The mortgage industry has spent billions of dollars digitizing the home-lending process. Borrowers can apply online, upload documents through portals and track their applications digitally. Yet behind that modern interface, much of the work required to manufacture a mortgage remains stubbornly manual.
Ace Watanasuparp, founder and CEO of Swish Holdings, believes that is the industry’s fundamental problem.
“For decades, the mortgage industry has manufactured loans by hand,” Watanasuparp said. “Swish is building the technology to manufacture them through software.”
Swish Holdings is a financial holding company building and investing in financial-services businesses, with a particular focus on mortgages. At the center of its technology strategy is Swish Labs, which is developing what the company describes as an AI-native, vertically integrated mortgage technology platform.
Rather than building another individual tool for lenders, Swish Labs is attempting to connect the mortgage lifecycle through a proprietary technology stack encompassing customer relationship management, point of sale, loan origination, pricing, AI-powered underwriting and, eventually, capital-markets technology.
The ambition is less about creating a better digital mortgage application and more about rethinking what happens after a borrower clicks “submit.”
Digitizing Mortgages Didn’t Necessarily Make Them Cheaper
According to Watanasuparp, much of the technology introduced to mortgage lending over the past two decades digitized existing processes rather than eliminating them.
“We’ve taken paper applications and turned them into online applications. We’ve taken physical documents and turned them into PDFs. We’ve created portals and dashboards,” he said.
Behind those interfaces, however, people may still need to review documents, move information between systems, check lending guidelines, clear conditions, price loans and manage files. Lenders can also rely on numerous separate technologies to originate a single mortgage, creating additional integrations and handoffs.
Swish Labs’ thesis is that AI can eliminate or automate portions of that work rather than merely digitize it.
The distinction matters because mortgage origination remains expensive. Watanasuparp says Swish has already demonstrated that it can move the cost structure from roughly $11,500 per loan toward $5,000, with a longer-term objective of pushing the figure to approximately $2,000 or less as more of the manufacturing process is automated.
The mechanism is relatively straightforward: reduce the number of times a human needs to touch a loan.
A traditional mortgage file can pass among loan officers, processors, disclosure teams, underwriters, closers and post-closing personnel. Swish is designing its technology so that software can perform more of the preparatory work before the file reaches those employees.
Its AI can ingest borrower documentation, extract and structure information, identify missing documents, analyze a file against lending requirements and assist in pre-underwriting. The idea is that instead of an employee spending hours preparing a file for an underwriter, the underwriter receives a file that has already undergone substantial analysis.
From Moving Data To Understanding It
Artificial intelligence could be particularly consequential in mortgages because home loans are difficult to automate using conventional rules-based software.
“If X happens, do Y” automation works well when information is standardized. Mortgages are different. Borrowers have varying income structures, documents can contain inconsistent information, and lending guidelines include large numbers of rules and exceptions.
Watanasuparp argues that AI can increasingly interpret that information rather than simply transport it between software systems.
A bank statement, for example, is more than a document that needs to be attached to a borrower’s file. An AI system can potentially analyze the activity within it, identify deposits and patterns, compare information against other documents and flag discrepancies requiring human attention.
“Traditional software manages workflows,” Watanasuparp said. “AI can increasingly understand the loan itself.”
The goal isn’t necessarily to remove humans from mortgage lending altogether. Instead, Swish wants to shift employees away from repetitive processing and toward reviewing exceptions and making decisions that require judgment.
That could make one of the most important measures of AI’s success decidedly unglamorous: loans per employee.
If AI handles more document review, guideline analysis and file preparation, Watanasuparp believes an underwriter could eventually manage multiples of today’s loan volume. Similar productivity improvements could extend to processors, loan officers, operations teams and eventually capital-markets personnel.
For lenders, that could be far more meaningful than simply adding an AI chatbot to a website. A mortgage company capable of substantially increasing loan volume without adding employees at the same rate would have a fundamentally different cost structure.
Building The Software While Making The Product
Swish is pursuing another strategy that distinguishes it from pure mortgage software providers: The company intends to both develop technology and participate in manufacturing mortgages.
That vertical integration is deliberate.
“It’s very difficult to build great mortgage technology if you’re separated from the actual mortgage manufacturing process,” Watanasuparp said.
By operating within mortgage origination, the company can observe where files get stuck, which conditions delay closings, where underwriters spend their time and where margins disappear. Engineers can then develop technology around those problems, deploy it into actual operations and measure whether it works.
The resulting loop—manufacture mortgages, identify friction, build technology, deploy it and measure the result—is central to Swish’s strategy.
It could also become increasingly important as AI systems depend on high-quality data and real-world outcomes. Instead of developing software in isolation, Swish Labs can potentially learn from the mortgages moving through the broader Swish ecosystem.
“We don’t view Swish simply as a SaaS company,” Watanasuparp said. “We’re building the technology and intelligence layer for mortgage manufacturing.”
The Bigger Opportunity May Come After The Mortgage Closes
Reducing origination costs is Swish’s immediate objective, but Watanasuparp’s longer-term vision extends into a much larger financial market.
Every mortgage generates a substantial amount of information about the borrower, property, credit profile, income, assets, interest rate and loan structure. After origination, another set of information begins accumulating: whether borrowers refinance, prepay, become delinquent or default.
Connecting those two sets of data could eventually allow Swish to move from helping manufacture mortgages to helping understand them as financial assets.
That raises questions with enormous implications for banks and investors: What is the probability that a particular mortgage prepays? How likely is a borrower to refinance? How might a pool of mortgages perform under different interest-rate environments? And ultimately, what should an investor pay for that risk?
Swish’s long-term strategy envisions connecting origination, underwriting, pricing, funding, loan performance and predictive modeling in a continuous feedback loop. Information about how mortgages ultimately perform could then feed back into future underwriting and pricing decisions.
If that model works, the addressable opportunity becomes considerably larger than improving the experience of getting a home loan.
Swish Labs would effectively be attempting to create an intelligence layer spanning the mortgage lifecycle—from the moment a prospective borrower enters the system to the point where mortgage risk is priced in the capital markets.
For an industry that has already undergone years of digitization, Watanasuparp believes the next transformation will be less about putting existing processes online and more about eliminating processes that software can perform itself.
“Swish isn’t simply helping lenders originate mortgages more efficiently,” he said. “We’re building an intelligence layer that could help lenders, banks, investors and capital-markets participants better understand and price mortgage risk.”
