How AI helped provide cash assistance to many of Togo’s poorest people

Big Idea

Governments and humanitarian groups can use machine learning algorithms and mobile data to provide assistance to those most in need during humanitarian crises, we find in new research.

As the COVID-19 pandemic spread in early 2020, our research team helped Togo’s Ministry of Digital Economy and GiveDirectly, a non-profit organization that sends money to people living in poverty, turn this vision into a new type of assistance program. approach, as we explained in the magazine nature On March 16, 2022, is that the rich use phones differently than the poor. Their phone calls and text messages follow different patterns, and they use different data plans, for example. Machine learning algorithms—great pattern recognition tools—can be trained to recognize these differences and infer whether a mobile subscriber is rich or poor.

First, we collected recent, reliable and representative data. Working on the ground with partners in Togo, we conducted 15,000 phone surveys to collect information on the living conditions of each household. After matching survey responses with data from mobile phone companies, we trained machine learning algorithms to recognize phone usage patterns that were characteristic of people living on less than $1.25 a day.

The next challenge was to see if a system based on machine learning and phone data would be effective in getting money to the poorest people in the country. Our assessment indicated that this new approach worked better than other options that the Togolese government was considering.

For example, focusing entirely on poorer cantons – which are similar to US provinces – would have brought benefits to only 33% of people living on less than US$1.25 a day. By contrast, the machine learning approach targeted 47% of this population.

We then partnered with the government of Togo, GiveDirectly and community leaders to design and pilot a cash transfer program based on this technology. In November 2020, the first beneficiaries were registered and paid. To date, the program has provided nearly $10 million to some 137,000 of the country’s poorest citizens.

why does it matter

Our work shows that data collected by mobile phone companies – when analyzed using machine learning technology – can help target assistance to those who need it most.

Even before the pandemic, more than half of the West African nation’s 8.6 million people lived below the international poverty line. As COVID-19 slows economic activity further, our surveys indicated that 54% of all Togolese residents are forced to skip meals each week.

The situation in Togo was not unique. The downturn caused by the COVID-19 pandemic has pushed millions of people into extreme poverty. In response, governments and charities have launched several thousand new aid programs, providing benefits to more than 1.5 billion people and families around the world.

But in the midst of the humanitarian crisis, governments are struggling to figure out who needs help most urgently. Under ideal conditions, these decisions could be based on comprehensive household surveys. But there was no way to gather this information in the midst of a pandemic.

Our work helps illustrate how new sources of big data — such as information from satellite and mobile phone networks — can make it possible to target assistance in crisis conditions when more traditional sources of data are not available.

What then

We are conducting follow-up research to assess the impact of cash transfers on beneficiaries. Previous findings suggest that cash transfers can help increase food security and improve psychological well-being in normal times. We are assessing whether that assistance has similar results during the crisis.

It is also necessary to find ways to sign up and pay people without phones. In Togo, nearly 85% of households had at least one phone, and phones are often shared within families and communities. However, it is not clear how many people in Togo needed and did not receive humanitarian assistance because they did not have access to a mobile device.

In the future, systems that combine new methods that take advantage of machine learning and big data with traditional survey-based methods are bound to improve targeting of humanitarian aid.

Emily Aiken, PhD student in information, University of California, Berkeley, and Joshua Blumenstock, associate professor of information; Co-Director of the Center for Effective Global Action, University of California, Berkeley