The pandemic has been tough for Eric Dossekpli. The 49-year-old farmer from Anfoin Avele, a town in the west African country of Togo, had trouble selling his peanuts, black-eyed peas, maize and cassava at the market. Customers couldn’t buy much because of their own pandemic income loss. Then he couldn’t afford fertilizer to keep growing his crops.
“I didn’t know how I was going to buy food, to buy what’s needed at home,” he says. And with four of his six children in school, he needed to pay for their tuition.
Then around October, he heard people in his community buzzing about a program: The government was giving away free money: $13 for men, $15 for women every month for 5 months (women get more because of their caregiving role). All he had to do was dial *855# to register to see if he qualified for an instant mobile payment.
“I didn’t believe it,” he says. But he gave it a try anyway — and to his surprise, “my name passed the registration and the money transferred to my phone.”
But how did the government confirm that Dossekpli needed the cash? As it turned out they couldn’t — without the aid of artificial intelligence.
In low-income countries, identifying people who have fallen on hard times due to the pandemic is no easy task. People in this economic bracket often work in the informal sector and don’t have documents to prove how much they earn. As a result, governments don’t have good data about who is poor. There are ways to find out — for example, going door-to-door and asking detailed questions about how much money a family earns — but that kind of in-person surveying is problematic in a pandemic.
Last April, Togo started a program called Novissi, or “solidarity” in the local Ewe language, to help folks who’d been pushed into poverty by the pandemic. Over one million people registered. Cina Lawson, the country’s digital transformation minister, led a team that used voter identification data to select the recipients — people who listed themselves as “informal workers,” a sign they were likely to be poor.
The program used mobile phone technology to quickly distribute $22 million in three monthly mobile phone payments to 600,000 citizens in urban parts of Togo: $20 for men and $22 for women.
Lawson and her team wanted to add rural citizens to the program — like Eric Dossekpli.
But the government didn’t have enough money to help the millions of rural residents registered as an “informal worker.” And they wanted to target the poorest informal workers in the poorest parts of the country but didn’t know which rural areas were the least well-off.
So Togo turned to artificial intelligence: a computer program that dives into data to pinpoint pockets of poverty. The government partnered with researchers at the University of California, Berkeley, and the U.S. charity GiveDirectly to use satellite imagery and mobile phone data to find citizens most in need.
This new phase of the program began in November 2020 and aims to distribute another $4 million in cash to 60,000 individuals through mobile phone payments.
Lawson’s team consulted with Esther Duflo, who in 2019 won a Nobel prize for her experimental approach to alleviating poverty — using randomized control trials to see if programs were working. Her suggestion: Get in touch with Joshua Blumenstock!
Blumenstock, an associate professor at the School of Information at Berkeley, has been researching new and different ways to measure poverty. His lab showed that computers can detect levels of wealth just by satellite imagery, and that the way people use their cell phones can be a pretty good indicator of how rich or poor they are.
GiveDirectly was willing to help implement this new methodology in Togo and distribute $4 million in funds from its donors in this second phase of the program, called Novissi GiveDirectly.
Together, they embarked on a quest to find the rural poor.
Their first challenge was to find out which villages and neighborhoods in Togo were home to residents likely to live under the poverty line — under $1.25 a day.
Blumenstock’s turned to high-resolution satellite images. There’s no exact set of factors that can determine poverty or wealth, says Blumenstock. But in general, “poorer regions have different houses, different roofing material, different quality roads, different size plots of land,” he says. “Bodies of water, like rivers, tend to be associated with wealthier regions.”
Through the computer program, he and his team were able to identify 100 of the poorest cantons with about 600,000 registered voters.
That led to another dilemma. GiveDirectly could only afford to give funds to 10% of that group. So Blumenstock tried another tactic: What if they could use a person’s mobile phone behavior to narrow down the list?
Mobile phone data can reveal a lot about income level, says Blumenstock. “Wealthier people tend to make international calls. They tend to buy airtime in larger denominations. They tend to make more calls than they receive,” he says. Poorer people, on the other hand, tend to make shorter calls and more local calls.
Blumenstock analyzed data from Togo’s two primary cell networks to identify mobile phone users with patterns of those living under the poverty line.
Merging the two data sets, the team came up with a list of 60,000 names. They rolled out a pilot in October, and by November, opened it up to the public.
And that’s how Dossekpli was able to get his cash. When he sent that text message to *855# and registered for the aid, the Novissi GiveDirectly program verified through his voter ID that he lived in one of the poorest 100 cantons — identified through the satellite imagery data — and that his cellphone number met the behavior criteria of someone living in poverty.
So far, Dossekpli has received four payments and used the funds to pay his children’s school fees. He will use his fifth and last payment in February to buy fertilizer for his crops.
Rachel Strohm is a consultant at Innovations for Poverty Action Lab (IPA) and is a Ph.D. candidate at Berkeley, where she is writing her dissertation on welfare programs in African countries. IPA did not carry out the Novissi GiveDirectly program, but Blumenstock and his team are affiliated with IPA.
She says a few things make Novissi GiveDirectly noteworthy. Because the new method uses readily available digital data, it can be deployed more quickly – than, let’s say, surveys — to find the poor in emergencies.
She adds, “This is the first program I’ve ever heard that is using mobile phone targeting as a technique,” she says.
In the next few months, GiveDirectly hopes to expand its leg of the Novissi program by distributing $10 million to 114,000 individuals, says Han Sheng Chia, special projects director of the organization.
“The context for all of this is that extreme poverty for the first time in 20 years is on the rise” due to the pandemic, he says. “More than 150 million people are about to be thrown in extreme poverty.”
For Dossekpli, the money has been a godsend. Before the program came along, he says he was pleading with the other farmers in his village to let him work on their fields.
“Now I can do what I want without begging the other farmers,” he says. “I can’t imagine how I was going to live if not for this money. All I can say is thanks.”
With translations from Floriane Acouetey in Togo.