(Keva headquarters in Helsinki. Source: Keva.)
Keva (Helsinki), Finland’s largest pension provider, faces a major demographic challenge: roughly one-third of public sector workers exit employment before reaching retirement age due to disability. “One third of the public sector workers currently leave the workforce before the old age pension,” says Petra Sohlman, Data Scientist, Keva. “And we need everyone, since our old-age dependency ratio is the highest in the EU at the moment. It’s getting worse, so we need everyone to work.”
From Registers to Unstructured Data
Traditional models at Keva rely on demographic and sickness absence data. “They are interesting, and the models work,” Sohlman comments. “But I’d like to add those factors that pertain to a person’s health and motivation. Once somebody starts to think that they can’t work anymore, the ship sort of turns—and it’s very hard to turn back.”
Petra Sohlman, Data Scientist, Keva.
Doctors’ notes, which accompany pension applications, are one of the most promising sources. “The doctor is required to fill in all the details that pertain to the person’s health, and then our specialists comment on the applications,” she explains. “That gives us two layers of text we can analyze and see how they correlate with actual disability.”
SAS Viya Pilot for Text Analytics
Keva has tested open-source Python tools in a master’s thesis project, but recently ran a pilot with SAS Viya. “In our pilot project we were able to see how much the prediction power of our models was enhanced by extracting risk factors from text,” Sohlman says.
While Keva is not yet a Viya customer, the pilot has made a strong impression. “We are still using SAS Enterprise Guide, but I think the transition to Viya is in the future. We have many coders who are happy with what they have, so it’s a ship that turns slowly. But this text analysis is something we would really like to have.”
Turning Insight Into Action
Insights from text analysis are already being shaped into practical employer tools.
- Risk dashboards show where depression or alcohol misuse cases are peaking.
- Cost models quantify how sick leave or disability risk translates into financial losses.
“We can provide public sector employers with tools that tell the situation,” says Sohlman. “For example, in one geographical area we saw a sudden peak in depression cases, and in another we saw alcohol problems and depression linked very powerfully. With that information, employers can better target their early interventions. And when they see the costs of absence and disability quantified, they can make a stronger case for investing in prevention.”
For Sohlman, the long-term value of text-based AI is in surfacing the human signals that are otherwise invisible. “This kind of text analysis can potentially give us a look into aspects of health and motivation that aren’t captured elsewhere,” she says. “That’s where I think the next breakthroughs will come.”