Firehorse’s vision for automating decisions came from hard experience underwriting loans and real estate assets around the world. It seemed like every new transaction required a new number that wasn’t in the existing checklist or was hidden inside another calculated number.  This led to manual intervention.  Wouldn’t it be nice if all the data were available and in one place?  We realized that we could do this. Current technology is like catching tuna by finding the exact spot and dropping a line there with the exact bait you need and then doing that a hundred times.  That’s why there are confidence levels and human error checking.  Our technology is a net that catches all the fish in that spot.  We then throw out the fish we don’t need leaving us the tuna. 

Firehorse’s vision is to make all factors behind a decision transparent.  Often it’s hard for two parties to agree on what data leads to a decision because it’s based upon a calculated number with many inputs.  We make those inputs visible which makes the calculation visible.  Now anything related to a decision is easy.  Once all the data is there, you can automate the decision.  You can articulate the decision to another party.  As market conditions change, decision-making can change too.  When other companies automate a process, how can they do so without having all the relevant numbers available?  And will that automation work a year from now?

Howard Petty is an expert in providing post-submission and forensic loan reviews.  Hoch Cho had an affinity for Optical Character Recognition (“OCR”) and automating underwriting processes.  They came together to create Firehorse.

Firehorse started by doing forensic loan reviews which required going through every inch of the loan file to find relevant information.  Often the most important information was locked in parts of the document that are traditionally not reviewed or analyzed.  By going through this process for thousands of loans we created a way to abstract almost all the quantitative data from the loan file and most importantly check it for accuracy.  Unlike the double blind, human check that many loan reviewers use, our algorithmic check is more accurate and consistent and most importantly can say definitively whether or not a number is right or wrong.  We did this process manually as OCR at the time wasn’t good enough.  We were able to offer a better forensic product than our competitors leading to more significant findings but at the same price.

Fast forward a few years and with a noticeable improvement in OCR, we were able to automate all that manual data entry and error checking.  Abstracting 10-20x more data than the industry does also required a new approach to taking in, storing and exchanging that data with partners.  We created a hybrid relational database and NoSQL database infrastructure to solve that challenge. 

Now we bring that forensic loan review engine to the mortgage, title and other industries.  We found that often market conditions change and the data & documents required five years ago are no longer enough to make a decision.  Our approach of bringing in all the data and documents means that we can automate the exceptions, or mortgages that need extra documentation, today and also new challenges that come up in the future.