![HW+ candor](https://www.housingwire.com/wp-content/uploads/2021/02/HW-candor.jpg?w=1024)
It wasn’t long ago that scribes were writing the obituary of the mortgage underwriter.
And yet, during the COVID-19 pandemic, the underwriter has become the most sought-after professional in the mortgage industry. Some underwriters were commanding five-figure bonuses on top of six-figure salaries. The problem is, there just aren’t enough of them.
That didn’t change the fact that mortgage lenders had hundreds or thousands of loans to process a day, and an underwriter can only move so quickly. It was apparent that the origination volume – even at a record high – was being artificially constrained because lenders couldn’t keep up. Many lenders took 60 days or more to underwrite and deliver a loan.
Thomas Showalter, the CEO and founder of mortgage AI firm Candor, says he has a solution – automating a significant part of the underwriting process, freeing them up to tackle more loans.
According to Showalter, Candor has reduced the cycle time by an average of 18.7 days, processor productivity improved 20% or more, and the underwriter touches reduced from 2.7 to 1 per file. The 6% pull through lift added $4 million in volume for every 100 loans closed, the company claims.
Showalter says he doesn’t want to put underwriters out of a job. Quite the opposite. He says he merely wants to help them become more productive.
His self- and angel-funded mortgage tech company passed its 50,000 loan stress test in the fall. They’ve already got a top five originator as a client, and a few others in the top 25.
HousingWire caught up with Showalter to talk about existing mortgage tech, the future of the underwriter, the complexities inherent in automating mortgages, and much more.
This interview has been lightly edited for length and clarity.
HousingWire: Could you walk us through how the product works? Is it like a crawler? How much of the loan application can Candor automate?
Thomas Showalter: So the way we distribute Candor is the product is through your lender’s loan origination system. Let’s say they’re an Encompass user. That’s about 40% of the users out there right now, and we’re affiliated with Encompass. So they would go into the Encompass, they would click on Candor, it will pull up their loan and then they would go within Candor.
Now, let’s assume we had a set of non-QM guidelines that are affiliated with that particular lender. The way the lender gets affiliated is they just submit the guideline to Candor for programming, and we just post them and then tell the lender they’re good to go on non-QM, for example. So your lender would click on the non-QM under Candor and Candor would then start processing your loan against those guidelines.
So it would create conditions dynamically that would identify what kind of information this lender is going to need to process your non-QM application. It might ask for a pay stub, it might ask for bank statements and so forth.
HW: So the prospective borrower uploads pay stubs, W2s, a letter from an employer, and the system verifies bank funds?
TS: Yes, it does. So let’s assume you uploaded all of those documents in one form or another, either you uploaded them via some form of PDF, in which case Candor has an ICR service that would go transform those documents into data fields that would then be fed into Candor, and also your lender’s LOM… So all of that gets imported into a data architecture in Candor that has approximately 600 data fields in it. And Candor then uses that data to process your loan.
HW: What happens if something isn’t quite right? Say I expected a $5,000 gift to help with the downpayment, but I’m short? It flags it automatically?
TS: Let’s just say for the sake of argument, your loans going to require $40,000 to close. And let’s say in your bank statement, there’s $38,500 in cash ready to go. So we would note that there’s only $38,500, that you need $40,000. Then we would ask the underwriter to update that in the most meaningful way. We in effect augment the intelligence of the underwriter by doing a lot of the legwork and a lot of the basic thinking, but the things that require true human judgment are then done by the underwriter.
HW: One thing I hear a lot from LOs these days is the amount of errors in underwriting, much of it attributed to such a high volume of work and the fact that many back-office staff are new. How do you solve this problem?
TS: The problem you’re posing is you have a $70,000 underwriter and $100,000 underwriting problem. Candor is the $100,000 underwriter. It would go in and resolve those problems; it doesn’t do stupid stuff. It has within its architecture, the flexibility and the intelligence to go down a variety of paths as per that loan, particular borrower data guideline combination, it goes down that path and it resolves everything in its way that it can possibly resolve. And so those anomalies are presented as conditions, and those are presented for the underwriter and the borrower to resolve.
And assuming all of those were accomplished, Candor then proceeds to a clear to close. We have a 20-day reduction in cycle time and the underwriter touch is going from 2.7 to 1, and in some cases, less than 1 per month. Some loans are produced so clean they no longer need any additional underwriter attention. It saves the underwriter to do the really challenging loans that require his or her expert attention. And so some of the other benefits include an improvement in throughput of 24% and also an increase in customer satisfaction statistically significant.
HW: Do underwriters feel threatened by AI coming in and doing a not insignificant percentage of their job?
TS: The underwriters that use Candor, they really like Candor. Because what Candor does, is the initial conforming log, which is a checklist nightmare. Candor does a lot of the checklist, a lot of the cross validation, a lot of the corroboration so that the conventional loan checklist is done by a machine which can do it faster and more thoroughly than a person.
For example, those anomalies that require human judgment to resolve, or those loans that are largely one collection of anomalies after another, those loans require underwriter attention. So the underwriters that know Candor knows that they can ask Candor to do the stuff that doesn’t interest them anyway and they can focus on stuff that really interests them…
So in my opinion, even though Candor is a very smart piece of software, it’s still just a machine, it’s not yet capable of human thought. I think it’s a nice complement that we’ve found a way to take the mundane forms of critical thinking that are more capable of being organized and taught to a machine, we’ve taught those to Candor. And there’s stuff that’s still out there requiring human intel, that’s still available.
HW: Sure, but what about five years from now when you can do so much more with the technology? Would that put average underwriters out of a job?
TS: I guess the question is, what do you mean by the average underwriter? You talk about a person with 10 years experience that can handle a wide variety of loans, but if there’s still a portion of that person’s loan portfolio that’s sufficiently complex and dynamic and hard to organize for a machine to learn, then that person’s still going to be employed for those. So the job keeps advancing.
The other thing that you have to factor in is what are the lenders going to do now that they have a business model that can be broader and deeper than it’s ever been before? Are they going to try and go out to those more esoteric types of lending problems? Are they going to challenge them now, because now they have the manpower to do it? Now they have the intelligence back there to process all the plain vanilla stuff, but they still need to go after those very esoteric borrowers that are going to require some sharp pencil work in order to underwrite, that still demand human intel.
HW: There’s a lot of mortgage tech out there. What’s different about your system?
TS: One way to differentiate Candor versus the rest, the rest are very static in their approach. Candor is very dynamic and adaptive. So you go down the path of taking that borrower as they presented themselves in their 1003, which is going to soon become the Fannie 3.4 file. And as the data about the borrower, such as the pay stubs, such as the bank statement and so forth, as they get absorbed into the system and presented to the Candor architecture, Candor takes those two data sources and compares them to the eligibility guidelines such as the non-QM that we’re talking about, and just makes the next most appropriate step in that particular loan.
Candor can handle a wide variety of different kinds of level problems. It basically has thought through that little triangle of the 1003, the borrower data such as pay stubs and the eligibility guidelines. It has an engine that’s specializes in anomaly detection, but wherever, for example, say the borrower says he made $15,000 a month, but we could only find $12,500 in documented income, that would create an anomaly within this architecture. And Candor would then resolve that and enlist the underwriter’s participation in that along the borrower.
HW: Can you share your client list?
TS: The contracts are confidential. We have a top five lender and we have some top 25 lenders currently using the product.
HW: Could you talk a little bit about the backend challenges? It seems like the front-facing technology for consumers is pretty good, but there are a lot of pain points with FHFA and GSE guidelines on the back-end. Plus, I don’t think anyone planned for these systems to handle this kind of volume. Does the advancement of this kind of technology ultimately rely on how the FHFA and the GSEs move?
TS: So we take the existing body of investor guidelines from Fannie, from Freddie, whomever, and we implement those. If you want to think of Candor as streamlined, it takes investor guidelines and applies them to the underwriting problems. That’s what it does. So as the investor groups such as Fannies and the Freddie’s of this world start getting their thoughts organized so that they can produce additional investor guidelines basically to expand their footprint… expand what it needs to be in a conventional loan, expand what it needs to be in jumbo or non-QM and enable Freddie and Fannie to be able to do that. So Candor provides a very good vehicle for them to just go do that.
HW: Has this kind of technology brought down the cost of funding a loan? It’s still awfully expensive these days, and mostly because a lot of humans are working on it.
TS: For loans which are appropriate for Candor’s process, the human processing costs are probably reduced by about 90%.
HW: Can you talk about what your plans are going forward? You guys have only been around since the fall.
TS: We have two sets of plans. One is the middle market. The middle market is in dire need of technology, such as Candor because they have no way, shape or form to fund the development of something that’s a multi-million dollar, multi-year investment. They just can’t. So there’s great demand for new market players that want to revolutionize their business model in the mortgage making business. And so they’re our primary target.
And then we’ve identified three additional top lenders that we’d like to get on board. And that will be something that we’ll pursue, those dual plans or parallel plans, we’ll pursue those in the next six months.