Can Lenders Learn from Google’s Artificial Intelligence Story

Real Time Loan Underwriting with Partial Data


At Lendit Fintech last month, Scott Penberthy, Director of applied Artificial Intelligence at Google gave a fascinating keynote about how Google came to develop their AI business.  I appreciated his layman’s description of how learning happens via computer (I particularly liked his cat dog graphic).  It was fascinating to hear how quickly they had applied AI in:

  • Videos + Pics = using AI, you can identify images or you can type in Google Photos a keyword and it will find them, such as “beaches”
  • Email – 12% of all Google email responses are AI driven with an app called Inbox
  • Speech translation – Google Translate is now almost real-time

It got me thinking about how lenders are applying technology to underwriting, particularly tying the need for real-time data and the need for speed versus the completeness and accuracy of the data.  At Lendit, Bluevine, the balance sheet small business lender, touted their ability to pull and analyze bank transaction data, not a trivial exercise. First you need a business’s permission to access the data.  Even if you have that permission, you have to interpret bank statements, which are highly variable between financial institutions and not user friendly.  For example, Merchant Cash Advance could show up as MCA, MerchCash, or Advance, or some other form.  Bluevine uses natural language processing to clean and analyze this data.

Ultimately, if you are to stay in business, you need to develop underwriting that provides a lift over traditional metrics.  In the consumer world, its the FICO score.  In B2B, well, its your secret sauce, or Mama’s home cooking as Shaq likes to say.  This can be challenging to do for small business, or any business for that matter, where seasonality plays an issue, other loans may or may not be transparent, and income and profit are difficult to evaluate.

This trend is happening with networks as well.  My company, Global Business Intelligence, is seeing much more data science taking place around contractually committed payables and eFactoring, and we see the industry moving to requesting a few pieces of information from P2P and supplier networks to develop a marketplace type model built on real-time data and speed.  For example:

  • Has a supplier been on network for more than 6 months?
  • Has the buyer approved the invoice?
  • Is the buyer investment grade?
  • What is the historical dilution? Many P2P networks still struggle assessing dilution risk, as that requires tying back payments to original invoices,  which is not an easy thing to do.

We know data driven finance can drive up business advance rates for all types of event financing – starting from purchase orders and raw material buys, to inventory to finally receivables.  Just as a case in point, factoring is traditionally financing 75% to 90% of invoice value, while many forms of approved invoice finance do either 100% less interest or a percentage off invoice value (ie, 1% to 3%)

This movement to bring real time data to lending continues, and while everyone wants to issue the next ICO coin, there are some exciting and quiet developments to help businesses access cash when and where needed.

Don't forget to sign up for TFMs weekly digest delivered to your inbox every Monday here.

Share on Procurious

Discuss this:

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

This site uses Akismet to reduce spam. Learn how your comment data is processed.