The word around Wall Street and Silicon Valley is machine learning is going to revolutionize every industry – including financial services. That is, if the technology lives up to its hype, it will solve the mother of all problems: how to price and predict borrower risk.
Revolutionizing lending through machine learning doesn’t sound plausible when you consider there are three major credit bureaus whose sole job is to provide borrower data and risk analysis. Fintech startups are betting they can do better than credit bureaus with less engineers, less data, and less resources devoted to the task. Even if they succeed, what will set them apart from other lenders?
The answer is data. Lenders can use their own loan servicing data with granularity - data that can't be accessed through a credit bureau report. For example, if a borrower is late three days on a payment, it doesn’t get reported to the credit bureaus (only if they are late 30 days or more), but it’s a likely sign that something may be wrong.
It’s not only internal loan servicing data. There are two examples of beautifully carved moats that may have slipped under the radar.
The first is Prosper’s acquisition of BillGuard. When Prosper bought BillGuard last year (renamed Prosper Daily), it gave them unprecedented access to banking and credit data of BillGuard users. Let me repeat. Prosper detailed granular data including credit card transactional data and banking transactional data. It also got access to it prior to the potential borrower requesting a loan. It’s a win-win. Prosper can now train its deep machine learning algorithms with a much broader set of detailed data, and it has a potential customer base to develop a relationship prior to originations. It can pick and choose which users of Prosper Daily to extend a loan offer to with a much stronger set of data than you can get from a Credit Bureau.
The other example is Square’s business lending arm. Square started as the famous company that allowed small businesses to accept credit cards from their mobile phones. It then created a POS (Point Of Sale) system for iPad and gave it away for free. It’s now using that POS and transactional data to calculate risk and provide loans to small businesses. As with Prosper, Square gets to build a relationship of trust with a potential borrower prior to lending. It also has detailed transaction data to determine cash flow. The data that Square has access to is next to impossible to replicate. There isn’t a good enough standard for collecting business credit data.
One competitive advantage of deep machine learning is the data you feed it. However, the results it comes up with (training) are fed back in. It’s recursive. It learns to learn. Its algorithms get smarter. There are hints that machine learning is already as good as traditional methods of analyzing credit risk. The exciting part is that it doesn’t stop there. The results that are fed back in become a differentiated data source.