From pilot to practise: the potential of AI in the Social Support Act

As in all other municipalities across the country, WMO consultants in Hengelo conduct kitchen table conversations with individuals who need support. These intake interviews aim to gain a comprehensive understanding of the person, identify their request for help, and determine appropriate solutions, which may include tailored provisions from the Social Support Act if necessary.

"In these discussions, we opt for a broad formulation of questions," explains WMO consultant Niels Bolks. "We aim to uncover the question behind the question. For example, if someone requests a mobility scooter, we need to understand their disability, what they can manage independently, and what provisions are necessary as a result."

According to Bolks, these conversations form the basis for file creation. "Because we strive to fully understand someone’s needs during the conversation and how the WMO can address them if applicable, creating these files is quite time-consuming."

Time-saving technology

Pia Rocks, a digital transformation driver at the municipality of Hengelo, explores how technology can assist the municipality and its residents. "During work sessions organized on digital transformation, the question arose whether we could use technology within the social domain to document the WMO kitchen table discussions," she explains.

"We proposed to the municipality to explore the use of an AI tool we developed to transcribe kitchen table conversations," adds Max van der Zwaag, manager at Deloitte. "This solution hasn't been marketed yet, but we've conducted several tests with doctors based on hypothetical patient-doctor conversations, yielding interesting results."

Rocks continues, "The way GPs converse with patients to reach a diagnosis shares similarities with how we assess support needs under the Social Support Act during kitchen table conversations. We didn’t want to create an AI tool from scratch for transcribing these conversations, but preferred to utilize an existing tool. The application we used provided a solid foundation but needed further development to align with the specific objectives of the municipality. Our idea was: if the model can accurately capture the resident's living environment and the challenges they face across various life areas, it would save WMO consultants a significant amount of administrative time following the kitchen table discussion."

Demo version

A demo version was developed in workshops alongside the users. "To achieve this, we collaborated closely to determine the process the WMO consultant and the client should follow and how this should be captured in a report through transcription," explains Rocks. Freekje Huisman, senior manager at Deloitte, adds, "We believe it is essential to involve end-users in the developments we support. That's why we always have our colleagues shadow the end-users."

This process was completed in just a few months. After four workshops, a version was ready for a pilot. "This means the WMO consultants actually started using the tool during the kitchen table discussions," says Rocks. "With the resident's permission, of course."

Bolks comments, "During the development process, we also used the tool to identify and fix any initial bugs. Subsequently, it proved to be very user-friendly: you press the button and the recording starts."

In terms of content, there was a limitation, Bolks notes. "The tool transcribes what is said. The transcription is then summarized and interpreted. For concrete issues, like mobility, the tool found it easier to describe the problem or environment than for more abstract issues—such as mental complaints and multiple problems. Therefore, one of our conclusions is that it is better to summarize the transcript literally, without making connections or drawing conclusions."

Looking to the future

The conclusion from the pilot was clear: the tool needs adaptation to be practically applicable in kitchen table conversations. "It didn't work as effectively for complex cases as it did for simple ones," Bolks emphasizes. "Overall, I still had to rely on my own notes. So it didn't save enough time."

What did this pilot achieve?

"It provided us with valuable insights into how the solution works in practice and where improvements can be made," says Van der Zwaag. "The language model behind this solution is becoming increasingly smarter, which benefits the tool. Additionally, we can continuously refine the instructions to the language model, enhancing its ability to interpret certain information. It is also possible to set how 'creative' the model is."

According to Van der Zwaag, further development is possible based on these points. This potential development is ultimately interesting for all civil society organizations. "It can save time and thus increase staffing effectiveness. But more importantly, it allows the focus to be on the conversation rather than the administration."

This article was previously published on ICT&health in June 2024