The shifting narrative on AI and what we think is right for our clients

By Courtney Dodd, Head of Marketing, Floify 

During the past few years, the mortgage industry has been swimming in a particular kind of noise: AI is coming for jobs, AI will replace loan officers, AI will automate the entire origination process end to end.

At Floify, we never bought into it. Our view has always been that AI works best as a layer of intelligent assistance that makes the people doing the work faster, more accurate and better equipped, not redundant.

The walk-back nobody predicted

Across industries, some of the loudest voices on AI-driven workforce replacement have quietly reversed course. Klarna replaced 700 customer service employees with AI in early 2024, and began rehiring humans 15 months later, with its CEO publicly acknowledging they had pushed too far. Amazon cut 14,000 corporate roles in late 2025 citing AI, only for its CEO to walk it back days later, saying "it's not even really AI-driven." Salesforce made similar headlines cutting 4,000 roles, with its CEO later calling the decision to blame AI "a lazy way out." 

The pattern is emerging, and the data reinforces it. According to McKinsey's 2025 State of AI survey, just 39% of organizations report any enterprise-level bottom-line impact from AI, and most of those say AI accounts for less than 5% of their EBIT. The majority of organizations remain in the piloting or experimentation phase, nowhere near scaled impact. AI performs best as a targeted layer of assistance integrated into proven workflows, not as a wholesale replacement for human judgment and process.

Yes, mortgage is different

In mortgage, the AI-as-replacement argument was always more theoretical than practical. Buying a home is the largest financial transaction most people will ever make. It involves documentation, deadlines, emotional stakes and regulatory complexity that demands accountability at every step. Borrowers want to talk to a real person. They want someone who can explain what just happened with their rate lock, answer a question at 8 p.m. the night before closing or navigate an underwriting condition that doesn't fit neatly into an automated decision tree. This is the reality of how mortgage transactions work, and it hasn't changed.

What's more, the regulatory environment makes fully autonomous AI in mortgage a genuine liability. Fair lending laws require explainability in credit decisions. TILA and RESPA require accuracy in disclosures. Adverse action notices require documented reasoning. An AI system that acts autonomously, without human oversight, audit trails or compliance safeguards, creates both operational risk and legal exposure that no lender should be willing to accept.

Thus, vendors positioning agentic AI as the future of mortgage origination are, in effect, asking lenders to become test cases in one of the most regulated industries in the country. That’s not innovation; that’s risk transfer. 

The role of humans in using AI

There’s a concept gaining traction in research and enterprise technology circles called “human in the loop.” The idea is straightforward: AI handles the analytical heavy lifting of processing data sets, surfacing patterns, automating repetitive tasks, while humans direct the work, make judgment calls and remain accountable for outcomes. Neither replaces the other; each does what it does best.

That model maps cleanly onto the mortgage industry. AI can extract borrower data from uploaded documents, adapt application flows dynamically by loan type, flag inconsistencies and eliminate manual steps that slow down the process. Loan officers can focus on the relationship, the complexity and the compliance. The technology supports humans; it doesn’t try to become one of them.

Build from scratch, or build on what works?

There’s a second question embedded in the AI conversation that doesn’t get enough attention: when a vendor says its platform is AI-powered, what does that actually mean in terms of features and functionality?

Some vendors are attempting to build an AI-native mortgage platform from the ground up, positioning the architecture’s novelty as the selling point. The problem is that a functioning, compliant POS is table stakes, and you can’t build effective AI on top of a weak foundation. Lenders evaluating these platforms should be asking direct questions: How many customers are fully live and in production? How does the AI handle compliance disclosures, adverse actions and audit logs? What happens when something goes wrong the night before a closing?

The more durable approach is to integrate best-in-class AI capabilities into a platform that is proven, compliant and trusted by lenders at scale. That means partnering with leading AI providers rather than attempting to build and maintain proprietary AI infrastructure, a resource-intensive undertaking that pulls focus away from mortgage-specific functionality.

Why we think this approach works best

Rather than building AI from scratch, Floify partners with top AI vendors to layer intelligent capabilities onto a proven, compliant POS foundation, keeping development energy focused on the mortgage-specific functionality lenders actually depend on. Tools like Dynamic AI extract and auto-populate borrower data from uploaded documents, while Dynamic Apps adapt the application experience by loan type without a single line of code. The result is a faster workflow where the loan officer stays in control and compliance stays intact, and that’s the point.

In a highly regulated industry, purpose-built isn’t a compromise. That framing misses the point. Mortgage’s future runs on loan officers who can do more, backed by AI that actually works.