Canada’s immigrant settlement service system follows a centralized model. This means people seeking settlement support —such as for employment, language, or counselling— are asked a standard set of questions, and their answers are recorded in a centralized database. Ideally, this helps settlement workers connect people to the services they’re eligible for, keep track of who’s receiving support, and equip policymakers with data to assess impact and improve future settlement programs.

Canada has even developed its own tool, iCARE (Immigration Contribution Agreement Reporting Environment), to assess immigrants’ needs and document their journeys through their settlement. But while this approach brings consistency, it also introduces challenges.

I identified three main goals that mediate data practices in settlement sector in Canada:

1. Connecting immigrants to services: Settlement workers gather data to refer immigrants to the right support.
2. Respecting immigrants and their data: Immigrants have the right to decide what they share. Settlement workers often face the challenge of recording information accurately during time-pressured conversations with immigrants, while also trying to provide quality support.
3. Reporting service data: Settlement workers are responsible for the routine reporting of service data to the federal government, frequently within tight timeframes.  
   

These goals often clash. For instance, immigrants are asked the same personal questions repeatedly—sometimes multiple times in a single day—because different service providers can’t see each other’s data. Or, if different funders support the service, workers may need to enter data in multiple ways to meet each funder’s requirements. This repetition, while meant to respect privacy, can frustrate immigrants and make them less willing to share. At times, tensions emerge between organizational data reporting requirements and the relational work needed to build long-term, trust-based connections with immigrants.  

People, Not Just Data Points

A key insight from my study is that immigrant data isn’t fixed—it changes over time as people build trust with settlement workers and feel more comfortable opening up. This runs counter to the assumptions behind many AI systems, which often treat data as stable and objective. The reality is more fluid: some immigrants may feel ready to share their full story early on, while others need time and trust, which rarely develops during a hurried initial intake with a frontline worker. 

Canada’s system embraces this by treating settlement support as a continuous process, rather than a one-time quiz. Settlement workers often note that they collect better information through relationship-building. Still, I found that not all settlement organizations are equally equipped to support this model. Variations in organizational culture, the urban–rural divide, workers’ digital literacy, access to reporting technologies, and personnel availability all influence what data is collected and how it is recorded. At some organizations, settlement workers barely have time to talk to immigrants beyond the intake. In others, workers can build deeper relationships that allow for more accurate and complete records.  

Why This Matters for AI and Immigrant Settlement?

Canada’s example challenges the idea that more data automatically leads to better outcomes. In fact, using inconsistent or incomplete data to train AI systems can backfire – leading to errors and biases compound over time. If settlement workers record different information about the same immigrant, which version should an algorithm trust? 

I argue that any governance of AI or data-driven technologies in settlement services must be grounded in a deep understanding of the contexts and practices through which data is produced. This means involving frontline workers and immigrants —who often don’t know how their data will be used—in the design and evaluation of new technologies. It also means questioning whether certain tools, especially those focused on service allocation, are appropriate at all.     

Importantly, the study also offers a hopeful vision: by centering refugees’ and immigrants’ lived experiences, social systems can leverage data to support—rather than replace—the judgment of frontline workers. Canadian model shows that it’s possible to design data practices that reflect the realities of both immigrants and settlement workers. 

Rethinking Success in Immigrant Support Services

Ultimately, this study pushes us to ask: what does success look like in data-driven settlement support services? For AI developers, it might mean improving prediction accuracy. For policymakers, this might mean immigrants integrating well into Canadian society, securing employment, and making contributions. But for frontline workers, success often looks like something else entirely: a refugee who comes back, opens up, and finds a moment of safety in a system that too often fails them.  

By recognizing these different perspectives – and designing tools that support them – we can build more ethical, equitable, and sustainable immigrant support systems. As I argue, data-driven interventions must always be grounded in context, shaped by those who use and are impacted by them, and evaluated by the long-term support they enable.