine learning, the chatbot will search a database for the correct solution to the problem, and then suggest it to the user. Thanks to its Q&A-style approach, the chatbot is able to solve some problems directly, which will reduce the number of tickets.

If a problem is not solved, the chatbot’s machine learning model will look for the correct service category and ask the questions required to create a ticket. The ticket will then be sent for resolution. The chatbot was constructed using components from the Azure cloud service, which the Finnish Defence Forces were already using.

Their close cooperation with Digia was smooth and pleasant.

“The agreed tasks were completed on schedule, resulting in a demo that was implemented on time and within budget,” says Vesanen.

So will the chatbot be sticking around?

“The pilot gave us good reasons to continue developing chatbots. We received plenty of input for our other budding chatbot projects, as well as new understanding and experiences that we’ll be sharing internally,” says Vesanen.