3 lessons from running an AI-powered start-up in Africa

In Africa, like everywhere else in the world, artificial intelligence (AI) is moving up the agenda as companies, entrepreneurs and governments work out how to keep pace with the Fourth Industrial Revolution. While the continent has a long way to go when it comes to AI adoption, these technologies already play a prominent role in many individual organizations: Nigerian mobile-lending platform Carbon uses machine learning to evaluate credit applications, South African fashion retailers rely on algorithms to predict the next season’s top sellers and Kenyan ride-hailing app Little has implemented AI to assess driver performance.

For the continent to remain relevant on the global stage, it is not only vital that companies embrace AI, but also that local entrepreneurs have equity in these technologies. That said, building an AI-powered start-up in Africa comes with a unique set of challenges not experienced by entrepreneurs in Silicon Valley, particularly in terms of raising capital, human resources and market receptiveness.

Entrepreneur Vian Chinner has first-hand experience of both worlds. Having founded, and sold, a well-funded start-up that applied machine learning to the rental real estate market in the US, he is now CEO of South Africa-based Xineoh, which uses AI to predict consumer behavior.

Here are three lessons running an AI start-up on the African continent has taught him:

“In a single morning in North America, more VC funding is raised than in an entire year in South Africa,” he says. According to the Southern African Venture Capital and Private Equity Association, the region’s VC industry made investments to the tune of $77 million (converted from 1.16 billion South African Rand) in 2017 while KPMG puts US VC deals at $84.24 billion for the same period – an average of $115 million per morning.

In Chinner’s experience, South African VCs, with a few exceptions, are much more risk averse than their Silicon Valley counterparts. Whereas US-based VCs are generally willing to take a bet on a high-innovation/high-risk idea; South African investors typically avoid companies that don’t have a proven cash flow and solid traction.

“In Silicon Valley you basically need to sell the direction in which you are going. There’s an understanding that a start-up won’t have an exact business plan until it has launched its product in the market,” says Chinner.

“Once the start-up has received initial feedback from customers, it will start iterating to create a product-market fit. US VCs put a very high premium on innovative products and ideas, but attracting investment based on a concept or idea is tough in South Africa.”

Another challenge faced by start-ups in South Africa is that VCs often don’t have the in-house expertise to adequately evaluate AI solutions. According to Chinner, most South African VCs have a banking background, unlike Silicon Valley where many investors are former techies.

Looking abroad for funding is an option but it’s not always an easy route. While the economic and political situation in many African countries may make some VCs nervous, distance is an even greater factor – early stage investors tend to prefer start-ups that are based close to them. Chinner says: “A San Francisco-based VC would be skittish about investing in a start-up in Phoenix or New York. People generally prefer to back teams based in the same city as them. They want to keep a close eye on them and be able to check in once a week.”

Working at a start-up on the US’ West Coast allowed Chinner to build a solid contact book, which helped with Xineoh’s fundraising efforts. To fund the company, he approached Canadian investors who were willing to back him at a valuation almost 10 times higher than what he could have raised in South Africa at the time.

Without such a network, Chinner believes Africa-based entrepreneurs will find it much tougher to attract investment from US-based investors.

Super-smart data scientists are critical for the success of AI companies; yet it is a skill that is in short supply on the African continent.

As Wim Delva, acting director of Stellenbosch University’s School for Data Science and Computational Thinking, writes, virtually every university in Europe and North America has responded to the challenges and opportunities of data science by establishing new institutes, departments and degree programmes in the field. In Africa, however, educational institutions have only recently begun to narrow the gap.

Although AI and machine learning have become hot topics at tech, employment and economic forums and workshops, Chinner hasn’t seen a meaningful increase in the number of trained data scientists. He says, “South Africa has enough smart people with the potential to become data scientists, but for some or other reason it hasn’t been a popular career choice.”

Chinner hires applied mathematics graduates and trains them in modern-day data science: “The best people to train, by far, are those who come from applied mathematics. I can’t explain it, but suspect it has to do with the way they view the world.”

Xineoh has also adopted some unorthodox recruitment strategies: “We normally ask recruitment agencies to send us the names of the people who interviewed worst. People who are bad at politics and social skills usually end up being good data scientists.”

To thrive commercially, AI companies also need salespeople who can explain complex algorithms in a way that corporate executives can understand. Finding them hasn’t been easy in South Africa and after several failed hires, Xineoh began appointing salespeople who also have an applied mathematics background. “They are scarce, but they are out there,” says Chinner.

AI sales pitches in South Africa usually have to include a significant educational component. Although large corporates generally recognize the importance of AI, they are mostly in the information-gathering or experimentation phases and are not close to adopting AI on an industrial scale. Chinner also found that South African companies often believe they can build their own AI solutions, but that in-house initiatives rarely get to the implementation phase.

One industry that has shown a particular willingness to adopt AI solutions is the brick-and-mortar retail sector. Retailers’ transaction data (on which AI algorithms feed) is generally in good shape. Chinner ascribes this to the highly competitive nature of the industry: “South African retailers are extremely focused on customer service and having the lowest prices. In order to succeed at both, they need highly accurate real-time data. It is the competitive tension in the industry that makes them open to innovation.”

Software is, however, a global game, with solutions from names such as Microsoft, Oracle or IBM used throughout the world. Xineoh’s two biggest international competitors in the consumer behavior prediction space both sell their solutions globally, including in South Africa.

So is there a case to be made for South African-based AI companies that cater to the local market? Yes, says Chinner: “Local players often have a better understanding of in-country nuances. For instance, in many African companies, even large ones, transactional data is not as structured and clean as it would be in an American multinational. This is mostly because they typically don’t have as much resources available to allocate to data administration. As a result, the algorithms used by AI players in the developed world more often than not struggle to cope with Africa’s unstructured and unclean data, whereas local companies build their solutions with this in mind.

“Think of it this way: a top US tech company is a bit like an F-16 fighter plane: highly efficient, but needs near perfect conditions to take off. By comparison, a MiG [an alternative jet fighter plane] can take off on a dirt road and doesn’t need the same ongoing maintenance. The software platforms created by local companies are MiGs; they tend to be more robust, flexible and suited to local conditions.”