How Can New AI Tools Improve Integration and Resettlement Outcomes for Migrants Globally?

Michael Hotard has some idea of how government decision-making works. Roughly ten years ago, he spent time on the inside of the Employment and Training Administration at the U.S. Department of Labor, where he oversaw the management of government grants to nonprofits providing workers with training and employment opportunities. 

It was enjoyable work, but there was also a sense of frustration.

“I didn’t think we were pursuing the best strategies for helping workers. In fact, I didn't think we were measuring the impact of our work in a way where we could even learn what those strategies should be.”

This hankering to pursue initiatives grounded in research and data brought Hotard to Stanford’s Immigration Policy Lab (IPL). Here he joined an array of academics – from social scientists to data scientists – united by one mission: to improve the lives of immigrants with hard evidence and innovation.

“IPL is really a group committed to working outside of academia with governments and practitioners to figure out what questions matter to them, what solutions are possible, and whether they actually make a difference.”

The lab opts out of academic isolation and instead works in deep partnership with practitioners across the globe on the frontlines of challenges like resettlement. From the U.S., Canada, and Europe, to the Middle East, Africa, and beyond, IPL’s objective is to develop programs that improve the lives of refugees, immigrants, and communities – and ultimately act as a model for governments and agencies worldwide.

GeoMatch-making for better lives

Hotard is now a Director of one such IPL program: GeoMatch, which received a 2021 Stage 2 investment from Stanford Impact Labs. At its core, GeoMatch is an algorithmic tool that provides data-driven recommendations for the settlement of immigrants and refugees in locales where they can thrive.

The idea had its genesis in a gathering IPL held in 2016 with the 9 of the ten large U.S. nonprofit organizations that, through contracts with the U.S. State Department, are responsible for refugee resettlement. Here it emerged that, for the 30% of refugees with no existing ties to the U.S., the decision-making that guided their placement was largely based on staff experience combined with some qualitative evidence. In other words, enormously significant decisions for refugees and communities alike were being made with limited data on how to estimate the likelihood for success.

Despite these agencies conscientiously tracking relevant success metrics, that data simply wasn’t being fully utilized.

“We started thinking with them,” says Hotard. “Would there be a way of using all that existing data to provide insights into where people could find success? And that’s where GeoMatch came from.”

Location is critical

When it comes to resettlement, the importance of placement should not be underestimated. There are over 35 million refugees globally. For the small proportion who are fortunate enough to be resettled in a third country, the road to successful, profitable integration is a challenging one.  Mostare relocated with no assets, receive scant (if any) local support, and often face the challenge of learning a new language. 

Where refugees make a home is a critical factor when it comes to success or failure – not just for the individuals themselves, but also for the receiving local communities and economies.

Working within a limited network of locations, as well as with prescriptive quotas, resettlement agency workers are faced with the high stakes task of finding relevant connections between immigrants and their prospective new locations.

Enter GeoMatch

To assist with that endeavor, the GeoMatch tool deploys an AI algorithm that dynamically learns important underlying synergies over time – including those that aren’t immediately obvious to a caseworker.

Which relevant skill sets are migrants bringing that may be a good match for local industries? In which locations is it important to have a strong grasp of English? How might a refugee's country of origin affect their likelihood of finding a job? And most importantly, how do all of these factors interact with one another in ways that would be unobservable without using machine learning techniques.

"GeoMatch is able to take in vast quantities of data and uncover patterns that would be impossible for the human mind to detect,” says Hotard.

The system can identify correlations in the rich historical data and make personalized suggestions for each newly arrived refugee, while also taking into account a variety of required constraints, quota considerations, and available services.

The GeoMatch recommendation comes in the form of an “impact score”; a ranking which conveys the likelihood of success at each location, with a higher score reflecting a better chance of success. From there, the human decision-maker who receives the recommendation can then introduce their own background information, using their experience and discretion to make the final decision – ensuring there is always a “human in-the-loop.” 

GeoMatch helps to reduce the kind of trial and error that can cause already troubled migrants to stumble upon arrival in their new home. When the algorithm has been tested using refugee applications and data from previous years the results suggested that those candidates placed by the algorithm would be between 40-70% more likely to find work, which can be a crucial factor for integration.

Unexpected benefits 

Thus far, the reception from the agencies deploying GeoMatch has been overwhelmingly positive. In the U.S., the tool has been deployed with Global Refuge (formerly the Lutheran Immigration and Refugee Service), and the team at IPL has begun early-stage data sharing and process mapping with three of the other largest resettlement agencies that were present at the original gathering in 2016. 

“I’m able to get everything I need in one dashboard…this is The Jetsons, compared to The Flintstones,” said Helen Purcel, a 22-year veteran of refugee resettlement, currently working for Global Refuge.

What makes GeoMatch a hit with users like Helen is that the system doesn’t seek to reinvent the existing resettlement decision-making process, but rather to enhance it with the algorithm. To do this, the GeoMatch team has dedicated time and energy to precisely documenting and modeling the processes resettlement experts have developed over many years – which often include a lot of intuitive knowledge – in order to build AI that replicates and improves upon them.

Hotard is emphatic: “We didn’t want something that would radically change resettlement agencies’ system. We wanted to mirror their system with enhancements. We spent a lot of time working with them to build a software tool that does that.”

The result is that processes that used to take as long as 8 hours can now be completed in just 30 minutes, freeing up staff time to focus on other important work. And perhaps even more encouragingly, employees at resettlement agencies have reported feeling more confident in their decision-making.

In their attempt to build a tool that improves the lives of migrants, the GeoMatch team has also inadvertently created a new standard for resettlement agencies and enhanced the working lives of those that run them.

Expanding globally

The impact of GeoMatch hasn’t been confined to just the U.S. The AI system has been used in Switzerland since 2020, where it helps balance allocations of asylum seekers across the country while optimizing three-year employment outcomes. It does this by sending newcomers to places with the best prospects for economic opportunity.

“In Switzerland, the tool is able to help meet the government's goal of fairly allocating cases across the entire country while also boosting economic outcomes for asylum seekers,” notes Hotard.

In each context GeoMatch also considers different, potentially predictive features of newly arriving asylum seekers or refugees. These can include age, country of origin, language abilities, work history, and family size. Working closely with partners, they agree on the outcomes they want to target for improvement and build around that.

The team is also embarking upon projects in The Netherlands and Canada. The latter initiative has an exciting new spin: putting the tool directly into the hands of economic migrants and allowing them to feed in their own preferences about their ideal community.

A brighter future

With the number of cases of displaced people at an all-time high (and only increasing), the GeoMatch team believes that what it has built is an eminently scalable tool that can help governments and multilateral agencies looking for solutions. So far, it has made recommendations for over 9,000 refugees and asylum seekers across the U.S. and Swiss contexts where it has been deployed. 

The tool’s design allows it to be adapted to new contexts easily, and Hotard believes that new iterations – like the customer-facing project in Canada – could one day be adopted in the U.S. to help migrants better understand the full picture of opportunity available to them.

He also thinks that GeoMatch could help communities better understand the economic benefits of immigration.

“More and more people are asking how we can use immigration to help communities. A tool like GeoMatch can also provide matches that have real benefits for the receiving countries. Finding a two-way match is an important opportunity.”