Försäkringskassan Deploys Neo4j to Automate Benefits Payments, Prevent Fraud, and Improve Life for Swedish Citizens
Neo4j worked with Sweden’s largest government agency to streamline data for millions of social security requests—leading to faster approvals for claimants, money saved for the agency, and payments distributed to the Swedish citizens who need them most.
- Neo4j compiles data logs in minutes, enabling Försäkringskassan to meet GDPR requirements
- Cost reduction in storage from SEK 20M to < 200k per year
- Försäkringskassan now has the ability to standardize and productize reporting
- Over 700 employees use Neo4j
“Our duty is to the Swedish people, to make sure they can live the best lives possible,” says Försäkringskassan IT Architect Frode Randers. Försäkringskassan is Sweden’s national social insurance agency, which provides financial assistance during pivotal life moments. Those caring for a newborn, navigating illness, or managing care for individuals with disabilities, for example, can breathe easier when granted some financial security.
The government agency allocates SEK 230 billion (USD 22.5 billion) in payments each year—an amount that accounts for approximately one-fifth of the country’s annual budget—managed by a team of 14,000 employees who make some 21 million decisions annually regarding compensation. These decisions must be correct and made quickly, as Sweden’s 10.4 million citizens depend on this social safety net.
Managing such a critical system for an entire country means that Randers’ team continually looks for ways to better serve the people of Sweden and ensure that Försäkringskassan’s IT infrastructure is as innovative and world-class as the services the agency provides. The team sought to expedite the process for claim approvals and fraud detection but needed state-of-the-art technology flexible enough to accommodate regular updates — particularly following election cycles, which can bring changes to policy and regulations that impact every aspect of this social safety net, from immigration to requirements for payouts.
Randers and his team knew that complex connected data challenges are well-suited to graphs and that moving off its relational database would allow Försäkringskassan to draw the kind of meaningful, dynamic insights it needs for the future. The IT team heard about Neo4j as early as 2008, just a year after Neo4j’s founding, so when the time was right, Randers says, “the choice felt natural.”
A Backend Overhaul: Paving the Way for Faster Claim Approvals
“If you have a good use case for a graph database, it makes change easy and the cost low. If instead you’ve invested in a relational database, you have a problem,” Randers explains.
“Our historical approach to making sense of all this data — and thus making compensation decisions — wasn’t optimized. It was slow, which meant slower approvals for claimants.”
On the front end, the process for Swedish citizens to apply for social insurance is simple, but data processing dragged on the backend — nowhere near as dynamic as it needs to be, especially for applying automation. “Our biggest time suck by far is gleaning insight from the vast swathes of data we collect—insight we need to approve claims,” says Randers.
Even with an IT team of approximately 660 people logging over 900,000 development hours and striving to improve the integrity of the systems they manage, their relational database wasn’t performant.
Consolidating Data into the Graph for a Holistic View of Claimants, Faster Approvals, and Better Fraud Detection
Försäkringskassan is a conglomeration of as many as 20 different historic agencies and IT networks. The first step to automating fraud detection and easy, positive payment decisions, was to work with Neo4j to map complex network dependencies to demonstrate, for example, how a payment decision in one network affects the outcome in another.
“Because we were unifying so many agencies into one, we had pools of data lying around that couldn’t be analyzed in unison,” explains Randers.
This was an IT-focused project, but one that had a significant business impact. Using the graph to connect disparate data provides Försakringskassan with an organized, holistic view of citizen claim data. The graph database functions as a consolidation layer, and the insight it provides leaves less room for error and leaves Försäkringskassan less vulnerable to fraud, as the claims that need further review are much easier to flag.
This optimizes both the staff’s time and leads to faster approvals for valid claimants, translating to money in their bank accounts and greater peace of mind during pivotal life moments.
Automating Medical Statement Reviews with Neo4j and Machine Learning
Försäkringskassan also built a machine learning (ML) pipeline using Neo4j and PyTorch. The model is trained on good medical and claim data and then used to assist in decision-making.
Medical claims are not always straightforward. Some cases require meticulous review, slowing down the approval process for everyone. A claimant requesting medical leave or applying for permanent retirement might list any number of ailments — though not all conditions are considered equal.
Previously, agency administrators evaluated paper medical statements and manually hand-coded the doctor’s statement with an ICF (International Classification of Functioning, Disability, and Health) code to categorize the problem, be it a stroke or the onset of dementia or poor eyesight. This was hard work and left a lot of room for human error, which “discouraged new colleagues from joining and also encouraged existing staff to leave,” says Randers.
The evaluation process was a good use case for automation, and Randers’ team decided to deploy Neo4j to create a proof of concept, with the idea that this AI model would later be replicated for other applications and use cases. The automation is built on top of Neo4j, and the ML pipeline is managed within it.
To do this, domain experts mapped the doctor statements and ICF codes into a graph, then worked to validate accuracy and raise the quality of response within the training data. The end result is powerful — now, when reviewing a medical claim, administrators can confidently query the system’s AI for a list of accurate suggestions.
The new Neo4j-powered system is used by more than 700 people at Försäkringskassan.
“A model like this only works when the data needed to train it is stored in such a way that relationships between data points can be established,” says Randers. “Doing that in a relational database is costly, time-consuming, and inefficient.”
Neo4j Compiles Data Logs in Minutes, Enabling Försäkringskassan to Meet GDPR Requirements
The results of adopting Neo4j are wide-ranging for Försäkringskassan — from time saved through automated approvals to reduced staff turnover to improved fraud detection. All of these equal dollars saved. Just the cost reduction in storage alone has reduced from SEK 20 million to less than 200,000 per year.
The team also realized that by using Neo4j to power their ML pipeline, they gain better oversight into who has accessed the claimant’s data—a crucial General Data Protection Regulation (GDPR) requirement.
“The new model allows us to understand and describe our data so much better, both historical and new,” he says. “We can trace the data we’ve used and how it was approved because the whole process of managing training data is captured, recorded, and easy to access. It makes our work so much easier because that regulatory obstacle is taken care of.”
Försäkringskassan achieved better oversight over claimant data access by choosing Neo4j to power its machine learning (ML) pipeline.
Data logs, also a GDPR requirement, that used to take weeks to compile are now completed in minutes. “We now have the ability to standardize and productize reporting—before graph, that was a pipe dream,” says Randers.
“Because of the system we developed with Neo4j, and how closely we worked to launch it, we’ve been able to rebrand ourselves as a data-driven agency. We could not be more excited for the many other efficiencies we’ll unlock with the system.”
Get in Touch
Curious about what insights you could unlock for your business with graph-powered solutions? Let’s talk – reach out, and we’ll get in touch.
