A Credit Score for People the Bureaus Cannot See

What we talked about.

In this episode, Sunny Ray talks with Farhad Bihhat, CEO of Farosian, a Johannesburg based company that has spent over a decade doing social and digital media screening for employers. Farhad explains how that work is expanding into two new areas: continuous monitoring for organizations and an alternative credit scorecard built from people's online presence, aimed at thin file and credit invisible individuals who banks cannot assess through traditional bureaus. He shares the origin story of the company, sparked by a viral tweet scandal he heard about on the radio, and describes how his wife, an anthropologist, shaped the human centered methodology behind the screening process. The conversation covers where Farhad draws ethical lines around protected information, why machines still fail to interpret satire and memes, which platforms are hardest to access for screening, and how 21 different behavioral signals can approximate a credit score. Farhad also opens up about his own debt history and how testing the new model on himself produced a score within 7 percent of his real one, before closing with the company's near term growth targets.

A screening company built on social media data is now scoring the creditworthiness of people traditional bureaus cannot see.

The questions, and the answers.

What are you building and why should people care?

We've done social and digital media screening for 11 and a half years, mostly for HR and recruitment. Now we're building two things: a continuous monitoring platform for organizations, and an alternative data credit scorecard that uses people's online presence and behavior, specifically aimed at thin file and credit invisible people who banks otherwise cannot assess.

When did the idea stop being just an observation and turn into a company?

I was driving home one day and heard a radio interview about an American woman who tweeted something on holiday in 2011 and went viral and got fired by 2015. It hit me that people already informally screen each other socially at bars or parties, and I realized that could be flipped into a professional context. I researched it for months, and it helped that my girlfriend, now my wife and co-founder, is an anthropologist.

Who was the first person willing to pay you, and what did they want to know?

Through cold calling and LinkedIn outreach, I met an executive headhunter named Rob Riddout, who has since moved to the Netherlands. He had two candidates he wanted screened. For the senior candidate we uncovered fraud allegations tied to a past CFO role. For the junior candidate, a former Groupon call center agent, we found glowing customer service reviews, which he used to help her land another job.

Where do you draw the line on what you'll look at in someone's digital footprint?

We exclude anything classified as protected information, like sexual orientation, religious belief, political affiliation, or pregnancy status. Women have been excluded from hiring because an employer spotted a sonogram on Instagram, and the onus is then on the employer to prove that wasn't part of the decision. We make sure what shouldn't be seen isn't seen, and we present both positives and negatives in a balanced way.

Can you give an example where a human and a machine read the same post completely differently?

Machine learning still can't distinguish satire, humor, and intent. There's a well known meme with Barack Obama on one side and Donald Trump on the other with identical text above and below. A human reads it as humor, but a machine just flags unprofessional language. That distinction is something AI still isn't able to make.

You claim coverage of more than 250 platforms. Which is the hardest to see into, and why does that matter?

Facebook is still the hardest. Since the Cambridge Analytica scandal around 2015 or 2016, they've gradually locked down more of what's publicly available. Facebook has essentially become a large ad agency that happens to be a social network, so the more they lock things down, the more data they have to upsell to their own customer base.

What signal in someone's digital activity actually predicts whether they'll repay credit?

We're building a scorecard with 21 different signals, drawing on thesis papers from universities including Oxford, that show online presence can predict propensity to repay. Signals include financially distressed language in posts and the strength and stability of someone's network, since people who are stable tend to associate with stable people. I tested it on myself and it landed within 7 percent of my real credit score.

Who does this reach that a traditional bureau cannot?

Bureaus rely on historical payment records from banks and retailers, so with no history you rarely get credit. In South Africa the township economy, the informal workforce, is worth about 500 billion rand and is largely unbanked. We see 92 to 97 percent penetration of Facebook and Instagram even in rural areas, so that online behavior becomes usable data to build a scorecard and act as an entry point into the formal credit system.