As the Biden Administration ramps up its efforts to regulate data collection and surveillance practices, fintechs are racing to develop software products that anticipate ethical considerations in mortgage underwriting and servicing processes. The government has long signaled its interest in acting as a watchdog over consumer information and automated decisioning. In both 2018 and 2020, the Consumer Financial Protection Bureau put out requests for input around data gathering and ownership. Furthermore, as part of his July executive order on competitive practices, President Biden asked regulators to look at data collection and surveillance practices. MeasureOneConsumer-permissioned data aggregator MeasureOne recently expanded its offerings into the mortgage business. As a technology provider, “we obviously don’t play any role in the ownership of the data, and we certainly will always fall on the side of the consumer because we believe not only is that the right thing but ultimately that’s where it’s going to end up,” said CEO Elan Amir In its setup, MeasureOne is the “trusted party.” The consumer gives their consent and provides the credentials so it can access the information, serving as a barrier between the data and the outside world. The business that will ultimately use the information does not get those credentials. “We enable the consumer to share that data and to not have the institution be able to restrict that data, by not providing an electronic interface publicly,” said Amir. “We allow the consumer to essentially say ‘look this is my data, I’ll just share it with the requesting business,’ and in that way everybody wins.” MeasureOne has access to more income and employment data than is used by sources the mortgage business has worked with for years to verify the information, Amir said. “We can share a much more accurate picture and much more up-to-date picture than you would have from a database that is reliant on being updated from payroll processors or other types of data sources.” In the current system, the institution will provide access to the data and can then sell it off, Amir said. “And so what we’ve done and what motivates us, is to use the consumer’s access to the data that they already have to [their own] benefit,” he said. “Ultimately, moving away from the aggregator monetization resale model that has dominated frankly the last 30-40 years.” Touchless Lending by TavantSoftware developer Tavant also recently launched a product aimed at reducing the number of hands in the pot during the origination process. Touchless Lending is designed to provide an end-to-end loan manufacturing experience. “It would be nirvana if we can take a loan from application submission all the way to clear to close without any human intervention,” said Mohammad Rashid, the head of Tavant’s fintech practice. “I doubt that can be done today.” Touchless Lending’s first iteration is mortgage specific. “It will take an application that is submitted to a loan origination system and basically take it forward in an automated manner through the hoops, right through the workflow that allows it to be a decision,” said Rashid. Tavant broke the underwriter’s workflow down into five steps. The software checks for consistency, assessing whether the borrower’s assets match up with the income or the property being bought, the loan program involved and so forth, he said. “Do all these component areas fit as a jigsaw puzzle? When it doesn’t fit, that’s where you want to highlight the fissures, the fault lines,” Rashid said. The underwriter is then informed of what things specifically they would need to further examine. The underwriter’s experience and expertise “is something that we need to be able to convert into a software process, an algorithmic processing capability,” Rashid said. “And what best to do that then is machine learning techniques that we bring in so our injection of AI is very, very judicious.” Tavant is capturing information so there’s a “data lake behind the scenes,” he said. The software records what users did with files, records of human-to-human contact regarding the loan, and what kinds of determinations were made. “And as we get more and more loans through, we basically can generate insights through deep learning capabilities, and then inject those insights back into our operational pipeline,” Rashid said. “These loans with these characteristics and these borrowers fall into this bucket and usually what happens with this bucket is that they go through this set of processes or a set of reviews, and maybe that insight gives [an originator] a better way of judging the pipeline.” InRule integrates simMachinesIn June, InRule, an automated decisioning platform, acquired simMachines, which offers technology that explains decisions made by AI and machine language systems. Even before a transaction, the software informs users about the system rules that came into play when a decision was made, said Chris Berg, InRule’s director of corporate development. By contrast, with machine learning models, “that decision could never say what factors were affecting the outcome, which were the critical criteria present that made that decision,” he said. “With simMachines we can explain, not just in aggregate the impact on a population, but we can say, ‘here’s the specific weights on data that affected that outcome, and not even the data that was present, but also the data that was missing.’” That allows users to test for fairness and ensure that for a protected class there is little or no difference between the outcomes of these decisions. If a lender’s going to use AI safely, it needs to be able to do that, Berg said. To control against unintended consequences, the lenders who use the software need to continually monitor its results. “You could have drift in your models too, so it’s a changing landscape, lending is always changing,” Berg said. “So, you need practice there as well to pick up on the drift and changes that are occurring in the population.” The aim is to lend to more consumers. Even with the changes to the qualified mortgage rule, the debt-to-income ratio ceilings are “a blunt instrument” when trying to expand the market, he said. Someone who doesn’t make much money but is careful with how they spend might actually be a better risk than that person who fits in the traditional ratios but spends their money freely. simMachines can assist in making those distinctions, Berg said. “I think we need a more refined set of tools to find the folks that lift up our points of view about what risk really is, and [thus] how we can help more people,” Berg said.