Scientists say satellite imagery and machine learning advances will boost African consumer data

Last month, the Nigeria Bureau of Statistics (NBS) published a consumer spending report which is meant to be published every three to five years. But the previous one was published 10 years ago.

Of a population of 198 million people only 22,110 households where covered. The data was collected via door-to-door surveys requiring visits to rural communities which in many parts of the country are often difficult to access and sometimes not safe for researchers unfamiliar with the terrain.

This highlights the challenges with consumer data collection and availability in many African countries, leading to surveys which don’t include enough households to generate a reliable national statistical picture. National consumer data is not regularly published and much of the available data isn’t always up-to-date. As a result, many figures used for analyses and decision-making are extrapolated estimates, often unreliable. This makes it difficult for business, government, and development agencies to make good data-based decisions on the continent.

But these challenges may soon be a thing of the past as new research from scientists says they have developed and improved computer programs (machine learning) that use publicly available satellite images (daytime, nightlight and multispectral) and available survey data to estimate household consumption and assets wealth variations in African countries.

After training computer programs with survey data from one part of a country, they were able to predict variation in household consumption for the rest of the country where they don’t have the data. They were able to predict variation down to the village or neighborhood level with up to 55% accuracy and assets wealth up to 75% accuracy.

This approach improves the quality of the consumer data as the approach can be scaled with little or no additional cost and produced in about 24 hours. The scientists say the technology can still be further improved by the integration of more imagery such as optical and radar imagery, and also with data from mobile phone and social media network.

The lack of disaggregated consumption data of African nations is a problem that can hardly be solved through traditional institutions or approach. It will cost some African country as much as $1 billion a year to upgrade their statistical systems and carry out regular data collection.

Cape Province from space

Africa is seen as the new market frontier and the continent’s consumer market is estimated to reach $2.1 trillion by 2025. Consumption behavior of households and neighborhoods, especially of the growing middle class, is of huge interest and value to businesses but there are still debates on the actual population size of the middle-class household in Africa.

US-based startup Fraym is one of the tech companies already using satellite data to develop African consumer data. The three-year old company has worked with organizations including MasterCard and the African Development Bank to provide neighborhoods and rural districts level data.

While African startups including Kasi Insight and mSurvey are already using internet, mobile phones and traditional in-person panel surveys to gather consumer data there is an opportunity to scale up their offerings with satellite-imagery-for-consumer- data approach in the continent to tap from the huge market opportunity.

“Start-ups are commercializing this technology in sectors like agriculture and construction in particular, but there are many specialized areas, like retail, traffic management, and market research, where niche operators are positioning themselves,” says Arthur Goldstuck, managing director of  Johannesburg-based World Wide Worx. “As the use of artificial intelligence becomes more accessible and cost-effective, it will drive many new opportunities.”

The sector faces challenges holding it back which ranges from government apathy and bureaucracy to lack of supporting infrastructure and low appetite from investors, cautions Goldstuck. “It is bureaucratic intransigence and efficiency rather than technical skills and technology adoption that will hold back these advances.”

Leave a Reply

Your email address will not be published. Required fields are marked *